Stackrig

How accurate is Stackrig?

Stackrig computes its live numbers with a fast flow model. We check that model against a slower, exact simulation that follows every request through the same designs. Below is every check of the current engine: where the two agree within the tolerance, and the few places where they do not, which we publish as known limits.

121 scenarios, 766 checks: 688 agree within the tolerance, 78 are published limits (explained below). No failures.

Tolerances: rates and utilization ±5 %, latency percentiles ±15 %, error shares ±0.5 points or ±15 %. Each value of the exact simulation is the mean of 8 independent runs.

Engine 0.1.0 · page build eefec0d (2026-10-14)

How we test

Every scenario below runs twice: once with the fast model that drives the live numbers in the playground, and once with a slow, exact simulation that follows every single request through the design.

The two must agree within ±5 % for throughput and utilization and within ±15 % for the latency percentiles; the exact simulation is repeated 8 times and averaged, so its own noise stays small.

Where they do not agree, we say so: the known limits below explain in plain words which situation it is (for example long overload, retries that feed on themselves, or a breaker right at its threshold) and in which direction the live numbers are off.

Agreeing with an exact simulation is not the same as agreeing with real machines, so we also run the same designs on real cloud VMs under load and publish how far the model is from those measurements, with the setup of every run.

Known limits

  • S7 · Overload with a finite buffer: With a full accept queue the live model’s p99 is about 20 % too high: the wait is nearly constant then, which the model’s distribution does not capture.
  • S8 · Timeouts and retries (stable): A retry right after a timeout meets the same congestion. The live model correlates retries through the shared wait and the orphaned work of timed-out calls, but still underrates long congestion episodes: the final error rate is about a fifth of the exact simulation’s.
  • S8b · Timeout below the healthy p99: When the retries plus the normal load are more than the service can handle, a retry storm is a second, self-sustaining state. The exact simulation falls into it after a random burst within seconds; the live model needs a trigger, then stays in it too (S37). The playground warns when a design is at risk of it.
  • S11 · Periodic traffic spikes (time averages): Over periodic spikes the p99 is about 18 % too high: transient queues are averaged, not replayed.
  • S16b Ca 2 · Bursty arrivals, 75 % busy: With very bursty arrivals the median is about 23 % too low; the tail (p95, p99) stays within tolerance.
  • S20b · Autoscaling step with cold starts (new instances 3× slower at first): A fresh instance is overloaded on its own while the live model averages its speed with the warm ones: the p99 is about 30 % too low and the autoscaler may decide one step higher.
  • S28b · Thundering herd: slow loads after a flush, with and without coalescing: Without coalescing and with the database deep in overload, the live model ends each load after the wait it met when it started, while in the exact simulation early loads of hot keys overtake the backlog: about 30–40 % too many misses reach the database. With coalescing it matches.
  • S33 · Circuit breaker against a dependency failing 60 %: it opens and the calls fail fast: While the breaker is open the dependency only sees the half-open probes, a few per cycle, and whether a cycle closes it is a coin toss: its arrival rate (under 1 per second here) is noisy in both models.
  • S33b · Circuit breaker near its threshold with fallback skip: it keeps opening and closing: Near its threshold a closed breaker reopens after exactly the minimum number of calls in the live model, while in the exact simulation a lucky window keeps it closed a little longer: the dependency sees about half as many calls (4 against 8 per second, of 200), and the callers about half as many errors (0.6 % against 1.1 %).
  • S35e · Saturated connection pool without a timeout: the caller turns the excess away: The pool is full and the caller’s threads all wait for it, so the caller turns the excess away. The exact simulation first queues it in the caller’s accept queue, the live model rejects it at once: error rate and throughput match, the p99 is about half.
  • S35f · Overloaded connection pool with a timeout and a retry: Overloaded with a timeout, the exact simulation tips into a collapse: once the database backs up, every waiting call gets a connection just before its timeout, so all calls reach the database and it stays overloaded. The live model stays in the state where the excess waits out its timeout. Both report the pool as full.
  • S36 · Blocking queue at 125 %: producers wait for space and the API turns the excess away: Saturated, a producer waits almost exactly the time the consumers need for the messages ahead. The live model spreads that wait wider: the median matches, the p99 is about a third too high.
  • S36b · Blocking queue with a 20 ms limit on the wait for space: With a 20 ms limit, most producers in the exact simulation wait close to it before they get space; the live model lets more of them in at once, so the median is lower. The tail and the error rate match.
  • S38h · Virtual users whose bottleneck is a read-only database of four equal servers: A few users spread over several equal database servers by chance, so at any moment some servers have a queue while others are idle. The live model spreads them evenly: its throughput is about 5 % too high and its p99 about a third too low.
  • S43b · Shared-core instance (GCP e2-small) at 85 % of the CPU it can sustain: On a shared core the requests that run at the same time split its limited speed. The live model gets the average right but underrates the slowest requests by about a fifth.
  • S44 · Several database reads per request, one after the other, through a connection pool, at 50 to 90 % load: With 4 reads through a pool of 2 at 70 % load the median is 15 % too low, at the edge of the tolerance. Every p95 and p99 agrees.
  • S44b · Several database reads per request on a database whose two threads share one core: With only 2 connections into a database whose threads share a core, the wait near full load moves a lot for a small change of the query time: the live model’s median and p95 are 15 to 18 % too low at 90 % load, and the median already from 70 % load with four reads per request. With 16 connections and four reads at 90 % load its p99 is 16 % below this run of the exact simulation, whose own p99 varies by a fifth between repeated runs there; against eight times as many runs it is 6 % low.
  • S46 · Very bursty arrivals (as measured for serverless functions) at one service, 50 to 90 % busy: With very bursty arrivals the slow requests are right (p95 within 11 %), but the typical request is 20 to 60 % too fast in the live model, and at half load its p99 is up to 40 % too high.
  • S46b · Very bursty arrivals at one service whose callers give up after 100 ms: the error rate: With very bursty arrivals and a short timeout the live model shows a quarter to a third too few failed requests at 75 % load and about a sixth too few at 90 %.
  • S46c · Very bursty arrivals at a service with four instances: With very bursty arrivals over four instances the slow requests are right (p95 and p99 within 10 %); the typical request is 15 to 35 % too fast in the live model.
  • S46d · Very bursty arrivals through a connection pool into a database: Behind a connection pool the p99 agrees from 75 % load. At half load it is a quarter to two fifths too high, as at a single service. The p95 at 75 % load is about a fifth too low, and the median is a quarter to three fifths too low: with bursts most requests wait longer than the live model says, while the slowest ones are right.
  • S47 n2 cv0.03 50 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 50 % busy: A server that gets every n-th request and does near-constant work hardly queues, and the little waiting that is left is more even than the live model assumes: its p99 is up to about a quarter too high. The median matches.
  • S47 n2 cv0.03 70 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 70 % busy: A server that gets every n-th request and does near-constant work hardly queues, and the little waiting that is left is more even than the live model assumes: its p99 is up to about a quarter too high. The median matches.
  • S47 n2 cv0.03 90 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 90 % busy: A server that gets every n-th request and does near-constant work hardly queues, and the little waiting that is left is more even than the live model assumes: its p99 is up to about a quarter too high. The median matches.
  • S47 n2 cv2 50 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 50 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n2 cv2 70 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 70 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n2 cv2 90 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 90 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n4 cv0.03 70 · Load balancer hands requests to 4 one-thread servers in turn: near-constant work, 70 % busy: A server that gets every n-th request and does near-constant work hardly queues, and the little waiting that is left is more even than the live model assumes: its p99 is up to about a quarter too high. The median matches.
  • S47 n4 cv2 50 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 50 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n4 cv2 70 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 70 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n4 cv2 90 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 90 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n8 cv0.03 90 · Load balancer hands requests to 8 one-thread servers in turn: near-constant work, 90 % busy: A server that gets every n-th request and does near-constant work hardly queues, and the little waiting that is left is more even than the live model assumes: its p99 is up to about a quarter too high. The median matches.
  • S47 n8 cv2 50 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 50 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n8 cv2 70 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 70 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • S47 n8 cv2 90 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 90 % busy: Work this uneven is outside the range the live model is validated for: its median is up to about a third too high, its p99 up to about a quarter.
  • Timeouts while a backlog grows: When a server falls behind for so long that requests wait out their timeout, the live model shows the failures up to one timeout too early: when the request that will time out arrives, not when its timeout runs out. Example: 2,000 requests a second into a server that finishes 1,600, with a timeout of 30 s. The live model shows the first failures after about 117 s; the exact simulation and the arithmetic say 150 s. From then on both agree.
  • A retry budget in a burst: A retry budget saves up the retries it did not use and can spend them when a burst begins. The live model saves up nothing: in the busiest one to two seconds of a burst it shows about a third to a half fewer calls than the exact simulation, so the load on the service behind it looks lower than it is for those seconds. After that both agree.
  • A database disk near its limit: Commits that arrive while the disk is flushing share the next flush. Up to 90 % of the disk’s throughput the live model’s wait for a flush is within about a tenth of a commit-by-commit simulation, and its p99 within about a seventh. Above that its p99 runs high: at 95 % by about a fifth.

Real-cloud measurements

The same designs, run on real cloud VMs and measured under load (as of 2026-09-30). Setup: the service and, where named, PostgreSQL each on its own Google Cloud n2-standard-2 VM (2 vCPU), the load generator on a third VM; the load rises in steps of 60 s, at a constant rate or with random (Poisson) arrivals. The table shows how far the live model is from the measurement on average, over the load steps up to 90 % CPU.

