LVM for OS/app or hardware solution?

[Mike Hassell]

I have been tasked to determine the best route for fault tolerance vs. performance for a protein folding application (CPU/Memory intensive). Unfortunately I don't have any details on the app itself yet, as I've been thrown into this thing in the last two days, but should have more info on that tomorrow.

Currently I'm looking at an HP9000 rp8420-32. The developers who I am working with have questions regarding the optimal configuration of the OS and application. My experience in this area is limited and I've always installed the OS an app using LVM on the local disks, keeping the datasets on a hardware RAID solution in the backend (typically EMC). The developers are worried about the latency that comes along with software mirroring for the OS and app itself and the redundancy it provides.

I would like to know the pros and cons of running the OS and app on the backend EMC or similar hardware array vs. running it on the local disks.

I know there are many factors that come into play when trying to determine the best route for this scenario. I hope to have much more detailed information to offer in the next few days when they give me info on the app itself.

Since this app takes a month or so to complete an average job, fault tolerance is first and foremost, but I'd like to find the best balance between fault tolerance and performance.

Your thoughts?

Thanks,
Mike

[Patrick Wallek · Accepted solution]

I would put the OS on the internal drives and mirror them.

I would then put the application and all the data on whatever external raid solution you use.

You will see very little latency with LVM mirroring.

[Sridhar Bhaskarla]

Hi Mike,

You need to have OS mirrored. One advantage of having LVM mirroring on the external storage is that you will get good benefits from the reads. If you make sure that the mirror disks are accessed through a different controller, then you can minimize the performance degradation due to writes. Try having four A6795A HBAs. They work at 2 GB and the latency is much lower than earlier cards.

-Sri

[Jannik]

I have installed a vpar on SAN disks and it worked like af dream :-). There are the possibility that U have a failure on the SAN.
Again I have tried double striping on 4 Eva cabinets where U have a stripeset on all the 4 controlers on 2 of the cabinets and then stripe them from the host into 1 disk, this disk you then mirror to the other 2 EVA cabinet - I got VERY high IO. And it was fault tolerant... Server in one builing SAN angain in 2 others :-).
This U could use for root disk as well, but it is possible that you don't have the time. But is possible (I used SP3.0A (secure path)- there you could hardcode SAN path).

[Victor BERRIDGE]

Hi Mike,
My preferences goes to local disks for strictly OS plus backup soft (tools why not) on the internal disks and mirrored and the rest on dual path SAN or SCSI. For the internal part the best is with 2 controllers: Ive been caught once with mirror disk failure the type " Im here - now Im not " with intempestive scsi reset and box crash (then all is ok for few days...)...
This does make your live more easy when upgrade time comes. I do load the swap on external also though to take advantage of the bandwidth...(but when do these boxes really swap?...)
I did some configurations with all external,
at the time the performances were far better than the traditionnal but now with ultra-scsi internal I doubt its worth the worries if things fail (with microcode upgrade on SANs you are regularly using half bandwidth...)

All the best
Victor