-
Notifications
You must be signed in to change notification settings - Fork 149
Home
##Throughput Tests
###Test 1
####Summary
We used the ouroboros-release to redirect PWS firehose data into a standalone Loggregator deployment (described below).
There were two clients (volley
job from ouroboros-release
) reading from our TrafficController.
The maximum throughput of data between Doppler and TrafficController was 1 Million envelopes per 60 secs as seen by our Datadog Dashboard.
For more info, refer to this tracker story
####Environment AWS - PWS
c3.large
####Deployment
Firehose Connection (ouroboros
) -> 1 Metron -> 1 Doppler -> 1 TrafficController -> 2 Firehose Connections (volley
)
###Test 2
####Summary Similar test setup to Test 1 above. However, we added grpc connector logic between Dopplers and TrafficControllers. This is to test the overall throughput of gRPC with production load from PWS.
####Relevant Stories Reduce network connections for a client reading application logs from Doppler Traffic controller finds and consumes new doppler endpoints
####Environment
- The test was run in AWS.
- All instances were
c3.large
- Number of producers (
ouroboros
): 1-
ouroboros
reads from PWS firehose and sends the raw data through to Metron Agent.
-
- Number of client connectors (
volley
): 2 - Number of Dopplers: 1
- Number of Traffic Controllers: 1
####Results We found a bug where the gRPC path doesn't have a truncating buffer in Doppler. Hence, Doppler is having a hard time receiving and sending data.