Mean absolute deviation of the live model from real machines
RunArrivalsThroughputp50p95p99
S1: one CPU-bound service (2 vCPU, n2)constant0.0 %6.6 %6.7 %6.3 %
S1: one CPU-bound service (2 vCPU, n2)Poisson1.0 %22.6 %23.1 %26.9 %
S2: service → Postgres (2 vCPU each, n2)constant0.0 %130.6 %97.6 %44.0 %
S2: service → Postgres (2 vCPU each, n2)Poisson0.2 %99.8 %30.7 %42.1 %

Throughput matches the real machines within 1 %. For a CPU-bound service the latency percentiles are within 7 % at a constant rate and within about 25 % under random (Poisson) arrivals. For a service in front of PostgreSQL the real requests take under a millisecond, so small fixed costs dominate and the model’s median is up to about 1.7× off. Near the database’s capacity the model still underestimates the slowest requests, because requests there also wait for a free database connection in the application, which we are still calibrating. Until then, read Stackrig’s tails near a database’s capacity as optimistic.

Every check

Showing 766 of 766 checks.
S1 · One service, 4 threads, 75 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
srate300300−0.0 %±5.0 %within
sutil0.7500.750+0.1 %±5.0 %within
end-to-endp5011.8411.60+2.1 %±15.0 %within
end-to-endp9541.5040.53+2.4 %±15.0 %within
end-to-endp9963.4863.480.0 %±15.0 %within
S2 · Load balancer → service → database with a replica
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate12101209+0.1 %±5.0 %within
sutil0.5500.550−0.0 %±5.0 %within
dbutil0.6820.680+0.3 %±5.0 %within
end-to-endp5016.3615.87+3.1 %±15.0 %within
end-to-endp9539.5536.62+8.0 %±15.0 %within
end-to-endp9955.1853.22+3.7 %±15.0 %within
S3 · Cache in front of the database
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate400401−0.4 %±5.0 %within
dbutil0.7000.704−0.6 %±5.0 %within
end-to-endp502.112.110.0 %±15.0 %within
end-to-endp9511.6011.600.0 %±15.0 %within
end-to-endp9922.2222.220.0 %±15.0 %within
S4 · Parallel and sequential fan-out
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate600603−0.4 %±5.0 %within
dbutil0.6000.603−0.5 %±5.0 %within
end-to-endp5035.6435.640.0 %±15.0 %within
end-to-endp9565.4365.430.0 %±15.0 %within
end-to-endp9986.9186.910.0 %±15.0 %within
S5 · Queue with workers
WhereMetricLive modelExact simulationDeviationToleranceResult
wrate199199+0.1 %±5.0 %within
wutil0.7450.743+0.3 %±5.0 %within
qqp9958.1155.18+5.3 %±15.0 %within
wnp9973.2473.240.0 %±15.0 %within
S6 · Rate limit at 1.2× the traffic
WhereMetricLive modelExact simulationDeviationToleranceResult
cerr0.1670.168−0.7 %±15.0 %within
end-to-endp501.791.790.0 %±15.0 %within
end-to-endp995.315.43−2.2 %±15.0 %within
S7 · Overload with a finite buffer
WhereMetricLive modelExact simulationDeviationToleranceResult
cerr0.1690.169−0.2 %±15.0 %within
sutil0.9970.998−0.1 %±5.0 %within
end-to-endp5069.3479.10−12.3 %±15.0 %within
end-to-endp95143127+12.3 %±15.0 %within
end-to-endp99193158+22.2 %±15.0 %limit
S7c · Overload with a finite buffer (constant arrivals)
WhereMetricLive modelExact simulationDeviationToleranceResult
cok per s from t = 0399399−0.1 %±5.0 %within
S8 · Timeouts and retries (stable)
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate268270−1.0 %±5.0 %within
s-db-readedgeErr0.06710.0793−15.3 %±15.0 %limit
cerr0.001170.00643−81.8 %±15.0 %limit
end-to-endp504.464.460.0 %±15.0 %within
end-to-endp9512.0812.33−2.0 %±15.0 %within
end-to-endp9918.8018.31+2.7 %±15.0 %within
S8b · Timeout below the healthy p99
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate290750−61.4 %±5.0 %limit
cerr0.05911.00−94.1 %±15.0 %limit
S9 · Thread-pool exhaustion
WhereMetricLive modelExact simulationDeviationToleranceResult
sslots0.8920.891+0.1 %±5.0 %within
cerr0.000.000.0 %±15.0 %within
end-to-endp501101100.0 %±15.0 %within
end-to-endp95236232+1.7 %±15.0 %within
end-to-endp99316309+2.5 %±15.0 %within
S10 · Highly variable service times
WhereMetricLive modelExact simulationDeviationToleranceResult
sutil0.8000.799+0.1 %±5.0 %within
end-to-endp5024.6622.22+11.0 %±15.0 %within
end-to-endp95127124+2.4 %±15.0 %within
end-to-endp99213197+7.9 %±15.0 %within
S11 · Periodic traffic spikes (time averages)
WhereMetricLive modelExact simulationDeviationToleranceResult
sutil0.6360.636−0.0 %±5.0 %within
end-to-endp5012.3312.330.0 %±15.0 %within
end-to-endp99146118+24.0 %±15.0 %limit
S12 · The “Web app with database” template from the live demo
WhereMetricLive modelExact simulationDeviationToleranceResult
dbutil0.4400.440−0.0 %±5.0 %within
apiutil0.1870.187+0.2 %±5.0 %within
end-to-endp5021.2421.240.0 %±15.0 %within
end-to-endp9555.1855.180.0 %±15.0 %within
end-to-endp9986.9186.910.0 %±15.0 %within
S13 · Slot-bound queue consumer
WhereMetricLive modelExact simulationDeviationToleranceResult
wrate60.0060.02−0.0 %±5.0 %within
wslots0.8250.826−0.1 %±5.0 %within
wutil0.07500.0752−0.2 %±5.0 %within
qqp991971970.0 %±15.0 %within
S14 · Writes into a read-through cache
WhereMetricLive modelExact simulationDeviationToleranceResult
krate22502254−0.2 %±5.0 %within
dbrate300300−0.2 %±5.0 %within
dbutil0.5250.525−0.1 %±5.0 %within
kutil0.02250.0225−0.1 %±5.0 %within
knp9914.2814.04+1.7 %±15.0 %within
end-to-endp503.453.39+1.8 %±15.0 %within
end-to-endp9510.139.89+2.5 %±15.0 %within
end-to-endp9917.3317.330.0 %±15.0 %within
S15 · 8 threads on 4 cores with hyperthreads, 80 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
srate10501052−0.2 %±5.0 %within
sutil0.7610.761−0.1 %±5.0 %within
end-to-endp5011.6011.35+2.2 %±15.0 %within
end-to-endp9538.5739.55−2.5 %±15.0 %within
end-to-endp9962.0163.48−2.3 %±15.0 %within
S15b · 2 vCPUs on 1 core, constant arrivals
WhereMetricLive modelExact simulationDeviationToleranceResult
srate230230−0.0 %±5.0 %within
sutil0.6980.698−0.0 %±5.0 %within
end-to-endp506.536.530.0 %±15.0 %within
end-to-endp957.267.39−1.7 %±15.0 %within
end-to-endp997.637.75−1.6 %±15.0 %within
S16a Ca 0.5 · Smoother-than-random arrivals, 75 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
srate300300−0.1 %±5.0 %within
sutil0.7500.749+0.1 %±5.0 %within
end-to-endp5010.3810.13+2.4 %±15.0 %within
end-to-endp9522.2221.24+4.6 %±15.0 %within
end-to-endp9930.5228.56+6.8 %±15.0 %within
S16b Ca 2 · Bursty arrivals, 75 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
srate300299+0.5 %±5.0 %within
sutil0.7500.747+0.4 %±5.0 %within
end-to-endp5014.0418.31−23.3 %±15.0 %limit
end-to-endp9554.2056.15−3.5 %±15.0 %within
end-to-endp9983.0183.010.0 %±15.0 %within
S16c Ca 0 chain · Constant arrivals through service → database, 80 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
sutil0.3000.300−0.0 %±5.0 %within
dbutil0.8000.800−0.0 %±5.0 %within
dbnp9913.5512.57+7.8 %±15.0 %within
end-to-endp508.188.42−2.9 %±15.0 %within
end-to-endp9514.2813.55+5.4 %±15.0 %within
end-to-endp9918.8016.85+11.6 %±15.0 %within
S17 · Timeout storm without retries
WhereMetricLive modelExact simulationDeviationToleranceResult
c (arrivalCv 1)goodput per s from t = 0138155−10.5 %±15.0 %within
c (arrivalCv 0)goodput per s from t = 0145144+0.8 %±15.0 %within
S17b · Finite queue whose full-buffer wait is about the timeout
WhereMetricLive modelExact simulationDeviationToleranceResult
sutil1.001.00+0.0 %±5.0 %within
cerr0.4580.521−12.2 %±15.0 %within
c-sedgeErr0.4580.521−12.2 %±15.0 %within
S18 · Network round trip in front of a tight timeout, 1 retry
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate442442+0.1 %±5.0 %within
