Skip to content

Latest commit

 

History

History
259 lines (240 loc) · 6.17 KB

loadtestresults.md

File metadata and controls

259 lines (240 loc) · 6.17 KB

Loadtest results

Below, we share the results of our load tests run in IONOS Cloud, which took place with a Jitsi single-shard-setup.
The loadtest test server configuration for both cases was set to 16 Cores Intel, 16 GB RAM, 8 server. All simulated participants did send a high-quality video-stream (1280x720 pixels, 60 frames per second) and sound. There are two different test cases we investigated:

1 One big conference

All test servers are connecting to one conference. This was achieved by a slight variation of run_loadtest.sh. We have tested two different setups for the limits and request of the JVB pods:

1.1 CPU limit is 4 cores, Memory limit is 1 GB

We have further varied the number channelLastN in web-configmap.yaml. It decides how many video streams are forwarded to every conference participant, where only the video of the most active N users are used.

1.1.1 ChannelLastN not set

Here are the results for channelLastN not set, that means all videos are streamed to all participants:

Participants total CPU usage test server video quality frame rate sound num participants stable Max Memory Usage per JVB Max CPU usage per JVB Network Transmitted max CPU-Throttling Run order Error Messages JVB and Remarks
8 ~ 7 % very high very high very good stable 830 MB 1,9 cores 4,7 MB 0% 2
14 ~ 7 % very high very high very good stable 830 MB 3,5 cores 10,7 MB 5% 1
24 ~20 % okay partly lagging very good stable 940 MB 4 Cores (pod limit) 14,6 MB 80% 3 Conference was moved to another JVB. Observed warnings: TCC packet contained received sequence numbers: 58852-59404. Couldn't find packet detail for the seq nums: 58852-59404. Latest seqNum was 63013, size is 1000. Latest RTT is 4310.241456 ms.

1.1.2 ChannelLastN set to 5

Here are the corresponding results for channelLastN set to 5:

Participants total CPU-usage test server video quality frame rate sound num participants stable Max Memory Usage per JVB in MB Max CPU-Usage per JVB Max Network Transmitted MB CPU-Throttling Error Messages JVB and Remarks Run order
8 ~ 7% very high very high very good stable 760 MB 1,9 Cores 4,1 MB 0% 5
16 ~17% very high very high very good stable 760 MB 3,5 Cores 9,1 MB 5% 4
24 ~20% high very high very good stable 800 MB 4 Cores (pod limit) 8,4 MB 80% 6 video quality still high, network traffic seems to be manageble for JVB

1.2 CPU limit is 8 cores, Memory limit is 1 GB

1.2.1 ChannelLastN not set

Not tested.

1.2.2 ChannelLastN set to 5

Participants total CPU-usage test server video quality frame rate sound num participants stable Max Memory Usage per JVB in MB Max CPU-Usage per JVB Max Network Transmitted MB CPU-Throttling Error Messages JVB and Remarks Run order
32 30% very high partly lagging very good 3 participants leaving 910 4,3 17 0 WARNING: 1891053881 large jump in sequence numbers detected (highest received was 52436, current is 52603, jump of 166), not requesting retransmissions , WARNING: TCC packet contained received sequence numbers: 21287-21371. Couldn't find packet detail for the seq nums: 21287-21371. Latest seqNum was 24965, size is 1000. Latest RTT is 4568.5094 ms. , INFO: timeout for pair: 217.160.210.116:30301/udp/srflx -> 157.97.109.136:52047/udp/prflx (stream-59888d67.RTP), failing. Error Messages indicate network issues. The traffic might be already to high. Conference stays mostly stable, though 1
16 15% very high very high very good stable 960 1,8 8,5 0 CPU usage inconsistent with other observations 2

2 Eight smaller conferences in parallel

Each of the eight test servers start exactly one conference with a certain number of participants. The script run_loadtest.sh was used for this purpose.

2.1 CPU limit is 4 cores, Memory limit is 1GB

There were two JVBs running for this test.

Participants per conference CPU-usage test server video quality frame rate sound num participants stable Max Memory Usage per JVB in MB Max CPU-Usage per JVB Max Network Transmitted MB (both JVBs) CPU-Throttling Error Messages JVB and Remarks Run order
5 ~ 50% very high very high very good stable ~ 900 MB ~ 2.5 Cores 9,1 MB; 6,0 MB 0% 4
10 ~75% high high very good several participants lost ~ 860 MB ~ 3,9 Cores (nearly pod limit= 12,5 MB ; 10,5 MB 50% 3

Tests with bigger conference sizes have not been carried out due to the high resource demands of the test servers.