AWS / RESOURCE STUDY

CPU and RAM under load.

A separate resource study on the four live t4g.nano hosts, using the current binaries and existing caches. 3 interleaved rounds per workload, 1-second warmups, 5-second measurements, 64 HTTP/1.1 keep-alive connections and four wrk threads on a separate c7g.xlarge in the same AZ. All 60 samples are included, with 0 reported request errors. Network burst capacity was not reset or guaranteed. These CPU/RAM figures belong to this pass, not to the earlier throughput chart. Read CPU and RAM together with achieved request rate and bandwidth: network throttling can lower CPU use. Expand each workload to see every round.

UTC: 2026-10-17T21:59:16.013214+00:00 to 2026-10-17T22:05:33.228476+00:00

All resource snapshots and load samples ↗ · Resources during the original throughput run ↗

Tiny text · HTTP

Server	Median req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
Superweb	37,800	33.9%	24.6%	19.0 MiB	200.7 MiB	0.163 Gbit/s	Network; Not established
nginx	97,675	100.0%	99.6%	9.5 MiB	192.9 MiB	0.306 Gbit/s	CPU; Network
Caddy	18,100	98.7%	98.3%	31.9 MiB	202.7 MiB	0.053 Gbit/s	CPU
lighttpd	18,880	29.5%	13.6%	9.8 MiB	187.8 MiB	0.032 Gbit/s	Network; Not established

All three rounds — Tiny text · HTTP

Round	Server	Req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
1	Superweb	177,772	84.9%	77.8%	13.6 MiB	200.7 MiB	0.720 Gbit/s	Not established
1	nginx	97,675	100.0%	99.6%	7.9 MiB	192.9 MiB	0.306 Gbit/s	CPU
1	Caddy	19,715	98.5%	97.9%	29.6 MiB	202.7 MiB	0.057 Gbit/s	CPU
1	lighttpd	106,176	93.1%	92.2%	7.5 MiB	186.5 MiB	0.299 Gbit/s	Not established
2	nginx	99,812	100.0%	99.6%	8.1 MiB	187.2 MiB	0.313 Gbit/s	CPU
2	Caddy	18,096	98.7%	98.3%	31.7 MiB	192.0 MiB	0.053 Gbit/s	CPU
2	lighttpd	14,809	23.4%	12.2%	9.8 MiB	187.8 MiB	0.032 Gbit/s	Network
2	Superweb	26,979	18.4%	6.5%	19.0 MiB	191.2 MiB	0.032 Gbit/s	Network
3	nginx	17,363	29.6%	15.0%	9.5 MiB	189.0 MiB	0.032 Gbit/s	Network
3	Caddy	18,100	99.1%	98.7%	31.9 MiB	190.6 MiB	0.053 Gbit/s	CPU
3	lighttpd	18,880	29.5%	13.6%	7.6 MiB	186.3 MiB	0.032 Gbit/s	Network
3	Superweb	37,800	33.9%	24.6%	19.0 MiB	191.4 MiB	0.163 Gbit/s	Network

Home page, gzip · HTTP

Server	Median req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
Superweb	125,423	87.0%	77.7%	19.0 MiB	192.3 MiB	4.892 Gbit/s	Network
nginx	92,831	100.0%	99.6%	9.5 MiB	189.2 MiB	3.530 Gbit/s	CPU; Network
Caddy	13,822	98.7%	98.3%	31.9 MiB	196.5 MiB	0.533 Gbit/s	CPU
lighttpd	89,121	100.0%	99.4%	10.0 MiB	202.2 MiB	3.376 Gbit/s	CPU; Network

All three rounds — Home page, gzip · HTTP

Round	Server	Req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
1	nginx	92,831	100.0%	99.6%	7.9 MiB	189.2 MiB	3.530 Gbit/s	CPU
1	Caddy	14,256	98.7%	98.3%	29.8 MiB	193.3 MiB	0.554 Gbit/s	CPU
1	lighttpd	89,792	100.0%	99.5%	7.6 MiB	196.2 MiB	3.397 Gbit/s	CPU
1	Superweb	128,255	89.1%	80.0%	13.9 MiB	191.0 MiB	5.003 Gbit/s	Network
2	nginx	94,144	100.0%	99.6%	8.1 MiB	186.8 MiB	3.584 Gbit/s	CPU
2	Caddy	13,822	98.6%	98.2%	31.8 MiB	196.5 MiB	0.533 Gbit/s	CPU
2	lighttpd	89,121	100.0%	99.4%	10.0 MiB	202.2 MiB	3.376 Gbit/s	CPU
2	Superweb	125,423	87.0%	77.7%	18.7 MiB	192.3 MiB	4.892 Gbit/s	Network
3	Caddy	13,707	98.9%	98.4%	31.9 MiB	192.3 MiB	0.532 Gbit/s	CPU
3	lighttpd	839	2.1%	1.7%	7.6 MiB	186.3 MiB	0.032 Gbit/s	Network
3	Superweb	812	1.8%	1.0%	19.0 MiB	191.5 MiB	0.032 Gbit/s	Network
3	nginx	831	3.0%	2.0%	9.5 MiB	188.4 MiB	0.032 Gbit/s	Network