s-db-readedgeErr0.1060.106+0.3 %±15.0 %within
snp9941.5041.500.0 %±15.0 %within
end-to-endp5020.7520.750.0 %±15.0 %within
end-to-endp9535.6435.640.0 %±15.0 %within
end-to-endp9943.4643.460.0 %±15.0 %within
S20 · Autoscaling: traffic steps to 3×, instances follow (Kubernetes HPA)
WhereMetricLive modelExact simulationDeviationToleranceResult
sinstance trajectory: largest gap (instances)0.000.000.0 %±100.0 %within
sinstances billed (mean)6.906.90+0.0 %±5.0 %within
cerr0.08700.0870+0.1 %±15.0 %within
end-to-endp99961914+5.1 %±15.0 %within
S20b · Autoscaling step with cold starts (new instances 3× slower at first)
WhereMetricLive modelExact simulationDeviationToleranceResult
sinstance trajectory: largest gap (instances)0.8750.00+87.5 %±100.0 %within
sinstances billed (mean)8.858.28+6.9 %±5.0 %limit
cerr0.1090.114−4.2 %±15.0 %within
end-to-endp9913591953−30.4 %±15.0 %limit
S21 · Scale from zero: the first request waits for the start
WhereMetricLive modelExact simulationDeviationToleranceResult
cfirst successful response (s)10.1010.00+1.0 %±5.0 %within
cok responses per s19.9919.86+0.7 %±5.0 %within
cerr9.94e-70.000.0 %±15.0 %within
S22 · Autoscaling over a daily traffic curve: instance-hours
WhereMetricLive modelExact simulationDeviationToleranceResult
sinstance trajectory: largest gap (instances)1.000.00+100.0 %±100.0 %within
sinstances billed (mean)3.093.19−3.2 %±5.0 %within
cerr2.85e-160.000.0 %±15.0 %within
end-to-endp9979.1075.20+5.2 %±15.0 %within
S27 · Cache flush under load: database load while the cache warms up
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrequests per s after the flush: first second757754+0.4 %±5.0 %within
dbrequests per s after the flush: seconds 1–1056445590+1.0 %±5.0 %within
dbrequests per s after the flush: seconds 10–601720817203+0.0 %±5.0 %within
S28 · Cache TTL: keys stored together expire together (with and without coalescing)
WhereMetricLive modelExact simulationDeviationToleranceResult
dbcoalesced: peak rate of the expiry wave (per 0.5 s)286296−3.4 %±10.0 %within
dbcoalesced: flush, first 5 s56285608+0.4 %±10.0 %within
dbcoalesced: expiry wave 34–40 s30152961+1.8 %±10.0 %within
dbherd: peak rate of the expiry wave (per 0.5 s)286297−3.5 %±10.0 %within
dbherd: flush, first 5 s56375638−0.0 %±10.0 %within
dbherd: expiry wave 34–40 s30242967+1.9 %±10.0 %within
S28b · Thundering herd: slow loads after a flush, with and without coalescing
WhereMetricLive modelExact simulationDeviationToleranceResult
dbcoalesced: first 2 s after the flush19641952+0.6 %±10.0 %within
dbcoalesced: seconds 2–2019511964−0.7 %±10.0 %within
dbherd: first 2 s after the flush47623663+30.0 %±10.0 %limit
dbherd: seconds 2–2050803636+39.7 %±10.0 %limit
S29 · Request classes: browse and checkout take different paths through a shared service and database
WhereMetricLive modelExact simulationDeviationToleranceResult
app [browse]rate450451−0.3 %±5.0 %within
db [browse]rate900903−0.3 %±5.0 %within
end-to-end [browse]p5014.7713.79+7.1 %±15.0 %within
end-to-end [browse]p9545.4144.43+2.2 %±15.0 %within
end-to-end [browse]p9971.2969.34+2.8 %±15.0 %within
app [checkout]rate50.0049.84+0.3 %±5.0 %within
db [checkout]rate10099.67+0.3 %±5.0 %within
end-to-end [checkout]p5065.4363.48+3.1 %±15.0 %within
end-to-end [checkout]p95124122+1.6 %±15.0 %within
end-to-end [checkout]p99162158+2.5 %±15.0 %within
apputil0.7500.747+0.4 %±5.0 %within
dbutil0.5500.550−0.0 %±5.0 %within
end-to-endp99110106+3.7 %±15.0 %within
S30 · Request classes under load: slow checkouts make browse wait at the shared service
WhereMetricLive modelExact simulationDeviationToleranceResult
app [browse]rate320321−0.3 %±5.0 %within
db [browse]rate640642−0.3 %±5.0 %within
end-to-end [browse]p5020.2617.82+13.7 %±15.0 %within
end-to-end [browse]p9584.9686.91−2.2 %±15.0 %within
end-to-end [browse]p991351350.0 %±15.0 %within
app [checkout]rate80.0079.74+0.3 %±5.0 %within
db [checkout]rate160159+0.3 %±5.0 %within
end-to-end [checkout]p5079.1077.15+2.5 %±15.0 %within
end-to-end [checkout]p951621620.0 %±15.0 %within
end-to-end [checkout]p99221217+1.8 %±15.0 %within
apputil0.8000.799+0.1 %±5.0 %within
S31 · Request classes through a queue: messages keep their class, per-class work at the worker
WhereMetricLive modelExact simulationDeviationToleranceResult
w [small]rate450450−0.1 %±5.0 %within
w [big]rate150150−0.2 %±5.0 %within
db [big]rate150150−0.2 %±5.0 %within
wutil0.8250.826−0.1 %±5.0 %within
dbutil0.3000.300+0.1 %±5.0 %within
qqp990.0009690.0009690.0 %±15.0 %within
end-to-end [small]p9913.0613.060.0 %±15.0 %within
end-to-end [big]p9912.8212.820.0 %±15.0 %within
S32 · Retry budget: three retries against a dependency failing 30 %, capped at 10 % of the calls plus 5 per second
WhereMetricLive modelExact simulationDeviationToleranceResult
xrate335334+0.4 %±5.0 %within
s-xshed65.4065.17+0.3 %±5.0 %within
cerr0.2180.219−0.1 %±15.0 %within
end-to-endp5021.7321.730.0 %±15.0 %within
end-to-endp9557.1356.15+1.7 %±15.0 %within
end-to-endp9986.9186.910.0 %±15.0 %within
S33 · Circuit breaker against a dependency failing 60 %: it opens and the calls fail fast
WhereMetricLive modelExact simulationDeviationToleranceResult
s-xopen1.000.998+0.2 %±5.0 %within
s-xshed199199+0.3 %±5.0 %within
cerr0.9990.998+0.1 %±15.0 %within
xrate0.6000.918−34.6 %±5.0 %limit
S33b · Circuit breaker near its threshold with fallback skip: it keeps opening and closing
WhereMetricLive modelExact simulationDeviationToleranceResult
s-xopen0.9870.968+1.9 %±5.0 %within
cerr0.006100.0114−46.6 %±15.0 %limit
xrate4.077.79−47.7 %±5.0 %limit
s-xshed196192+1.9 %±5.0 %within
S34 · Rate limit on a service: 600 requests per second against a limit of 500
WhereMetricLive modelExact simulationDeviationToleranceResult
serr0.1670.168−0.8 %±15.0 %within
sutil0.06250.0623+0.3 %±5.0 %within
end-to-endp501.911.910.0 %±15.0 %within
end-to-endp956.046.040.0 %±15.0 %within
end-to-endp9910.3810.380.0 %±15.0 %within
S35 · Connection pool: 4 connections per instance, 75 % busy, the wait for a connection
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil0.7500.750+0.0 %±5.0 %within
dbrate24002401−0.0 %±5.0 %within
cerr4.06e-90.00+100.0 %±15.0 %within
end-to-endp504.824.820.0 %±15.0 %within
end-to-endp9511.3511.350.0 %±15.0 %within
end-to-endp9915.8715.870.0 %±15.0 %within
S35b · Bulkhead: 2 slots and no waiting, the excess fails at once
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil0.5170.518−0.2 %±5.0 %within
dbrate828831−0.4 %±5.0 %within
cerr0.3100.308+0.6 %±15.0 %within
s-db-readshed372371+0.5 %±5.0 %within
end-to-endp503.513.510.0 %±15.0 %within
end-to-endp9912.3312.330.0 %±15.0 %within
S35c · Connection pool at 87 % with a 20 ms timeout that includes the wait
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil0.8740.871+0.4 %±5.0 %within
dbrate27982801−0.1 %±5.0 %within
cerr0.004290.00366+17.3 %±15.0 %within
end-to-endp506.776.53+3.7 %±15.0 %within
end-to-endp9516.3615.87+3.1 %±15.0 %within
end-to-endp9920.7519.78+4.9 %±15.0 %within
S35d · Connection pool near capacity with a 5 ms limit on the wait
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil0.8880.890−0.2 %±5.0 %within
dbrate28422849−0.2 %±5.0 %within
cerr0.05250.0501+4.9 %±15.0 %within
s-db-readshed158150+4.9 %±5.0 %within
end-to-endp506.416.410.0 %±15.0 %within
end-to-endp9917.8217.820.0 %±15.0 %within
S35e · Saturated connection pool without a timeout: the caller turns the excess away
