Hardware Requirements
A common question about Load Tester™ is "what type of hardware do I need to test my website?". The short answer is that even with old hardware, Load Tester™ can generate a massive amount of load. An old machine can simulate up to 500 VUs and a modern machine can simulate over 1500 VUs, easily saturating a 100Mb network.
The long answer depends on the goal of the tests and the nature of the testcases, so an accurate answer would require detailed knowledge of the tests you intend to run. Using a internal reference testcase yielded the following estimates. The reference testcase has 25 pages, 389 URLs, a total size of 2.5MB and the duration is just under 2 minutes. It does not use SSL, cookies or have any dynamic fields or validators (beyond the defaults) applied to it.
Note that these estimates are based on a stand-alone load engine. The local load engine (which is built into the Load Tester™ GUI) will have signifantly lower capacity due to the additional functions it must perform during a test.
Best-case engine capacity:
The following examples illustrate the maximum number of VUs that could be generated on a realistic, but simple, testcase.| Hardware | Virtual Users |
Result | Metrics |
|---|---|---|---|
| Dual 800MHz PIII and 1G RAM | 500 | Limited by CPU | 9M bytes/sec, 100 pages/sec, 1600 hits/sec >80% CPU, 10% memory |
| 2.8 GHz Xeon and 2G RAM | 1800 | Limited by CPU | 37M bytes/sec, 380 pages/sec, 5900 hits/sec >80% CPU, 20% memory |
| Dual 2.8 GHz Xeon and 2G RAM | 1900 | Limited by OS threads | 40M
bytes/sec, 400 pages/sec, 6000 hits/sec <40% CPU, 20% memory |
These examples were run using the Instant Load Engine boot disk version 3.5, which runs a Linux 2.6 SMP kernel. In the above example, the PIII-class dual-processor machine could generate at most 500VUs using the reference testcase. With the single-processor Xeon, the threading overhead started increasing non-linearly after 1200 VUs but still scaled to 1800 VUs. With the dual-processor Xeon, the OS thread limit was exceeded shortly after 1900 users. Note that during these tests, there was almost no degredation in server performance (when server performance degrades, it eases the load on the engines since each VU spends more time waiting on responses).
What items affect engine capacity?
- SSL will raise CPU usage, lowering the capacity of an engine by ~40%.
- The content-processing searches required for dynamic fields will increase both memory and CPU utilization. The affect is dependent on the number of dynamic fields, the type of content inspection required to locate the values and the size of the content searched.
- Large file uploads or downloads (such as audio/video) will greatly increase memory utilization in the engine.
- Each content validator will require additional CPU, depending primarily on the size of the content searched and the location of the target in the content.
- When running a high number of simultaneous tasks, the overhead of the operating system's context-switching management becomes a factor. On Linux (2.6 kernel), this overhead becomes non-linear somewhere around 1200 VUs.
Best-case engine capacity with SSL:
Here is some examples of engine capacity with SSL. As before, these tests were run with the Instant Load Engine boot disk. The reference testcase was nearly identical to the above-described testcase but all pages and resources were accessed vi SSL URLs.| Hardware | Virtual Users |
Result | Metrics |
|---|---|---|---|
| Dual 800MHz PIII and 1G RAM | 275 | Limited by CPU | 6M bytes/sec, 60 pages/sec, 1000 hits/sec >80% CPU, 10% memory |
| 2.8 GHz Xeon and 2G RAM | 710 | Limited by CPU | 15M bytes/sec, 150 pages/sec, 2500 hits/sec >80% CPU, 20% memory |
| Dual 2.8 GHz Xeon and 2G RAM | 1070 | Limited by CPU | 21M
bytes/sec, 230 pages/sec, 3500 hits/sec >80% CPU, 35% memory |
