[Lustre-discuss] what performance is expected with more ost
valiantljk at gmail.com
Wed Jul 10 23:59:08 PDT 2013
I'm doing several tests,
Test 1: 1 process from 1 compute node, accessing 1 stripe on 1 ost.
Test 2: 10 MPI process from 1 compute node, each process accessing 1
stripe on different ost.
I think Test 2 will cost more, because even the 10 ost can respond
concurrently, there will be more latency, i.e., network latency, etc.
But I think such latency shouldn't dominate the I/O cost.
the experimental result shows me a linear trend as I increase the ost
My question is how the multiple ost interact with the same client?
If multiple ost sends back the message to the same client, is there any
contention issue (e.g., waiting queue on the client)?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the lustre-discuss