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil1.001.00−0.0 %±5.0 %within
dbrate32003201−0.0 %±5.0 %within
cerr0.1110.111+0.3 %±15.0 %within
end-to-endp5032.7198.63−66.8 %±15.0 %limit
end-to-endp9959.08118−50.0 %±15.0 %limit
S35f · Overloaded connection pool with a timeout and a retry
WhereMetricLive modelExact simulationDeviationToleranceResult
s-db-readpoolUtil1.001.00+0.0 %±5.0 %within
dbrate32006086−47.4 %±5.0 %limit
cerr0.9451.00−5.5 %±15.0 %within
S36 · Blocking queue at 125 %: producers wait for space and the API turns the excess away
WhereMetricLive modelExact simulationDeviationToleranceResult
wrate397401−1.2 %±5.0 %within
cerr0.2070.198+4.6 %±15.0 %within
end-to-endp50387395−2.0 %±15.0 %within
end-to-endp99633480+31.7 %±15.0 %limit
qqp99127170−25.3 %±15.0 %limit
S36b · Blocking queue with a 20 ms limit on the wait for space
WhereMetricLive modelExact simulationDeviationToleranceResult
wrate400400−0.1 %±5.0 %within
cerr0.2000.200−0.1 %±15.0 %within
end-to-endp503.8815.87−75.6 %±15.0 %limit
end-to-endp9523.1922.22+4.4 %±15.0 %within
end-to-endp9924.6623.68+4.1 %±15.0 %within
S36c · Blocking queue at 87 %: it rarely blocks
WhereMetricLive modelExact simulationDeviationToleranceResult
wrate350349+0.2 %±5.0 %within
cerr0.000.000.0 %±15.0 %within
qqp9988.8790.82−2.2 %±15.0 %within
end-to-endp502.232.230.0 %±15.0 %within
end-to-endp996.656.53+1.9 %±15.0 %within
S37 · Metastable retry storm: after a 5 s spike to twice the load, the storm lasts
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate750751−0.1 %±5.0 %within
cerr1.001.00−0.0 %±15.0 %within
S38 · Virtual users: 6 users with 0.1 s think time, below the knee
WhereMetricLive modelExact simulationDeviationToleranceResult
crate51.3451.39−0.1 %±5.0 %within
sutil0.5130.515−0.2 %±5.0 %within
cusers0.8660.860+0.6 %±15.0 %within
end-to-endp5011.8411.60+2.1 %±15.0 %within
end-to-endp9973.2479.10−7.4 %±15.0 %within
S38b · Virtual users: 11 users with 0.1 s think time, the knee
WhereMetricLive modelExact simulationDeviationToleranceResult
crate83.4683.59−0.2 %±5.0 %within
sutil0.8350.839−0.6 %±5.0 %within
cusers2.652.61+1.6 %±15.0 %within
end-to-endp5025.1524.66+2.0 %±15.0 %within
end-to-endp99118120−1.6 %±15.0 %within
S38c · Virtual users: 33 users with 0.1 s think time, throughput at its cap
WhereMetricLive modelExact simulationDeviationToleranceResult
crate100.0099.58+0.4 %±5.0 %within
sutil1.001.00−0.0 %±5.0 %within
cusers23.0023.00−0.0 %±15.0 %within
end-to-endp50213225−5.2 %±15.0 %within
end-to-endp99434387+12.1 %±15.0 %within
S38d · Virtual users: 40 users through a service and a database
WhereMetricLive modelExact simulationDeviationToleranceResult
crate193193+0.1 %±5.0 %within
apiutil0.1930.193+0.1 %±5.0 %within
cusers1.391.39+0.3 %±15.0 %within
end-to-endp506.046.040.0 %±15.0 %within
end-to-endp9924.1724.170.0 %±15.0 %within
S38e · Virtual users without think time: 20 users always waiting, like a load test without sleep
WhereMetricLive modelExact simulationDeviationToleranceResult
crate100100−0.0 %±5.0 %within
sutil1.001.00+0.0 %±5.0 %within
cusers20.0020.00+0.0 %±15.0 %within
end-to-endp50186193−4.0 %±15.0 %within
end-to-endp99355332+7.1 %±15.0 %within
S38f · Virtual users whose bottleneck is the database behind the app: a primary with writes and one replica
WhereMetricLive modelExact simulationDeviationToleranceResult
crate192192+0.0 %±5.0 %within
dbrate385385−0.0 %±5.0 %within
end-to-endp50277285−2.7 %±15.0 %within
end-to-endp998528520.0 %±15.0 %within
S38g · Virtual users whose bottleneck is the database behind the app: a primary with writes and three replicas
WhereMetricLive modelExact simulationDeviationToleranceResult
crate294296−0.8 %±5.0 %within
dbrate588593−0.8 %±5.0 %within
end-to-endp50254225+13.0 %±15.0 %within
end-to-endp99773727+6.5 %±15.0 %within
S38h · Virtual users whose bottleneck is a read-only database of four equal servers
WhereMetricLive modelExact simulationDeviationToleranceResult
crate500475+5.2 %±5.0 %limit
dbrate1000951+5.2 %±5.0 %limit
end-to-endp50114120−4.9 %±15.0 %within
end-to-endp99213309−31.0 %±15.0 %limit
S39 · Zones: an API in three zones calls a backend in one, two of three calls cross zones
WhereMetricLive modelExact simulationDeviationToleranceResult
berate900901−0.1 %±5.0 %within
beutil0.4500.451−0.2 %±5.0 %within
end-to-endp5011.1111.35−2.2 %±15.0 %within
end-to-endp9930.0328.56+5.1 %±15.0 %within
S40 · Regions: Frankfurt calls Virginia (115 ms) with a 120 ms timeout
WhereMetricLive modelExact simulationDeviationToleranceResult
berate300300+0.2 %±5.0 %within
cerr0.1820.181+0.5 %±15.0 %within
end-to-endp501201200.0 %±15.0 %within
end-to-endp991271270.0 %±15.0 %within
S41 · Read replicas replay 300 writes per second next to their reads: how far they lag
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate13001299+0.0 %±5.0 %within
dbutil0.3770.376+0.2 %±5.0 %within
dblag1.070.989+8.6 %±15.0 %within
end-to-endp503.693.63+1.7 %±15.0 %within
end-to-endp9920.2619.78+2.5 %±15.0 %within
S41b · Read replicas replay a busy write stream with their one replay thread at 72 %
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate14501452−0.1 %±5.0 %within
dblag5.965.93+0.6 %±15.0 %within
end-to-endp504.464.33+2.8 %±15.0 %within
end-to-endp9934.6733.69+2.9 %±15.0 %within
S42 · Failover: the primary’s zone goes down, a standby takes over after 20 s
WhereMetricLive modelExact simulationDeviationToleranceResult
dboutage: seconds with most requests failing20.0020.000.0 %±5.0 %within
cerror share over the run0.2220.223−0.4 %±15.0 %within
dbarrivals: before the outage2300023001−0.0 %±5.0 %within
dbarrivals: after the failover3496334896+0.2 %±5.0 %within
S42b · Failover: the primary’s zone goes down, a read replica takes over after 30 s
WhereMetricLive modelExact simulationDeviationToleranceResult
dboutage: seconds with most requests failing30.0030.000.0 %±5.0 %within
cerror share over the run0.3000.300+0.0 %±15.0 %within
dbarrivals: before the outage2300023043−0.2 %±5.0 %within
dbarrivals: after the failover3496335002−0.1 %±5.0 %within
S43 · Burstable instance: its CPU credits run out and it slows to its baseline
WhereMetricLive modelExact simulationDeviationToleranceResult
apitime the credits run out (s)3003000.0 %±5.0 %within
cok per s at the baseline40.0039.70+0.8 %±5.0 %within
cerror share over the run0.09310.0971−4.2 %±15.0 %within
S43b · Shared-core instance (GCP e2-small) at 85 % of the CPU it can sustain
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate70.0069.67+0.5 %±5.0 %within
apiutil0.7910.789+0.3 %±5.0 %within
end-to-endp5056.1553.22+5.5 %±15.0 %within
end-to-endp95221270−18.1 %±15.0 %limit
end-to-endp99340426−20.2 %±15.0 %limit
S43c · Shared-core instance at full load against our measurement of a real e2-small
WhereMetricLive modelExact simulationDeviationToleranceResult
apisaturated throughput vs the vm12 measurement (views/s)766772−0.7 %±10.0 %within
S44 · Several database reads per request, one after the other, through a connection pool, at 50 to 90 % load
WhereMetricLive modelExact simulationDeviationToleranceResult
pool 2, 1 call, 50 %p502.902.900.0 %±15.0 %within
pool 2, 1 call, 50 %p954.704.700.0 %±15.0 %within
pool 2, 1 call, 50 %p996.165.92+4.1 %±15.0 %within
pool 2, 1 call, 70 %p503.203.27−1.9 %±15.0 %within
pool 2, 1 call, 70 %p956.656.53+1.9 %±15.0 %within
pool 2, 1 call, 70 %p999.168.91+2.7 %±15.0 %within
pool 2, 1 call, 80 %p503.693.81−3.2 %±15.0 %within
