It is less that 4 gigs is not enough but rather if the incoming ram buffer is 4 gigs and my test data is 1 gig something is still not working. In theory I should be able to ingest at line speed until the ram buffer is full but something is slowing down.
Unfortunately I can’t edit the title post but my signature has the accurate hardware. Client and server are both using 100G fiber. For me this is less about the write speed and more about finding the bottleneck as read and write are currently identical.
1 Like
So it took less than half a second. TCP/IP doesn’t utilise full bandwidth instantly. Perhaps it could be the reason why you didn’t see the desired 10+GB/s. You can try to transfer a 4GB file (or slightly less just in case) and see whether the numbers got bigger.
However, IMO, with sub-second transfers, windows explorer’s speed reporting (I assume you used it for the test) should not be taken seriously. OTOH, I don’t know how to properly benchmark these scenarios.
So you have 16 HDDs in a stripe? Just to ensure, can you post the output of sudo zpool list -v <your-pool-name>?
What was the number of tests? Did Q8 and Q1 have significantly different values?