JavaScript, identity · HTTP

Server	Median req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
Superweb	12	0.6%	0.1%	19.0 MiB	203.7 MiB	0.032 Gbit/s	Network
nginx	3,392	30.8%	10.1%	8.1 MiB	187.4 MiB	5.005 Gbit/s	Network
Caddy	3,391	58.0%	41.7%	31.1 MiB	190.3 MiB	5.006 Gbit/s	Network
lighttpd	3,390	28.1%	8.8%	9.8 MiB	189.7 MiB	5.004 Gbit/s	Network

All three rounds — JavaScript, identity · HTTP

Round	Server	Req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
1	Caddy	3,392	55.2%	40.7%	29.5 MiB	189.3 MiB	5.004 Gbit/s	Network
1	lighttpd	3,390	28.1%	8.8%	7.6 MiB	189.3 MiB	5.005 Gbit/s	Network
1	Superweb	3,393	31.8%	11.6%	14.3 MiB	196.1 MiB	5.004 Gbit/s	Network
1	nginx	3,393	30.4%	10.1%	7.7 MiB	187.4 MiB	5.004 Gbit/s	Network
2	Caddy	3,390	59.8%	41.7%	30.9 MiB	189.8 MiB	5.006 Gbit/s	Network
2	lighttpd	3,391	31.8%	9.3%	7.5 MiB	188.0 MiB	5.004 Gbit/s	Network
2	Superweb	12	0.6%	0.1%	18.7 MiB	203.7 MiB	0.032 Gbit/s	Network
2	nginx	3,392	31.0%	10.3%	8.1 MiB	186.4 MiB	5.005 Gbit/s	Network
3	Caddy	3,391	58.0%	41.9%	31.1 MiB	190.3 MiB	5.006 Gbit/s	Network
3	lighttpd	15	0.4%	0.1%	9.8 MiB	189.7 MiB	0.032 Gbit/s	Network
3	Superweb	12	0.5%	0.1%	19.0 MiB	202.3 MiB	0.032 Gbit/s	Network
3	nginx	3,392	30.8%	10.0%	8.1 MiB	186.7 MiB	5.005 Gbit/s	Network

JavaScript, gzip · HTTP

Server	Median req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
Superweb	59	0.8%	0.2%	19.0 MiB	193.9 MiB	0.032 Gbit/s	Network
nginx	10,837	30.0%	18.6%	8.1 MiB	186.8 MiB	5.003 Gbit/s	Network
Caddy	9,421	99.1%	98.6%	31.8 MiB	200.0 MiB	4.360 Gbit/s	CPU
lighttpd	10,875	30.5%	19.3%	7.6 MiB	187.8 MiB	5.004 Gbit/s	Network

All three rounds — JavaScript, gzip · HTTP

Round	Server	Req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
1	lighttpd	10,876	30.5%	19.3%	7.6 MiB	187.8 MiB	5.004 Gbit/s	Network
1	Superweb	10,821	40.9%	16.0%	14.5 MiB	192.6 MiB	5.004 Gbit/s	Network
1	nginx	10,838	30.0%	18.6%	7.7 MiB	185.4 MiB	5.003 Gbit/s	Network
1	Caddy	9,156	99.1%	98.7%	29.5 MiB	188.8 MiB	4.257 Gbit/s	CPU
2	lighttpd	10,875	33.3%	19.3%	7.5 MiB	185.7 MiB	5.006 Gbit/s	Network
2	Superweb	59	0.5%	0.2%	18.7 MiB	193.8 MiB	0.032 Gbit/s	Network
2	nginx	10,837	30.6%	18.7%	8.1 MiB	186.7 MiB	5.004 Gbit/s	Network
2	Caddy	9,440	99.3%	96.0%	31.5 MiB	200.0 MiB	4.363 Gbit/s	CPU
3	lighttpd	61	0.8%	0.3%	7.6 MiB	187.0 MiB	0.032 Gbit/s	Network
3	Superweb	58	0.8%	0.2%	19.0 MiB	193.9 MiB	0.032 Gbit/s	Network
3	nginx	61	0.8%	0.4%	8.1 MiB	186.8 MiB	0.032 Gbit/s	Network
3	Caddy	9,421	99.1%	98.6%	31.8 MiB	198.4 MiB	4.360 Gbit/s	CPU