pool 2, 1 call, 80 %p959.168.91+2.7 %±15.0 %within
pool 2, 1 call, 80 %p9912.8212.57+1.9 %±15.0 %within
pool 2, 1 call, 90 %p505.315.43−2.2 %±15.0 %within
pool 2, 1 call, 90 %p9516.3615.87+3.1 %±15.0 %within
pool 2, 1 call, 90 %p9923.6823.680.0 %±15.0 %within
pool 2, 2 calls, 50 %p504.824.94−2.5 %±15.0 %within
pool 2, 2 calls, 50 %p958.678.91−2.7 %±15.0 %within
pool 2, 2 calls, 50 %p9912.3311.84+4.1 %±15.0 %within
pool 2, 2 calls, 70 %p505.435.92−8.2 %±15.0 %within
pool 2, 2 calls, 70 %p9514.2813.79+3.5 %±15.0 %within
pool 2, 2 calls, 70 %p9920.7519.78+4.9 %±15.0 %within
pool 2, 2 calls, 80 %p506.777.26−6.7 %±15.0 %within
pool 2, 2 calls, 80 %p9520.7520.26+2.4 %±15.0 %within
pool 2, 2 calls, 80 %p9930.5229.05+5.0 %±15.0 %within
pool 2, 2 calls, 90 %p5011.3511.60−2.1 %±15.0 %within
pool 2, 2 calls, 90 %p9541.5039.55+4.9 %±15.0 %within
pool 2, 2 calls, 90 %p9962.0156.15+10.4 %±15.0 %within
pool 2, 4 calls, 50 %p508.678.91−2.7 %±15.0 %within
pool 2, 4 calls, 50 %p9516.8517.33−2.8 %±15.0 %within
pool 2, 4 calls, 50 %p9924.6623.68+4.1 %±15.0 %within
pool 2, 4 calls, 70 %p509.6411.35−15.1 %±15.0 %limit
pool 2, 4 calls, 70 %p9529.5429.05+1.7 %±15.0 %within
pool 2, 4 calls, 70 %p9943.4640.53+7.2 %±15.0 %within
pool 2, 4 calls, 80 %p5013.0614.53−10.1 %±15.0 %within
pool 2, 4 calls, 80 %p9544.4343.46+2.2 %±15.0 %within
pool 2, 4 calls, 80 %p9967.3863.48+6.2 %±15.0 %within
pool 2, 4 calls, 90 %p5023.6824.66−4.0 %±15.0 %within
pool 2, 4 calls, 90 %p9590.8286.91+4.5 %±15.0 %within
pool 2, 4 calls, 90 %p99139131+6.0 %±15.0 %within
pool 16, 1 call, 50 %p502.902.900.0 %±15.0 %within
pool 16, 1 call, 50 %p954.334.330.0 %±15.0 %within
pool 16, 1 call, 50 %p995.435.19+4.7 %±15.0 %within
pool 16, 1 call, 70 %p503.143.20−1.9 %±15.0 %within
pool 16, 1 call, 70 %p955.805.68+2.2 %±15.0 %within
pool 16, 1 call, 70 %p997.637.39+3.3 %±15.0 %within
pool 16, 1 call, 80 %p503.513.57−1.7 %±15.0 %within
pool 16, 1 call, 80 %p957.517.39+1.7 %±15.0 %within
pool 16, 1 call, 80 %p9910.3810.13+2.4 %±15.0 %within
pool 16, 1 call, 90 %p504.584.82−5.1 %±15.0 %within
pool 16, 1 call, 90 %p9512.3312.82−3.8 %±15.0 %within
pool 16, 1 call, 90 %p9917.8218.31−2.7 %±15.0 %within
pool 16, 2 calls, 50 %p504.824.94−2.5 %±15.0 %within
pool 16, 2 calls, 50 %p957.637.51+1.6 %±15.0 %within
pool 16, 2 calls, 50 %p9910.139.64+5.1 %±15.0 %within
pool 16, 2 calls, 70 %p505.315.55−4.4 %±15.0 %within
pool 16, 2 calls, 70 %p9511.3511.350.0 %±15.0 %within
pool 16, 2 calls, 70 %p9915.8715.50+2.4 %±15.0 %within
pool 16, 2 calls, 80 %p506.166.53−5.6 %±15.0 %within
pool 16, 2 calls, 80 %p9515.8715.870.0 %±15.0 %within
pool 16, 2 calls, 80 %p9923.1922.71+2.2 %±15.0 %within
pool 16, 2 calls, 90 %p509.169.64−5.1 %±15.0 %within
pool 16, 2 calls, 90 %p9529.0531.01−6.3 %±15.0 %within
pool 16, 2 calls, 90 %p9943.4645.41−4.3 %±15.0 %within
pool 16, 4 calls, 50 %p508.428.91−5.5 %±15.0 %within
pool 16, 4 calls, 50 %p9514.2814.280.0 %±15.0 %within
pool 16, 4 calls, 50 %p9919.2918.80+2.6 %±15.0 %within
pool 16, 4 calls, 70 %p509.4010.38−9.4 %±15.0 %within
pool 16, 4 calls, 70 %p9522.7123.19−2.1 %±15.0 %within
pool 16, 4 calls, 70 %p9932.7131.74+3.1 %±15.0 %within
pool 16, 4 calls, 80 %p5011.6012.57−7.8 %±15.0 %within
pool 16, 4 calls, 80 %p9533.6933.690.0 %±15.0 %within
pool 16, 4 calls, 80 %p9948.3447.36+2.1 %±15.0 %within
pool 16, 4 calls, 90 %p5018.8019.78−4.9 %±15.0 %within
pool 16, 4 calls, 90 %p9563.4865.43−3.0 %±15.0 %within
pool 16, 4 calls, 90 %p9996.6896.680.0 %±15.0 %within
S44b · Several database reads per request on a database whose two threads share one core
WhereMetricLive modelExact simulationDeviationToleranceResult
pool 2, 1 call, 50 %p503.393.390.0 %±15.0 %within
pool 2, 1 call, 50 %p955.806.41−9.5 %±15.0 %within
pool 2, 1 call, 50 %p998.188.42−2.9 %±15.0 %within
pool 2, 1 call, 70 %p503.974.09−3.0 %±15.0 %within
pool 2, 1 call, 70 %p959.169.64−5.1 %±15.0 %within
pool 2, 1 call, 70 %p9912.8213.06−1.9 %±15.0 %within
pool 2, 1 call, 80 %p504.705.07−7.2 %±15.0 %within
pool 2, 1 call, 80 %p9512.5713.79−8.8 %±15.0 %within
pool 2, 1 call, 80 %p9918.3119.29−5.1 %±15.0 %within
pool 2, 1 call, 90 %p507.268.42−13.8 %±15.0 %within
pool 2, 1 call, 90 %p9523.1927.59−15.9 %±15.0 %limit
pool 2, 1 call, 90 %p9934.6740.53−14.5 %±15.0 %within
pool 2, 2 calls, 50 %p505.805.92−2.1 %±15.0 %within
pool 2, 2 calls, 50 %p9511.1112.57−11.7 %±15.0 %within
pool 2, 2 calls, 50 %p9916.3617.33−5.6 %±15.0 %within
pool 2, 2 calls, 70 %p507.027.93−11.5 %±15.0 %within
pool 2, 2 calls, 70 %p9519.7820.75−4.7 %±15.0 %within
pool 2, 2 calls, 70 %p9928.5630.03−4.9 %±15.0 %within
pool 2, 2 calls, 80 %p509.1610.62−13.8 %±15.0 %within
pool 2, 2 calls, 80 %p9530.0331.74−5.4 %±15.0 %within
pool 2, 2 calls, 80 %p9944.4348.34−8.1 %±15.0 %within
pool 2, 2 calls, 90 %p5016.3619.29−15.2 %±15.0 %limit
pool 2, 2 calls, 90 %p9560.0673.24−18.0 %±15.0 %limit
pool 2, 2 calls, 90 %p9990.82104−13.1 %±15.0 %within
pool 2, 4 calls, 50 %p5010.6211.11−4.4 %±15.0 %within
pool 2, 4 calls, 50 %p9522.2225.63−13.3 %±15.0 %within
pool 2, 4 calls, 50 %p9932.7135.64−8.2 %±15.0 %within
pool 2, 4 calls, 70 %p5013.0615.50−15.7 %±15.0 %limit
pool 2, 4 calls, 70 %p9540.5344.43−8.8 %±15.0 %within
pool 2, 4 calls, 70 %p9961.0463.48−3.8 %±15.0 %within
pool 2, 4 calls, 80 %p5017.8221.24−16.1 %±15.0 %limit
pool 2, 4 calls, 80 %p9563.4871.29−11.0 %±15.0 %within
pool 2, 4 calls, 80 %p9996.68103−5.7 %±15.0 %within
pool 2, 4 calls, 90 %p5034.6742.48−18.4 %±15.0 %limit
pool 2, 4 calls, 90 %p95135154−12.7 %±15.0 %within
pool 2, 4 calls, 90 %p99201217−7.2 %±15.0 %within
pool 16, 1 call, 50 %p503.513.45+1.8 %±15.0 %within
pool 16, 1 call, 50 %p955.806.41−9.5 %±15.0 %within
pool 16, 1 call, 50 %p998.188.42−2.9 %±15.0 %within
pool 16, 1 call, 70 %p504.094.21−2.9 %±15.0 %within
pool 16, 1 call, 70 %p959.169.40−2.6 %±15.0 %within
pool 16, 1 call, 70 %p9912.8212.57+1.9 %±15.0 %within
pool 16, 1 call, 80 %p504.825.07−4.8 %±15.0 %within
pool 16, 1 call, 80 %p9512.5712.82−1.9 %±15.0 %within
pool 16, 1 call, 80 %p9918.3118.310.0 %±15.0 %within
pool 16, 1 call, 90 %p507.147.51−4.9 %±15.0 %within
pool 16, 1 call, 90 %p9521.7322.71−4.3 %±15.0 %within
pool 16, 1 call, 90 %p9931.7432.71−3.0 %±15.0 %within
pool 16, 2 calls, 50 %p506.166.04+2.0 %±15.0 %within
pool 16, 2 calls, 50 %p9510.8612.08−10.1 %±15.0 %within
pool 16, 2 calls, 50 %p9915.5016.36−5.2 %±15.0 %within
pool 16, 2 calls, 70 %p507.397.75−4.7 %±15.0 %within
pool 16, 2 calls, 70 %p9518.8019.29−2.5 %±15.0 %within
pool 16, 2 calls, 70 %p9927.5927.590.0 %±15.0 %within
pool 16, 2 calls, 80 %p509.169.89−7.4 %±15.0 %within
pool 16, 2 calls, 80 %p9528.0828.56−1.7 %±15.0 %within
pool 16, 2 calls, 80 %p9941.5040.53+2.4 %±15.0 %within
pool 16, 2 calls, 90 %p5015.0116.36−8.2 %±15.0 %within
pool 16, 2 calls, 90 %p9552.2555.18−5.3 %±15.0 %within
pool 16, 2 calls, 90 %p9981.0584.96−4.6 %±15.0 %within
pool 16, 4 calls, 50 %p5011.3511.11+2.2 %±15.0 %within
pool 16, 4 calls, 50 %p9521.2424.17−12.1 %±15.0 %within
pool 16, 4 calls, 50 %p9931.0133.69−8.0 %±15.0 %within
pool 16, 4 calls, 70 %p5013.7915.01−8.1 %±15.0 %within
pool 16, 4 calls, 70 %p9538.5741.50−7.1 %±15.0 %within
pool 16, 4 calls, 70 %p9957.1359.08−3.3 %±15.0 %within
pool 16, 4 calls, 80 %p5017.8220.26−12.0 %±15.0 %within
pool 16, 4 calls, 80 %p9559.0863.48−6.9 %±15.0 %within
pool 16, 4 calls, 80 %p9988.8792.77−4.2 %±15.0 %within
pool 16, 4 calls, 90 %p5031.0134.67−10.6 %±15.0 %within
