Silent Authentication Load Test

Test Overview

Item	Details
Test Date	December 11, 2025
Target Endpoint	`GET /authorize?prompt=none`
Purpose	Measure maximum throughput and performance limits for silent authentication

Test Environment

K6 Cloud Configuration

Component	Details
Load Generator	K6 Cloud (amazon:us:ashburn)
Target	https://conformance.authrim.com
Protocol	HTTPS with PKCE (S256)

Infrastructure

Component	Technology
Compute	Cloudflare Workers
State	Durable Objects (Sharded)
Database	Cloudflare D1
Cache	Cloudflare KV

Sharding Configuration

Durable Object	Standard	Extended
AuthorizationCodeStore	64	128
SessionStore	64	128
TokenStore	48	64

Test Methodology

Scenario

Pre-create 500 user sessions via Admin API
Each VU picks a random session and performs silent auth
Success criteria: HTTP 302 redirect with code parameter

Load Pattern

{
  executor: 'ramping-arrival-rate',
  startRate: 0,
  timeUnit: '1s',
  preAllocatedVUs: 2500,
  maxVUs: 3500,
  stages: [
    { target: 1000, duration: '15s' },  // ramp-up
    { target: 2000, duration: '180s' }, // sustain
    { target: 0, duration: '15s' },     // ramp-down
  ],
}

Test Duration

Ramp-up: 15-20 seconds
Sustain: 180 seconds
Ramp-down: 15-20 seconds
Total: ~3.5 minutes per RPS target

Test Configuration

Authentication Flow

Parameter	Value
response_type	code
prompt	none
code_challenge_method	S256
Session Source	Pre-created cookies

Success Criteria

HTTP 302 status code
code parameter in redirect URL
No HTTP failures

Results - Performance Metrics

Summary (64 Shards)

RPS	Total Requests	HTTP Failures	CF DO Errors	Status
500	73,124	0	13	✅
1,000	146,249	0	0	✅
1,500	219,374	0	0	✅
2,000	292,499	0	0	✅
2,500	365,624	0	0	✅
3,000	445,378	0	0	✅
3,500	521,646	0	0	✅
4,000	599,440	160	1,223	⚠️

128 Shards Test

RPS	HTTP Failures	CF DO Errors	Status
4,000	0	2,112	✅
4,500	287	1,376	⚠️

Effect: 128 shards achieved zero HTTP failures at 4,000 RPS (vs 160 failures with 64 shards)

K6 Client Latency (ms)

RPS	P50	P95	P99	Max
500	406.62	453.68	535.93	1,802
1,000	403.44	453.39	528.18	2,978
1,500	404.12	471.06	529.63	3,320
2,000	404.65	451.84	528.25	3,014
2,500	652.44	793.59	837.71	5,181
3,000	1,243.49	1,582.82	1,641.73	5,482
3,500	614.85	1,631.44	1,727.45	8,077
4,000	458.43	668.64	5,621.68	58,618

Results - Infrastructure Metrics

Worker Duration (ms)

RPS	P50	P75	P90	P99	P999
500	48.14	49.96	51.27	68.10	141.93
1,000	49.44	50.45	51.79	70.15	155.28
2,000	49.74	50.86	52.15	68.27	157.83
2,500	82.86	101.78	116.63	134.59	157.84
3,000	157.24	180.27	194.63	222.03	254.70
3,500	76.16	148.35	193.47	231.27	257.43
4,000	50.23	56.26	74.84	175.81	3,750.00

Worker CPU Time (ms)

RPS	P50	P75	P90	P99	P999
500	2.24	2.78	3.23	8.01	21.93
1,000	2.35	2.78	3.17	10.47	23.27
2,000	2.34	2.83	3.45	6.40	20.30
3,000	2.25	2.85	3.55	5.80	17.38
4,000	2.27	2.83	3.53	13.68	15.55

Key Finding: CPU time remains stable (2-3ms P50) regardless of RPS

Durable Objects Wall Time (ms)