JavaScript, gzip · HTTPS

Server	Median req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
Superweb	61	1.4%	0.8%	19.0 MiB	199.1 MiB	0.032 Gbit/s	Network
nginx	10,781	88.7%	85.5%	9.5 MiB	192.8 MiB	5.002 Gbit/s	Network
Caddy	7,329	99.3%	98.8%	32.9 MiB	199.8 MiB	3.419 Gbit/s	CPU
lighttpd	6,157	74.2%	73.8%	10.0 MiB	201.5 MiB	3.504 Gbit/s	CPU + Network; Network

All three rounds — JavaScript, gzip · HTTPS

Round	Server	Req/s	Host CPU	Service CPU	Server RAM used · peak PSS	Host RAM used · peak	Transmit	Constraint
1	Superweb	10,752	69.4%	62.0%	18.7 MiB	199.1 MiB	5.005 Gbit/s	Network
1	nginx	10,781	90.4%	87.3%	8.0 MiB	187.2 MiB	5.003 Gbit/s	Network
1	Caddy	7,329	99.3%	98.8%	32.1 MiB	197.9 MiB	3.419 Gbit/s	CPU
1	lighttpd	10,700	100.0%	98.3%	10.0 MiB	192.7 MiB	4.944 Gbit/s	CPU + Network
2	Superweb	61	0.8%	0.5%	19.0 MiB	195.1 MiB	0.032 Gbit/s	Network
2	nginx	10,787	88.7%	85.5%	8.1 MiB	192.8 MiB	5.002 Gbit/s	Network
2	Caddy	7,471	99.3%	98.7%	32.6 MiB	194.4 MiB	3.480 Gbit/s	CPU
2	lighttpd	6,157	74.2%	73.8%	9.8 MiB	201.5 MiB	3.504 Gbit/s	Network
3	Superweb	60	1.4%	0.8%	19.0 MiB	193.4 MiB	0.032 Gbit/s	Network
3	nginx	61	1.1%	0.7%	9.5 MiB	189.5 MiB	0.032 Gbit/s	Network
3	Caddy	7,220	99.5%	98.9%	32.9 MiB	199.8 MiB	3.355 Gbit/s	CPU
3	lighttpd	60	1.4%	0.7%	9.9 MiB	192.1 MiB	0.032 Gbit/s	Network

How to read the measurements

Resource intervals must fit entirely inside the timed load test; warmup and partial boundary intervals are excluded. CPU and transmit rates are duration-weighted within each run, then reported as the median across rounds. 100% host or service CPU means both vCPUs are occupied. Service CPU covers its cgroup; host CPU also includes kernel/network work and monitoring. Server RAM used is the sampled peak of summed process PSS (resident memory with shared pages apportioned), including every worker; it excludes swapped-out pages and unmapped filesystem cache. Host RAM used is the sampled peak of MemTotal minus MemAvailable, including the OS and other processes, and excluding memory Linux estimates is available for reuse. MemTotal is usable Linux RAM, not advertised instance capacity. RSS, swap PSS, available host RAM and cgroup memory including charged filesystem cache are also retained in the raw data. The sampler can miss subsecond peaks. No server restarts or cache drops were performed; live services carry their existing cache and allocation history.

Constraint labels are inferences: CPU when run-average host CPU is at least 95%; network when transmit reaches 4.75 Gbit/s or bandwidth-allowance counters increase with CPU below 80%; both when both conditions hold. Memory pressure is flagged by new cgroup high/max/OOM events. Otherwise the constraint is not established. Multiple labels mean the inferred constraint differed between rounds. An ENA exceedance counts queued/dropped packets, not HTTP errors. Lack of a memory-pressure event does not prove RAM could never become a limit.

Sampling interval: 1 second. Each snapshot reads /proc/stat, /proc/meminfo, ENA counters, the service cgroup CPU/memory counters, and smaps_rollup for every worker process. The sampler runs outside the service cgroup. This is a short resource study of five repeated files, not a maximum-cache or memory-capacity test. No memory exhaustion is deliberately induced.

Server versions: Superweb 2026.10 preview, nginx 1.24.0, Caddy 2.6.2, lighttpd 1.4.74. Competitor versions are Ubuntu 24.04 packages, not latest upstream.

Superweb executable SHA-256: 8f30ea377d566560dfda9f8533ef65600b2485f50817f0af299302aa023f319d. SiteVue commit: 2ac4fda96182660d819e91830e33029467c80fd5.

For each run, the raw data retains CPU averages, observed duration, sampled memory peaks, bandwidth allowance counters and errors. PSS includes shared pages proportionally; summing RSS can count shared pages multiple times. Cgroup memory includes charged page cache and kernel memory and must not be substituted for process PSS.

Linux process memory definitions ↗ · Cgroup accounting ↗ · AWS allowance counters ↗

Repeat the resource study

Create the temporary client described in deploy/aws/README.md, then run the commands below. Retrieve the samples before deleting only the temporary client stack.

uv run --with boto3 python deploy/aws/resource_run.py build/aws/benchmark-config.json
python3 deploy/aws/resource_report.py build/aws/resource-pass