[Openstack] [SWIFT] Bad replication performance after adding new drives
klaus.schuermann at mediabeam.com
Tue Feb 10 10:33:27 UTC 2015
I know your blogpost very well and I made a lot of changes to my cluster.
I set the vfs_cache_pressure to 10 and moved container- and account-server to SSD harddrives.
The normal performance for object writes and reads are quite ok.
But why takes moving some partions to only two new harddisks so much time?
Will it be faster if I add more memory?
Von: Robert van Leeuwen [mailto:Robert.vanLeeuwen at spilgames.com]
Gesendet: Dienstag, 10. Februar 2015 11:17
An: Klaus Schürmann; 'openstack at lists.openstack.org'
Betreff: RE: [Openstack] [SWIFT] Bad replication performance after adding new drives
> I have a swift cluster with 7 storagenodes. The cluster has about 360 million objects with an average size of 170kB.
> I added 2 new harddrives to my cluster and now it takes weeks for one replication round:
> Is this a normal behavior?
> DELL PowerEdge R720xd
> 32 GB RAM
> Harddisks 3TB/4TB SATA
I created a blogpost about this a while ago when we hit performance issues with lots of small files with a few things you can look at and tune:
As a small update on the blogpost above:
We removed all config 1.X nodes to nodes with config 2.0.
We stopped using flashcache, since it was a bit of a pain and without it is also fine for us.
As a result a full sync when adding an empty new node is now down to 1-2 days while this used to be 2+ weeks.
Robert van Leeuwen
More information about the Openstack