[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] RHCS Coraid/ATAoe compatibility



The newer units are actually quite a bit faster.  Also, I highly recommend connecting your Coraids in a dual-network / not-cross-connected fashion.

In other words, given machines A,B,C each with eth2 and eth3; switches Alpha and Beta; and Coraids Aleph and Beth each with internal connections Uno and Dos, that you connect them as follows.  For switch Alpha, plug in A-eth2, B-eth2, C-eth2, Aleph-Uno, Beth-Uno.  For switch Beta, plug-in A-eth3, B-eth3, C-eth3, Aleph-Dos, Beth-Dos.  Do not cross-connect Alpha and Beta!  In this configuration, with Jumbo frames, we get pretty good performance.

I've experimented with Quorum Disks and find that I quite like them, although we've just never needed them as we also fence with APC power bars.  We use GFS extensively and it works quite well.

On Mar 28, 2007, at 1:19 AM, Ramon van Alteren wrote:

Franco Martin Bladilo wrote:
We are considering the option to build a 4 node HA cluster using RHCS and Coraid products (SR1521). ATAoE seems to be the most cost effective solution for our case and I was wondering if there are any comments/ feedback for such configuration.
We're extensively using gfs + coraid 1521's in our setup.
Currently running 3 5-node clusters with 2 - 5 coraid devices attached.

You should not expect the same performance from gfs over AoE as listed in the coraid docs.
Depending on the model version you will not get better performance by enabling the secondary nic on the coraid devices, rather the opposite.
They claim to have solved this with their newer models but I haven't tested yet.

Make sure you use the coraid supplied AoE drivers, performance is much better than the in-kernel ones.
Apart from that, install the cec utility, it allows you to login on the coraids remotely and thus enables monitoring of raid arrays by a NMS of your choice

On the whole we are reasonable statisfied with the solution we managed to build on top of gfs and coraids.
It certainly is cost-effective.
Is the quorum partition issue (shared concurrent access) still a limitation for these appliances? Can these "service" partitions be formatted using GFS? Any other workarounds?
Can't comment, we use the dlm + fencing via APC powerbars.

Regards,

Ramon van Alteren

--
Linux-cluster mailing list

-- 
Jayson Vantuyl
Systems Architect



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]