Skip to content
Warren Fernandes edited this page Sep 22, 2016 · 7 revisions

##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.

Clone this wiki locally