[lustre-discuss] Benchmarking Lustre, reduce caching

[Vicker, Darby J. (JSC-EG111)[Jacobs Technology, Inc.]]

I'd recommend using the io500 benchmark.  This runs both bandwidth and metadata tests and has checks to run prevent caching from tainting the results (like forcing it to run for a specified amount of time).  I've found it useful for benchmarking all of our file systems (lustre, NFS, local, etc.) and is very useful for comparing performance.

https://io500.org/

Darby

From: lustre-discuss <lustre-discuss-bounces at lists.lustre.org> on behalf of Athinagoras Skiadopoulos via lustre-discuss <lustre-discuss at lists.lustre.org>
Reply-To: Athinagoras Skiadopoulos <askiad at stanford.edu>
Date: Wednesday, May 19, 2021 at 1:36 PM
To: "lustre-discuss at lists.lustre.org" <lustre-discuss at lists.lustre.org>
Subject: [EXTERNAL] [lustre-discuss] Benchmarking Lustre, reduce caching

Hello,

We are using Lustre in our cluster as an external service and
would like to measure the write/read performance we can achieve.

Right now, we are creating a number of files, and iteritavely:
- pick a file to open
- write/read it in 1MB chunks (block size)
- close it

Each of our servers has a 3000Mbps bandwidth,
but we are often reporting higher read/write throughput.
There should be some caching involved.
What would be a good way to alleviate this?

Thank you in advance.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20210519/0853e531/attachment.html>