[lustre-discuss] lru_size question

Andreas Dilger adilger at whamcloud.com
Tue Sep 14 16:14:18 PDT 2021


On Sep 9, 2021, at 02:49, Thomas Roth <t.roth at gsi.de<mailto:t.roth at gsi.de>> wrote:

Hi all,

I have checked the lru_size on an (2.12.5) system that has just been restarted.
The defaults have never been touched on that system, and so I see lru_size=0 for all OSTs, on the MDS as on a client, as it should be.
The client also reports
> ldlm.namespaces.MGC10.20.3.0 at o2ib5.lru_size=12800
while the three MDTs of the system also are shown with lru_size=0

The MDS reports
> ldlm.namespaces.MGC10.20.3.0 at o2ib5.lru_size=1600
> ldlm.namespaces.MGS.lru_size=1600
> ldlm.namespaces.hebe-MDT0000-lwp-MDT0000.lru_size=494
> ldlm.namespaces.hebe-MDT0001-osp-MDT0000.lru_size=1600
> ldlm.namespaces.hebe-MDT0002-osp-MDT0000.lru_size=1600
> ldlm.namespaces.mdt-hebe-MDT0000_UUID.lru_size=1600

(hebe being the name of the fs)

These values show up immedeately after starting the MDS.

For static-sized LRU the default is to use 100x core count on each node.  The MGS is always static, since it doesn't need many locks, and less effort to manage LRU size.

Cheers, Andreas
--
Andreas Dilger
Lustre Principal Architect
Whamcloud







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20210914/2ea1cd32/attachment.html>


More information about the lustre-discuss mailing list