Transition to new versions of our PBXware 5 and SERVERware 3 products saw a lot of changes happening in the back. As such, this imposes questions in regards to performance of these products on various hardware for which we did many tests. One of the most important rounds of tests included three common servers, and the scenarios being run suggested use of multiple Serverware VPSe. This meant using various types of SIP calls, which combined, should represent a common Serverware stand alone system in use.

For brevity of testing, and to limit the output of testing stats, we focused on three VPSes:

  1.  First VPS is running simple G711 ulaw calls
  2.  Second VPS is running G711 ulaw calls that are recorded
  3.  Last VPS is running G711 ulaw calls that are transcoded to G729 codec

Again, three VPSes were used since anything more would present an unmanageable data stats that would lose their significance in the testing scenario.

Testing procedure was run in following manner:

  1. - First VPS would initiate first 200 calls
  2. - Statistics data is collected
  3. - While previous calls are running, second VPS make 200 calls
  4. - Statistics are collected again
  5. - While previous 400 calls are running, third VPS makes 200 calls
  6. - Stats are collected
  7. - First VPS makes additional 200 calls,
  8. - And this repeats, until the audio on a control call starts breaking, popping or glitching.

Stats are collected in appropriate tables, and include CPU, RAM, Network and disk IO. This showed what happened with each step of the test and how much of system resources were busy.

Server specifications used in the testing:

1. HP DL380e Gen8

 1. 2x Xeon(R) CPU E5-2420 0 @ 1.90GHz
 2. 12 cores, 24 threads
 3. 32 GB RAM

2. Dell R730xd

 1. 2x Xeon(R) CPU E5-2620 v3 @ 2.40GHz
 2. 12 cores, 24 threads
 3. 32 GB RAM

3. HP DL180 G9

 1. 2x Xeon(R) CPU E5-2650 v4 @ 2.20GHz
 2. 24 cores, 48 threads

Also, for each server we made two different test:

  1.  VPSes have unlimited disk IO speed
  2.  VPSes have severely limited disk IO speed equaling to 1MB/s
  3. | DellR370 Ideal testing Environment PDF
  4. | HPDL180 Ideal testing Environment PDF
  5. | HPDL380 Ideal testing Environment PDF

Of course, these tests are almost synthetic in nature as these calls don't contain background SIP traffic that happens in everyday systems ,like SIP REGISTER, OPTIONS, NOTIFY, etc

To get close to the the real life scenario, we were able to make couple of tests that included REGISTER and OPTIONS SIP packets for each call and the results are present here:

  1. | Ideal testing Environment PDF

In the end we must note that these test were done with technical limitations in mind, where saturated hardware resources made immediate impact on control call audio.

In real world usage, we always set the limit to 60% of calls made in these tests which gives a safe performance limit on such hardware.

 NOTE: For small deployments, where one server would be used in a SERVERware Standalone configuration, one server configured as the previously mentioned above, should provide ability to safely make between 80 and 100 concurrent calls in mixed scenario usage case. HP Gen8 and Gen9, Dell R730xd can be obtained worldwide as a cost effective hardware solutions to upgrade from smaller SERVERware 1.x deployments to newer 3.x generation software, which supports newest versions of our PBXware VoIP software. 

Needed Hardware for 2600 Extensions In real world usage