The amount of data you're testing is far too small. Try upping it 10x so
you have a longer run time and you achieve a more reasonable average.
On Tue, Sep 28, 2021 at 6:02 PM Nagmat Nazarov
wrote:
> Dear Engineers,
>
> I have started working on a lustre file system. I have done a couple of
>
Dear Engineers,
I have started working on a lustre file system. I have done a couple of
experiments so far:
On the first experiment I am writing 100 files each 10MB and the batch size
is 4K. I got *704MB/s* bandwidth.
On the second experiment I am (writing 10 files each 10MB and reading 1 10
MB
You are more than right. The IB interface 172.21.164.116 is not
registered, only the TCP one is
- { index: 234, event: add_uuid, nid:
172.21.156.102@tcp1(0x20001ac159c66), node: 172.21.156.102@tcp1 }
- { index: 240, event: add_uuid, nid:
172.21.156.102@tcp1(0x20001ac159c66), node:
Hi Riccardo,
I would check if the OSTs on this OSS have been registered with the correct
NIDs (o2ib1) on the MGS:
$ lctl --device MGS llog_print -client
and look for the NIDs in setup/add_conn for the OSTs in question.
Best,
Stephane
> On Sep 28, 2021, at 9:52 AM, Riccardo Veraldi
>
Hello.
I have a lustre setup where the MDS (172.21.156.112) is on tcp1 while
the OSSes are on o2ib1.
I am using Lustre 2.12.7 on RHEL 7.9
All the clients can see the MDS correctly as a tcp1 peer:
peer:
- primary nid: 172.21.156.112@tcp1
Multi-Rail: True
peer ni:
-
On the second experiment, you’re writing a total of 1000MB and reading 100MB.
It could simply be that you’re not putting enough load on the system for long
enough to get full performance.
-Ben Evans
From: lustre-discuss on behalf of
Colin Faber via lustre-discuss
Reply-To: Colin Faber
Dear Lustre community,
LAD'21 starts today!
If you haven't already done so, you can still register for the webinar that
will take place in the next 3 days. The presentations are broadcast twice a day
to cover all the timezones:
- One broadcast at 8AM UTC (10AM CEST Paris, Berlin,