Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
> -----Original Message----- > From: drbd-user-bounces at lists.linbit.com [mailto:drbd-user- > bounces at lists.linbit.com] On Behalf Of Bart Coninckx > Sent: Monday, March 31, 2014 6:51 AM > To: Arnold Krille > Cc: drbd-user at lists.linbit.com > Subject: Re: [DRBD-user] Single versus multiple DRBD devices > > > On 28 Mar 2014, at 22:04, Arnold Krille <arnold at arnoldarts.de> wrote: > > > On Fri, 28 Mar 2014 10:40:28 +0100 Bart Coninckx > > <bart.coninckx at telenet.be> wrote: > >> All, > >> > >> > >> building a storage cluster which will offer SCST targets. These will > >> be LVM based (nested LVM probably). In terms of performance, what is > >> the most optimal strategy: one big DRBD device where the different > >> Logical Volumes are installed on or a DRBD device per LV? If a resync > >> needs to be done, the latter will probably better, as it allows us to > >> choose which devices should be resynced first. > > > > I do have several drbd volumes instead of one big. Then you can > > choose which ones to sync first. And if your volumes are on different > > disks, the order can be defined to sync volumes from each disk one > > after the other but volumes from different disks in parallel. That way > > one can max out a 2x10G connection with three volumes from three disks > > all reading/writing sequential, instead of the disk stepping in its own > > way by reading from several volumes on the same disk while the other > > disks are idle... > > > > And several drbd-volumes can fail-over individually. > > > > Have fun, > > > > Arnold > > > Sounds reasonably! Thank you for your input! > > BC > What Bart is suggesting is matching your drbd definition to your physical environment. Your OP was about the logical environment. You'll probably have a combination of factors drive your design. For example, I have four DRBD resources servicing about 20 VMs. The four DRBD resources mirror my physical environment and the VMs are spread around on these four devices to minimize the impact of a failure or slow-down on any one. Dan