pool 16, 4 calls, 90 %p95114131−12.7 %±15.0 %within
pool 16, 4 calls, 90 %p99178213−16.5 %±15.0 %limit
S44c · Several reads per request, check points kept aside while the model was built, and a cache plus a database in one request
WhereMetricLive modelExact simulationDeviationToleranceResult
pool 2, 3 calls, 60 %p506.907.51−8.1 %±15.0 %within
pool 2, 3 calls, 60 %p9516.3616.360.0 %±15.0 %within
pool 2, 3 calls, 60 %p9923.1923.190.0 %±15.0 %within
pool 2, 3 calls, 85 %p5012.5713.06−3.7 %±15.0 %within
pool 2, 3 calls, 85 %p9544.4341.50+7.1 %±15.0 %within
pool 2, 3 calls, 85 %p9965.4361.04+7.2 %±15.0 %within
pool 2, 2 calls, 60 %p505.075.31−4.6 %±15.0 %within
pool 2, 2 calls, 60 %p9510.8610.62+2.3 %±15.0 %within
pool 2, 2 calls, 60 %p9915.5014.53+6.7 %±15.0 %within
pool 2, 4 calls, 60 %p508.919.64−7.6 %±15.0 %within
pool 2, 4 calls, 60 %p9521.7322.22−2.2 %±15.0 %within
pool 2, 4 calls, 60 %p9931.7431.740.0 %±15.0 %within
pool 16, 3 calls, 60 %p506.907.39−6.6 %±15.0 %within
pool 16, 3 calls, 60 %p9513.3113.55−1.8 %±15.0 %within
pool 16, 3 calls, 60 %p9918.3118.310.0 %±15.0 %within
pool 16, 3 calls, 85 %p5010.6211.35−6.5 %±15.0 %within
pool 16, 3 calls, 85 %p9531.7432.71−3.0 %±15.0 %within
pool 16, 3 calls, 85 %p9947.3647.360.0 %±15.0 %within
pool 16, 2 calls, 60 %p504.945.19−4.7 %±15.0 %within
pool 16, 2 calls, 60 %p959.168.91+2.7 %±15.0 %within
pool 16, 2 calls, 60 %p9912.3311.84+4.1 %±15.0 %within
pool 16, 4 calls, 60 %p508.919.40−5.2 %±15.0 %within
pool 16, 4 calls, 60 %p9517.3317.330.0 %±15.0 %within
pool 16, 4 calls, 60 %p9924.1723.68+2.1 %±15.0 %within
cache + database, 2 calls each, 60 %p506.166.41−3.8 %±15.0 %within
cache + database, 2 calls each, 60 %p9510.3810.13+2.4 %±15.0 %within
cache + database, 2 calls each, 60 %p9913.5513.06+3.7 %±15.0 %within
cache + database, 2 calls each, 85 %p508.428.91−5.5 %±15.0 %within
cache + database, 2 calls each, 85 %p9521.7321.24+2.3 %±15.0 %within
cache + database, 2 calls each, 85 %p9931.0130.03+3.3 %±15.0 %within
S44d · Several reads per request on two threads that share one core, check points kept aside while the model was built
WhereMetricLive modelExact simulationDeviationToleranceResult
pool 2, 3 calls, 60 %p508.919.64−7.6 %±15.0 %within
pool 2, 3 calls, 60 %p9521.7324.17−10.1 %±15.0 %within
pool 2, 3 calls, 60 %p9931.7433.69−5.8 %±15.0 %within
pool 2, 3 calls, 85 %p5017.3320.26−14.5 %±15.0 %within
pool 2, 3 calls, 85 %p9563.4871.29−11.0 %±15.0 %within
pool 2, 3 calls, 85 %p9996.68101−3.9 %±15.0 %within
pool 2, 2 calls, 60 %p506.296.65−5.5 %±15.0 %within
pool 2, 2 calls, 60 %p9514.5315.87−8.5 %±15.0 %within
pool 2, 2 calls, 60 %p9921.2421.73−2.2 %±15.0 %within
pool 2, 4 calls, 60 %p5011.6012.57−7.8 %±15.0 %within
pool 2, 4 calls, 60 %p9529.0532.71−11.2 %±15.0 %within
pool 2, 4 calls, 60 %p9943.4646.39−6.3 %±15.0 %within
pool 16, 3 calls, 60 %p509.409.64−2.5 %±15.0 %within
pool 16, 3 calls, 60 %p9520.7522.71−8.6 %±15.0 %within
pool 16, 3 calls, 60 %p9930.0331.01−3.1 %±15.0 %within
pool 16, 3 calls, 85 %p5016.8518.80−10.4 %±15.0 %within
pool 16, 3 calls, 85 %p9557.1362.01−7.9 %±15.0 %within
pool 16, 3 calls, 85 %p9984.9690.82−6.5 %±15.0 %within
pool 16, 2 calls, 60 %p506.656.650.0 %±15.0 %within
pool 16, 2 calls, 60 %p9514.0415.01−6.5 %±15.0 %within
pool 16, 2 calls, 60 %p9920.2620.75−2.4 %±15.0 %within
pool 16, 4 calls, 60 %p5012.3312.57−1.9 %±15.0 %within
pool 16, 4 calls, 60 %p9527.5931.01−11.0 %±15.0 %within
pool 16, 4 calls, 60 %p9940.5344.43−8.8 %±15.0 %within
S45 · A database 20 % over its capacity behind a service, with and without a second service in front: when the collapse comes
WhereMetricLive modelExact simulationDeviationToleranceResult
2 calls per request, service in frontcollapse relative to no service in front1.001.05−4.6 %±15.0 %within
2 calls per request, service in frontcollapse: s until less than half succeed7.799.45−17.6 %±25.0 %within
2 calls per request, no service in frontcollapse: s until less than half succeed7.799.02−13.7 %±25.0 %within
2 calls per request, service in frontok per s in the first 4 s422377+11.8 %±15.0 %within
2 calls per request, no service in frontok per s in the first 4 s423376+12.3 %±15.0 %within
4 calls per request, service in frontcollapse relative to no service in front0.9980.969+2.9 %±15.0 %within
4 calls per request, service in frontcollapse: s until less than half succeed8.357.03+18.8 %±25.0 %within
4 calls per request, no service in frontcollapse: s until less than half succeed8.377.25+15.5 %±25.0 %within
4 calls per request, service in frontok per s in the first 4 s203177+14.7 %±15.0 %within
4 calls per request, no service in frontok per s in the first 4 s204178+14.4 %±15.0 %within
S46 · Very bursty arrivals (as measured for serverless functions) at one service, 50 to 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
CV 3, 50 %p5010.3813.31−22.0 %±15.0 %limit
CV 3, 50 %p9531.7432.71−3.0 %±15.0 %within
CV 3, 50 %p9956.1545.41+23.7 %±15.0 %limit
CV 3, 75 %p5015.8730.03−47.2 %±15.0 %limit
CV 3, 75 %p9598.63110−10.6 %±15.0 %within
CV 3, 75 %p99158174−9.0 %±15.0 %within
CV 3, 90 %p5059.0877.15−23.4 %±15.0 %limit
CV 3, 90 %p953093090.0 %±15.0 %within
CV 3, 90 %p99480434+10.8 %±15.0 %within
CV 3.7, 50 %p5010.3814.53−28.6 %±15.0 %limit
CV 3.7, 50 %p9539.5537.60+5.2 %±15.0 %within
CV 3.7, 50 %p9975.2054.20+38.7 %±15.0 %limit
CV 3.7, 75 %p5016.8541.50−59.4 %±15.0 %limit
CV 3.7, 75 %p95143154−7.6 %±15.0 %within
CV 3.7, 75 %p99232240−3.3 %±15.0 %within
CV 3.7, 90 %p5083.01114−27.4 %±15.0 %limit
CV 3.7, 90 %p95449488−8.0 %±15.0 %within
CV 3.7, 90 %p99711695+2.2 %±15.0 %within
S46b · Very bursty arrivals at one service whose callers give up after 100 ms: the error rate
WhereMetricLive modelExact simulationDeviationToleranceResult
CV 3, 50 %err0.0005730.00+100.0 %±15.0 %within
CV 3, 75 %err0.04830.0650−25.6 %±15.0 %limit
CV 3, 90 %err0.3410.398−14.3 %±15.0 %within
CV 3.7, 50 %err0.003420.000115+2859.3 %±15.0 %within
CV 3.7, 75 %err0.1040.153−32.2 %±15.0 %limit
CV 3.7, 90 %err0.4500.554−18.7 %±15.0 %limit
S46c · Very bursty arrivals at a service with four instances
WhereMetricLive modelExact simulationDeviationToleranceResult
CV 3, 50 %p5011.6014.04−17.4 %±15.0 %limit
CV 3, 50 %p9535.6438.57−7.6 %±15.0 %within
CV 3, 50 %p9956.1555.18+1.8 %±15.0 %within
CV 3, 75 %p5018.8025.63−26.7 %±15.0 %limit
CV 3, 75 %p9584.9688.87−4.4 %±15.0 %within
CV 3, 75 %p991311310.0 %±15.0 %within
CV 3, 90 %p5052.2562.01−15.7 %±15.0 %limit
CV 3, 90 %p95229240−4.9 %±15.0 %within
CV 3, 90 %p99355348+2.2 %±15.0 %within
CV 3.7, 50 %p5011.6015.26−24.0 %±15.0 %limit
CV 3.7, 50 %p9542.4843.46−2.2 %±15.0 %within
CV 3.7, 50 %p9967.3863.48+6.2 %±15.0 %within
CV 3.7, 75 %p5021.2431.74−33.1 %±15.0 %limit
CV 3.7, 75 %p95108114−5.1 %±15.0 %within
CV 3.7, 75 %p991701700.0 %±15.0 %within
CV 3.7, 90 %p5065.4379.10−17.3 %±15.0 %limit
CV 3.7, 90 %p95309340−9.2 %±15.0 %within
CV 3.7, 90 %p99473488−3.2 %±15.0 %within
S46d · Very bursty arrivals through a connection pool into a database
WhereMetricLive modelExact simulationDeviationToleranceResult
CV 3, 50 %p504.335.68−23.7 %±15.0 %limit
CV 3, 50 %p9514.7713.55+9.0 %±15.0 %within
CV 3, 50 %p9924.1718.80+28.6 %±15.0 %limit
CV 3, 75 %p507.3914.28−48.3 %±15.0 %limit
CV 3, 75 %p9542.4852.25−18.7 %±15.0 %limit
CV 3, 75 %p9973.2475.20−2.6 %±15.0 %within
CV 3, 90 %p5025.1539.55−36.4 %±15.0 %limit
CV 3, 90 %p95146166−11.8 %±15.0 %within
CV 3, 90 %p99232254−8.5 %±15.0 %within
