viniarck
commented
2 months ago I've stress tested with the following cases:
- case 1 -> 1 EVC with failover
- case 2 -> 402 EVCs with failover
- case 3 -> 1 static EVC
- case 4 -> 102 static EVCs
I've also collected convergence metric during a link_down for these cases:
In summary, telemetry_int is performing relatively similar to mef_eline, it's always going to be slightly worse then mef_eline since it follows it, it's pretty much on par especially for EVCs with failover. For static 100+ static EVCs the convergence takes longer for both mef_eline and telemetry_int:
| case 1 -> 1 EVC with failover | Initial Gbps | Avg Gbps / 30 secs | TCP retransmissions | Data plane reaction time (back at initial rate) |
|---|---|---|---|---|
| telemetry_int | 4.78 | 4.66 | 2588 | < 1s |
| mef_eline | 4.78 | 4.66 | 2411 | < 1s |
| ratio (telemetry_int/mef_eline) | 1 | 1 | 1.073413521 | |
| case 2 -> 402 EVCs with failover | Initial Gbps | Avg Gbps / 30 secs | TCP retransmissions | Data plane reaction time (back at initial rate) |
| telemetry_int | 4.78 | 3.72 | 1672 | 5s |
| mef_eline | 4.78 | 3.74 | 1704 | 5s |
| ratio (telemetry_int/mef_eline) | 1 | 0.9946524064 | 0.9812206573 | |
| case 3 -> 1 static EVC | Initial Gbps | Avg Gbps / 30 secs | TCP retransmissions | Data plane reaction time (back at initial rate) |
| telemetry_int | 4.78 | 4.53 | 1823 | < 1s |
| mef_eline | 4.78 | 4.67 | 1941 | < 1s |
| ratio (telemetry_int/mef_eline) | 1 | 0.9700214133 | 0.9392065945 | |
| case 4 -> 102 static EVCs | Initial Gbps | Avg Gbps / 30 secs | TCP retransmissions | Data plane reaction time (back at initial rate) |
| telemetry_int | 4.78 | 2.65 | 7059 | 14 s |
| mef_eline | 4.78 | 2.66 | 5805 | 14 s |
| ratio (telemetry_int/mef_eline) | 1 | 0.9962406015 | 1.216020672 |
This is a stress test task placeholder.
I'll use this use to document the next results of stress tests with the events being implemented on https://github.com/kytos-ng/telemetry_int/issues/90
iperf3too (it should perform similar to what has been shared before on Slack)