Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 2017-08-25 02:49 AM, Gionatan Danti wrote: > Hi list, > in my endless search for information :) I was testing DRBD in dual > primary mode + GFS2. The goal is "the same": running a replicated, > 2-node KVM setup where disk images are located on the GFS2 filesystem. > > I know how to integrate DRBD with corosync/pacemaker, but I wonder how > the system will perform under load. More specifically, it is my > understanding that GFS2 inevitably has some significan overhead compared > to a traditional, one-node filesystem, and this can lead to decreased > performance. > > This should be especially true when restarting (or live-migrate) a > virtual machine on the other host: as the first node has cached > significan portion of the vm disk, GFS2 will, on every first read on the > new host, fire its coherency protocol to be sure to update the first's > node in-memory cached data. The overhead should be lowered by not using > cache at all (ie: cache=none, which implies O_DIRECT), but this will > also case degraded peformance in the common case (ie: when all is > working correctly and VMs run on the first node only). > > So I ask: has some of you direct experience with a similar setup? How do > you feel about? Any other suggestions? > Thanks. Our Anvil! project (https://www.alteeve.com/w/Build_an_m2_Anvil!) is basically this, except we put the VMs on clustered LVs and use gfs2 to store install media and the server XML files. I would NOT put the image files on gfs2, as the distributed locking overhead would hurt performance a fair bit. -- Digimer Papers and Projects: https://alteeve.com/w/ "I am, somehow, less interested in the weight and convolutions of Einstein’s brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops." - Stephen Jay Gould