CV 3.7, 50 %p504.336.29−31.1 %±15.0 %limit
CV 3.7, 50 %p9518.3116.36+11.9 %±15.0 %within
CV 3.7, 50 %p9932.7123.19+41.1 %±15.0 %limit
CV 3.7, 75 %p508.4220.26−58.4 %±15.0 %limit
CV 3.7, 75 %p9563.4877.15−17.7 %±15.0 %limit
CV 3.7, 75 %p99108120−9.8 %±15.0 %within
CV 3.7, 90 %p5036.6260.06−39.0 %±15.0 %limit
CV 3.7, 90 %p95221225−1.7 %±15.0 %within
CV 3.7, 90 %p99348371−6.3 %±15.0 %within
S47 n2 cv0.03 50 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate100.0099.95+0.0 %±5.0 %within
apiutil0.5000.500+0.1 %±5.0 %within
end-to-endp5012.5712.82−1.9 %±15.0 %within
end-to-endp9934.6728.08+23.5 %±15.0 %limit
S47 n2 cv0.03 70 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate140139+0.4 %±5.0 %within
apiutil0.7000.697+0.4 %±5.0 %within
end-to-endp5013.3114.04−5.2 %±15.0 %within
end-to-endp9952.2543.46+20.2 %±15.0 %limit
S47 n2 cv0.03 90 · Load balancer hands requests to 2 one-thread servers in turn: near-constant work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate180180+0.2 %±5.0 %within
apiutil0.9000.899+0.1 %±5.0 %within
end-to-endp5025.6326.12−1.9 %±15.0 %within
end-to-endp99131108+20.7 %±15.0 %limit
S47 n2 cv1 50 · Load balancer hands requests to 2 one-thread servers in turn: work that varies like a typical handler, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate100.00100−0.2 %±5.0 %within
apiutil0.5000.500+0.1 %±5.0 %within
end-to-endp5013.3112.82+3.8 %±15.0 %within
end-to-endp9988.8788.870.0 %±15.0 %within
S47 n2 cv1 70 · Load balancer hands requests to 2 one-thread servers in turn: work that varies like a typical handler, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate140140+0.1 %±5.0 %within
apiutil0.7000.699+0.2 %±5.0 %within
end-to-endp5019.7818.80+5.2 %±15.0 %within
end-to-endp99139131+6.0 %±15.0 %within
S47 n2 cv1 90 · Load balancer hands requests to 2 one-thread servers in turn: work that varies like a typical handler, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate180180+0.0 %±5.0 %within
apiutil0.9000.902−0.3 %±5.0 %within
end-to-endp5054.2053.22+1.8 %±15.0 %within
end-to-endp99371387−4.0 %±15.0 %within
S47 n2 cv2 50 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate100.00100−0.2 %±5.0 %within
apiutil0.5000.500+0.0 %±5.0 %within
end-to-endp5014.5312.33+17.8 %±15.0 %limit
end-to-endp99332293+13.3 %±15.0 %within
S47 n2 cv2 70 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate140140−0.0 %±5.0 %within
apiutil0.7000.705−0.7 %±5.0 %within
end-to-endp5030.5225.63+19.0 %±15.0 %limit
end-to-endp99539434+24.3 %±15.0 %limit
S47 n2 cv2 90 · Load balancer hands requests to 2 one-thread servers in turn: very uneven work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate180179+0.4 %±5.0 %within
apiutil0.9000.901−0.1 %±5.0 %within
end-to-endp50127118+7.4 %±15.0 %within
end-to-endp9912971016+27.7 %±15.0 %limit
S47 n4 cv0.03 50 · Load balancer hands requests to 4 one-thread servers in turn: near-constant work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate200199+0.3 %±5.0 %within
apiutil0.5000.498+0.3 %±5.0 %within
end-to-endp5012.5712.570.0 %±15.0 %within
end-to-endp9921.2419.78+7.4 %±15.0 %within
S47 n4 cv0.03 70 · Load balancer hands requests to 4 one-thread servers in turn: near-constant work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate280280+0.1 %±5.0 %within
apiutil0.7000.700+0.1 %±5.0 %within
end-to-endp5012.5712.82−1.9 %±15.0 %within
end-to-endp9932.7127.59+18.6 %±15.0 %limit
S47 n4 cv0.03 90 · Load balancer hands requests to 4 one-thread servers in turn: near-constant work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate360360−0.1 %±5.0 %within
apiutil0.9000.901−0.1 %±5.0 %within
end-to-endp5016.3618.80−13.0 %±15.0 %within
end-to-endp9973.2467.38+8.7 %±15.0 %within
S47 n4 cv1 50 · Load balancer hands requests to 4 one-thread servers in turn: work that varies like a typical handler, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate200200+0.2 %±5.0 %within
apiutil0.5000.499+0.3 %±5.0 %within
end-to-endp5012.0811.84+2.1 %±15.0 %within
end-to-endp9977.1583.01−7.1 %±15.0 %within
S47 n4 cv1 70 · Load balancer hands requests to 4 one-thread servers in turn: work that varies like a typical handler, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate280280+0.2 %±5.0 %within
apiutil0.7000.701−0.1 %±5.0 %within
end-to-endp5017.3316.36+6.0 %±15.0 %within
end-to-endp99118127−6.9 %±15.0 %within
S47 n4 cv1 90 · Load balancer hands requests to 4 one-thread servers in turn: work that varies like a typical handler, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate360360−0.1 %±5.0 %within
apiutil0.9000.899+0.1 %±5.0 %within
end-to-endp5046.3943.46+6.7 %±15.0 %within
end-to-endp993093090.0 %±15.0 %within
S47 n4 cv2 50 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate200200−0.1 %±5.0 %within
apiutil0.5000.503−0.5 %±5.0 %within
end-to-endp5013.5511.35+19.4 %±15.0 %limit
end-to-endp99316270+17.4 %±15.0 %limit
S47 n4 cv2 70 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate280280−0.1 %±5.0 %within
apiutil0.7000.698+0.3 %±5.0 %within
end-to-endp5028.0821.73+29.2 %±15.0 %limit
end-to-endp99508434+17.1 %±15.0 %limit
S47 n4 cv2 90 · Load balancer hands requests to 4 one-thread servers in turn: very uneven work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate360360−0.0 %±5.0 %within
apiutil0.9000.895+0.5 %±5.0 %within
end-to-endp50120101+19.4 %±15.0 %limit
end-to-endp9912341422−13.2 %±15.0 %within
S47 n8 cv0.03 50 · Load balancer hands requests to 8 one-thread servers in turn: near-constant work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate400399+0.2 %±5.0 %within
apiutil0.5000.499+0.2 %±5.0 %within
end-to-endp5012.5712.570.0 %±15.0 %within
end-to-endp9913.7915.50−11.0 %±15.0 %within
S47 n8 cv0.03 70 · Load balancer hands requests to 8 one-thread servers in turn: near-constant work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate560560−0.1 %±5.0 %within
apiutil0.7000.701−0.1 %±5.0 %within
end-to-endp5012.5712.570.0 %±15.0 %within
end-to-endp9920.7519.29+7.6 %±15.0 %within
S47 n8 cv0.03 90 · Load balancer hands requests to 8 one-thread servers in turn: near-constant work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate720720−0.0 %±5.0 %within
apiutil0.9000.900−0.0 %±5.0 %within
end-to-endp5013.0615.01−13.0 %±15.0 %within
end-to-endp9944.4338.57+15.2 %±15.0 %limit
S47 n8 cv1 50 · Load balancer hands requests to 8 one-thread servers in turn: work that varies like a typical handler, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate400400+0.1 %±5.0 %within
apiutil0.5000.498+0.4 %±5.0 %within
end-to-endp5011.6011.35+2.2 %±15.0 %within
end-to-endp9971.2977.15−7.6 %±15.0 %within
S47 n8 cv1 70 · Load balancer hands requests to 8 one-thread servers in turn: work that varies like a typical handler, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate560561−0.1 %±5.0 %within
apiutil0.7000.701−0.2 %±5.0 %within
end-to-endp5015.8715.26+4.0 %±15.0 %within
end-to-endp99108118−8.3 %±15.0 %within
S47 n8 cv1 90 · Load balancer hands requests to 8 one-thread servers in turn: work that varies like a typical handler, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate720721−0.2 %±5.0 %within
apiutil0.9000.900+0.0 %±5.0 %within
end-to-endp5041.5039.55+4.9 %±15.0 %within
end-to-endp99277293−5.3 %±15.0 %within
S47 n8 cv2 50 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate400401−0.2 %±5.0 %within
apiutil0.5000.500−0.1 %±5.0 %within