RPS	Total DO Req	DO Errors	P50	P75	P90	P99
500	186,520	13	122.04	180.60	320.99	1,779
1,000	399,282	0	170.87	179.58	189.45	556
2,000	747,073	0	169.92	177.78	183.42	321
2,500	1,097,668	0	180.54	193.23	408.75	645
3,000	1,336,694	0	183.39	731.48	1,040.33	1,312
3,500	1,565,263	0	178.94	190.19	690.62	1,381
4,000	1,796,556	1,223	172.92	182.15	190.99	442

Sharding Analysis

64 vs 128 Shards at 4,000 RPS

Metric	64 Shards	128 Shards	Improvement
HTTP Failures	160	0	✅ Eliminated
clientDisconnected	160	0	✅ Eliminated
DO Errors	1,223	2,112	⚠️ Increased
Max Error-Free RPS	3,500	4,000	+500 RPS

Note: DO Errors increased due to cold start costs of new shards. This stabilizes after warm-up.

128 Shard Configuration Details

RPS	Worker P99	DO P99
4,000	812ms	1,301ms
4,500	3,750ms (timeout)	1,445ms

Capacity Recommendations

Usage	Recommended RPS	Rationale
Normal Operation	≤2,000	P99 < 600ms, 0% errors
Peak Handling	≤3,000	P99 < 1,700ms, 0% errors
Absolute Limit	≤3,500	P99 < 1,800ms, 0% DO errors
Hard Limit (128 shards)	~4,000	HTTP failures begin

Key Findings

1. Worker is Lightweight

CPU time is 2-3ms P50 across all RPS levels - the Worker code itself is not the bottleneck.

2. DO Queuing is the Bottleneck

At high RPS, DO wall time increases dramatically due to request queuing, not processing time.

3. 3,500 RPS Achieved with Zero Errors

With 64 shards, the system handles 3,500 RPS with no HTTP failures and no DO errors.

4. Sharding Extends Capacity

Increasing from 64 to 128 shards pushes the error-free limit from 3,500 to 4,000 RPS (+14%).

5. 4,500 RPS is the Hard Ceiling

Even with 128 shards, 4,500 RPS causes timeouts (Worker Duration hits 3.75s ceiling).

Performance Visualization

xychart-beta
    title "RPS vs K6 P95 Latency (ms)"
    x-axis [500, 1000, 1500, 2000, 2500, 3000, 3500, 4000]
    y-axis "Latency (ms)" 0 --> 2000
    bar [454, 453, 471, 452, 794, 1583, 1631, 669]

Note: 4,000 RPS value (669ms) excludes timeout requests. Latency begins rising at 2,500 RPS.

Bottleneck Analysis

Layer	500-2000 RPS	2500-3500 RPS	4000 RPS
Worker CPU	Stable (2-3ms)	Stable (2-3ms)	Stable (2-3ms)
Worker Duration	Stable (50ms)	Rising (80-160ms)	Spike
DO Wall Time	Stable (170-320ms)	Rising (600-1400ms)	Errors
Verdict	Headroom	Under load	At limit

Test Execution Details

K6 Cloud Run URLs

RPS	K6 Cloud URL
500	https://authrim.grafana.net/a/k6-app/runs/6297034
1,000	https://authrim.grafana.net/a/k6-app/runs/6297223
2,000	https://authrim.grafana.net/a/k6-app/runs/6297317
3,000	https://authrim.grafana.net/a/k6-app/runs/6297364
4,000	https://authrim.grafana.net/a/k6-app/runs/6298253
4,000 (128)	https://authrim.grafana.net/a/k6-app/runs/6298610

Conclusion

Authrim’s Silent Authentication endpoint achieves:

64 shards: 3,500 RPS with zero errors
128 shards: 4,000 RPS with zero errors (+14% improvement)

Recommended Settings:

Normal operation: 64 shards (up to ~3,000 RPS)
High load: 128 shards (up to ~4,000 RPS)

The Cloudflare Workers + Durable Objects architecture demonstrates effective horizontal scaling through DO sharding. However, 128 shards appears to be a practical upper limit due to cold start overhead.