Rased performance: measured numbers, not promises
These figures were actually measured in Rased's test environment. No uptime percentages and no service-level agreements: every figure carries its date, its method and its limits.
Method and figures
Every figure on this page: measured in Rased's test environment, 3 October 2026, by the method and within the limits set out below.
How we measured
Measured in Rased's test environment on 3 October 2026: one service process limited to half of one processor core, synthetic data, up to 20 simulated users, and a direct connection without internet delay. This is a measurement, not a service-level commitment.
- Synthetic data at two sizes: small (50 organisations, 250 devices, 63,000 attack events) and medium (300 organisations, 2,400 devices, 577,800 attack events).
- Simulated users who really sign in and use the same calls the pages use.
- Customer-page figures are the median of three rounds. No outside service was called: AI and outside data sources were switched off, and so were the monitoring and security engines and the background checks (health checks and vulnerability matching) during measurement.
- The 20 calls shown are the customer-page data calls that do not read every device; the screens that do are shown separately.
- p95 means 95% of requests were faster than this time, and p99 that 99% were.
What we did not measure
- More than 300 organisations and 2,400 devices: a larger size was built but has not yet been run.
- More than 20 users at the same time.
- The internet delay between you and us, and TLS: we measured over a direct connection.
- The engines' own time: the monitoring and security engines were switched off during measurement, so the figures do not include it. The device-metrics and device-hardware calls, which read from them, answered “unavailable” at once, so we do not show them.
- How the service behaves over weeks: our rounds lasted minutes, not days, so we do not claim there are no memory leaks.
- The speed of this public website itself: not measured yet.
Load
≈33 requests a secondWith 20 simulated users at once, with zero errors.
Measured in Rased's test environment, 3 October 2026Light page calls
1.6–8.5 msp95 of 20 customer-page calls that do not read every device. Screens that read every device are heavier: between 16.5 and 70 ms.
Measured in Rased's test environment, 3 October 2026Memory
135–140 MBProcess memory after normal use, against a 600 MB limit set on the measured service.
Measured in Rased's test environment, 3 October 2026Log analysis
≈0.4 secondUntil the rule-based report is ready for a 10,000-line log, without the AI step.
Measured in Rased's test environment, 3 October 2026Light customer-page response time as the data grows
p95 of each of 20 light customer-page calls, at two data sizes. Each dot is one call; dots further right are slower. When the data grew from 50 to 300 organisations, no call's p95 rose by more than 0.9 ms.
Show the numbers as a table
| Call | Small data | Medium data | Change |
|---|---|---|---|
| Devices list | 1.6 | 1.8 | +0.2 |
| Device applications | 2.0 | 1.8 | -0.2 |
| Device audit | 5.2 | 4.0 | -1.2 |
| Device events | 2.2 | 2.5 | +0.3 |
| Dashboard | 3.0 | 3.2 | +0.2 |
| Vulnerabilities | 3.2 | 3.7 | +0.5 |
| Attack map | 2.7 | 3.0 | +0.3 |
| Attack sources (7 days) | 4.5 | 5.0 | +0.5 |
| Attack sources (30 days) | 6.1 | 7.0 | +0.9 |
| Issues | 2.0 | 2.1 | +0.1 |
| Issue detail | 2.2 | 2.2 | 0.0 |
| Log analyser catalogue | 8.5 | 4.6 | -3.9 |
| Log reports | 2.1 | 2.1 | 0.0 |
| Account | 2.7 | 3.1 | +0.4 |
| Alert settings | 1.8 | 1.8 | 0.0 |
| Reports list | 1.7 | 2.4 | +0.7 |
| Team | 5.8 | 1.8 | -4.0 |
| Telegram link | 1.6 | 1.7 | +0.1 |
| Guide (Arabic) | 3.6 | 4.1 | +0.5 |
| Guide (English) | 3.6 | 3.9 | +0.3 |
Values in milliseconds. Two calls are not shown: device metrics and device hardware, because they read from the engines that were switched off and answered “unavailable” at once (2.4 and 1.7 ms).
Rising load: requests per second
From 4 to 20 simulated users with normal think time. Zero errors at every step, and the portal used between 2.3% and 4.8% of one core. Twenty is the most we tested, not a limit of the platform.
- Small data
- Medium data
Simulated users at once · vertical axis: requests per second
Show the numbers as a table
| Users | Small data | Medium data | ||||||
|---|---|---|---|---|---|---|---|---|
| req/s | p95 | p99 | CPU % | req/s | p95 | p99 | CPU % | |
| 4 | 6.5 | 99.7 | 116.2 | 2.3 | 6.1 | 31.4 | 3243.9 | 2.4 |
| 8 | 13.4 | 26.0 | 104.7 | 2.3 | 12.3 | 17.7 | 3232.3 | 2.5 |
| 12 | 20.5 | 15.5 | 99.1 | 2.9 | 19.3 | 18.5 | 38.8 | 3.1 |
| 16 | 26.6 | 16.5 | 39.0 | 3.6 | 25.6 | 22.0 | 45.5 | 4.8 |
| 20 | 33.7 | 12.5 | 36.1 | 3.8 | 32.8 | 14.8 | 44.9 | 3.8 |
p95 and p99 are in milliseconds. The figures cover all requests and include the platform-staff pages, which are the slowest, so they are conservative for customers. Errors: zero at every step. The p99 at 4 and 8 users on the medium size (about 3.2 seconds) comes from a single slow request to one of the platform-staff pages, not to a customer page. The highest p95 on the small size belongs to the first step (4 users, 15 seconds); our report does not isolate its cause, so we show it as measured.
Log analysis time
Time until the rule-based report is ready, without the outside AI step, which was switched off in the test. Measured at 100 ms resolution. The 10,000-line file was accepted in 176 to 205 ms. A real analysis with AI takes longer.
Show the numbers as a table
| Log | Size (KB) | Accepted in | Report ready in |
|---|---|---|---|
| 1,000 lines | 87 | 51–59 | 152–160 |
| 10,000 lines | 869 | 176–205 | 362–396 |
Memory
Memory of the measured portal process: it started near 94 MB and settled after normal use, peaking in the toughest round, with no growth across three consecutive rounds. The 600 MB is the limit set on the measured service.
Show the numbers as a table
| State | MB |
|---|---|
| At process start | 94 |
| After normal use | 135–140 |
| Peak in the toughest round | 191–197 |
| Limit set on the measured service | 600 |
Screens that read every device are heavier
The Telegram device list and the organisation and device reports read every device in the organisation, so they grow with the data: p95 between 16.5 and 70 ms in our measurement, against 1.6 to 8.5 ms for the light calls.
| Screen | Small data | Medium data |
|---|---|---|
| Telegram device list | 16.5–19.3 | 24.9–27.2 |
| Telegram organisation report | 37–46 | 52.5–64.2 |
| Printed device report | 36.1–70 | 27.2–33.2 |
| Printed organisation report | 42.2–59.1 | 53.7–57.8 |
Values in milliseconds. Each range covers the Arabic and English versions (median of three rounds), which differ slightly.
Measured in Rased's test environment, 3 October 2026Want a figure for your own case?
Ask us what matters to you, and we will tell you what we measured and what we did not.