Hi,

I am trying to tune Undertow for our high traffic application and as part of that I am running a load test from a client host (different from the host where undertow is running) that calls endpoints that involve writing multiple 1-1.5 MB files to disk.

These are two scenarios that I run the load test:

1) Single client with 100 concurrent threads , each thread uploads  files of size > 1 MB and repeats the process for 2 minutes

2) Two clients each with 50 concurrent threads, each thread uploads  files of size > 1 MB and repeats the process for 2 minutes

Observations:

In the first scenario, I am seeing a latency of about 5 secs per upload.

In the second scenario, I am only seeing a latency of 2.5 secs per upload.

I am calculating the latency at the client.

I've checked the number of connections open to the server in both cases and they are around 100 at the server.

Is there any setting on the server side that would explain this behavior? Since the number of connections are same shouldn't the latency be same more or less in both scenarios?

I have 32 IO Threads and 160 Worker threads with a Backlog of 1000 configured at the server.

Thank you for all the help.

-Ravi