On 12/07/2010 04:39 PM, Jeff Sturm wrote: >> -----Original Message----- >> From: linux-cluster-bounces@xxxxxxxxxx > [mailto:linux-cluster-bounces@xxxxxxxxxx] >> On Behalf Of Fabio M. Di Nitto >> Sent: Tuesday, December 07, 2010 5:42 AM >> To: linux-cluster@xxxxxxxxxx >> Subject: Re: GFS on AOE >> >> The problem being that AOE (as you suspect) adds a different level of > caching. > > Note however that the AoE protocol does not specify caching, except for > optional asynchronous writes. (The aoe Linux module does not utilize > asynchronous writes.) In our testing we did have several issues with the setup described above and trimmed down the problem to have: node A -> controller/driver X -> harddisk node B -> (any network block device, including AOE) -> controller/driver X -> harddisk. And isolated the issue to the asymmetry of the setup. > > Nevertheless, the configuration suggested by the OP is unusual, and > won't be very useful in my opinion. Having node B rely on a hard disk > in node A leaves node A as a single point of failure. Yes absolutely. It does not make any sense, but for basic testing is "good enough". > > We use GFS over AoE extensively, and find it works well. However we use > an AoE target that runs independent of the cluster and provides > high-availability on its own. Yes, this is also tested and works fine. As you might have noticed in the FAQ, we only describe the asymmetric setup as "not-working". Fabio > > -Jeff > > > > -- > Linux-cluster mailing list > Linux-cluster@xxxxxxxxxx > https://www.redhat.com/mailman/listinfo/linux-cluster -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