end-to-endp5013.0610.62+23.0 %±15.0 %limit
end-to-endp99301277+8.5 %±15.0 %within
S47 n8 cv2 70 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate560561−0.2 %±5.0 %within
apiutil0.7000.696+0.5 %±5.0 %within
end-to-endp5027.1020.26+33.7 %±15.0 %limit
end-to-endp99496418+18.7 %±15.0 %limit
S47 n8 cv2 90 · Load balancer hands requests to 8 one-thread servers in turn: very uneven work, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate720719+0.1 %±5.0 %within
apiutil0.9000.895+0.5 %±5.0 %within
end-to-endp5011494.73+20.6 %±15.0 %limit
end-to-endp9912031109+8.5 %±15.0 %within
S47b n4 50 · Load balancer in front of 4 servers with 2 threads each, 50 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate400399+0.2 %±5.0 %within
apiutil0.5000.498+0.4 %±5.0 %within
end-to-endp5010.6210.38+2.4 %±15.0 %within
end-to-endp9956.1555.18+1.8 %±15.0 %within
S47b n4 70 · Load balancer in front of 4 servers with 2 threads each, 70 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate560561−0.2 %±5.0 %within
apiutil0.7000.701−0.1 %±5.0 %within
end-to-endp5013.3112.82+3.8 %±15.0 %within
end-to-endp9971.2971.290.0 %±15.0 %within
S47b n4 90 · Load balancer in front of 4 servers with 2 threads each, 90 % busy
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate720721−0.2 %±5.0 %within
apiutil0.9000.902−0.2 %±5.0 %within
end-to-endp5027.5927.10+1.8 %±15.0 %within
end-to-endp991621620.0 %±15.0 %within
S47c · Half of the requests through the load balancer, half sent to the servers directly
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate280281−0.2 %±5.0 %within
apiutil0.7000.701−0.1 %±5.0 %within
end-to-endp5016.3616.85−2.9 %±15.0 %within
end-to-endp9977.1577.150.0 %±15.0 %within
S47d · Load balancer in front of a single server
WhereMetricLive modelExact simulationDeviationToleranceResult
apirate70.0070.07−0.1 %±5.0 %within
apiutil0.7000.701−0.2 %±5.0 %within
end-to-endp5024.1723.68+2.1 %±15.0 %within
end-to-endp991741740.0 %±15.0 %within
S49 · Time in a queue through a one-minute burst of jobs, workers limited by CPU
WhereMetricLive modelExact simulationDeviationToleranceResult
qtime in the queue p50, 10 s into the burst33443219+3.9 %±15.0 %within
qtime in the queue p99, 10 s into the burst34063531−3.5 %±15.0 %within
qtime in the queue p50, at the end of the burst19750197500.0 %±15.0 %within
qtime in the queue p99, at the end of the burst2025020750−2.4 %±15.0 %within
qtime in the queue p50, 30 s after the burst29750297500.0 %±15.0 %within
qtime in the queue p99, 30 s after the burst3025031250−3.2 %±15.0 %within
qtime in the queue p50, 60 s after the burst1725017750−2.8 %±15.0 %within
qtime in the queue p99, 60 s after the burst1875019250−2.6 %±15.0 %within
S49b · Time in a queue through a one-minute burst of jobs, workers limited by a slow call
WhereMetricLive modelExact simulationDeviationToleranceResult
qtime in the queue p50, 10 s into the burst33443219+3.9 %±15.0 %within
qtime in the queue p99, 10 s into the burst34693594−3.5 %±15.0 %within
qtime in the queue p50, at the end of the burst19750197500.0 %±15.0 %within
qtime in the queue p99, at the end of the burst2025020750−2.4 %±15.0 %within
qtime in the queue p50, 30 s after the burst29750297500.0 %±15.0 %within
qtime in the queue p99, 30 s after the burst3025030750−1.6 %±15.0 %within
qtime in the queue p50, 60 s after the burst1725018250−5.5 %±15.0 %within
qtime in the queue p99, 60 s after the burst1875019750−5.1 %±15.0 %within
S49c · Time in a queue while one of two workers is gone for a minute, workers limited by CPU
WhereMetricLive modelExact simulationDeviationToleranceResult
qtime in the queue p50, 10 s after the worker is gone28442594+9.6 %±15.0 %within
qtime in the queue p99, 10 s after the worker is gone30943156−2.0 %±15.0 %within
qtime in the queue p50, when the worker returns17250172500.0 %±15.0 %within
qtime in the queue p99, when the worker returns1725018250−5.5 %±15.0 %within
qtime in the queue p50, 30 s after the worker returned44384188+6.0 %±15.0 %within
qtime in the queue p99, 30 s after the worker returned60635438+11.5 %±15.0 %within
S49d · Time in a queue while one of two workers is gone for a minute, workers limited by a slow call
WhereMetricLive modelExact simulationDeviationToleranceResult
qtime in the queue p50, 10 s after the worker is gone29062656+9.4 %±15.0 %within
qtime in the queue p99, 10 s after the worker is gone32813031+8.2 %±15.0 %within
qtime in the queue p50, when the worker returns17250172500.0 %±15.0 %within
qtime in the queue p99, when the worker returns17750177500.0 %±15.0 %within
qtime in the queue p50, 30 s after the worker returned44384313+2.9 %±15.0 %within
qtime in the queue p99, 30 s after the worker returned60635563+9.0 %±15.0 %within
S50 1500/s · Postgres writes on a fast disk, commits share flushes: 1,500 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate15001502−0.1 %±5.0 %within
end-to-endp502.172.170.0 %±15.0 %within
end-to-endp992.842.840.0 %±15.0 %within
S50 300/s · Postgres writes on a fast disk, commits share flushes: 300 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate300300−0.1 %±5.0 %within
end-to-endp501.601.600.0 %±15.0 %within
end-to-endp992.662.660.0 %±15.0 %within
S50 3000/s · Postgres writes on a fast disk, commits share flushes: 3,000 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate30003004−0.1 %±5.0 %within
end-to-endp502.352.350.0 %±15.0 %within
end-to-endp993.083.14−1.9 %±15.0 %within
S50b 30 % · Postgres writes on a slow disk, its throughput 30 % used
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate750750−0.1 %±5.0 %within
end-to-endp502.902.96−2.1 %±15.0 %within
end-to-endp994.825.31−9.2 %±15.0 %within
S50b 60 % · Postgres writes on a slow disk, its throughput 60 % used
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate15001502−0.1 %±5.0 %within
end-to-endp504.824.820.0 %±15.0 %within
end-to-endp999.6410.13−4.8 %±15.0 %within
S50b 85 % · Postgres writes on a slow disk, its throughput 85 % used
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate21252126−0.1 %±5.0 %within
end-to-endp5011.6011.84−2.1 %±15.0 %within
end-to-endp9928.5628.08+1.7 %±15.0 %within
S50c 1600/s · MySQL writes (two flushes per commit): 1,600 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate16001602−0.1 %±5.0 %within
end-to-endp505.435.430.0 %±15.0 %within
end-to-endp9910.3810.86−4.5 %±15.0 %within
S50c 200/s · MySQL writes (two flushes per commit): 200 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate200200+0.2 %±5.0 %within
end-to-endp502.112.110.0 %±15.0 %within
end-to-endp993.883.97−2.3 %±15.0 %within
S50c 800/s · MySQL writes (two flushes per commit): 800 writes a second
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate800801−0.1 %±5.0 %within
end-to-endp503.333.39−1.8 %±15.0 %within
end-to-endp995.315.80−8.4 %±15.0 %within
S50d · Writes that do not wait for the disk to flush, its throughput 60 % used
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate15001502−0.1 %±5.0 %within
end-to-endp501.171.11+5.5 %±15.0 %within
end-to-endp993.693.57+3.4 %±15.0 %within
S50e · An API with 2 reads and 0.3 writes per request; only the writes wait for the disk
WhereMetricLive modelExact simulationDeviationToleranceResult
dbrate27602762−0.1 %±5.0 %within
dbutil0.7800.780−0.0 %±5.0 %within
apiutil0.3000.300+0.0 %±5.0 %within
end-to-endp508.428.18+3.0 %±15.0 %within
end-to-endp9927.5927.590.0 %±15.0 %within