[gpfsug-discuss] IO-500 and POWER9

Simon Thompson S.J.Thompson at bham.ac.uk
Mon May 13 14:11:06 BST 2019


Hi,

I was wondering if anyone has done anything with the IO-500 and POWER9 systems at all?

One of the benchmarks (IOR-HARD-READ) always fails. Having slack’d the developers  they said:
“It looks like data is not synchronized” and “maybe a setting in GPFS is missing, e.g. locking, synchronization, ...?”

Now I didn’t think there was any way to disable locking in GPFS. We tried some different byte settigns for the read and this made the error go away which apparently indicates “lockicg issue -> false sharing of blocks”.

We found that 1 or 2 nodes = OK. > 2 nodes breaks with 2ppn, > 2 nodes is OK with 1ppn.

(We also got some fsstruct errors when running the mdtests … I have a PMR open for that).

Interestingly I ran the test on a bunch of x86 systems, and that ran fine.

So … anyone got any POWER9 (ac922) they could try see if the benchmarks work for them (just run the ior_hard tests is fine)? Or anyone any suggestions?

These are all running Red Hat 7.5 and 5.0.2.3 code BTW.

Thanks

Simon
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20190513/7146b1e0/attachment-0001.htm>


More information about the gpfsug-discuss mailing list