Re: some EVMS questions

"Eric A. Hall" <[email protected]>
Newsgroups gmane.linux.evms.devel
Message-ID <[email protected]>
On 3/17/2007 6:36 PM, Scott Kaelin wrote:
> On 3/17/07, Eric A. Hall <[email protected]> wrote:
>> Hi guys,
>>
>> I'm putting together a storage server for NAS and iSCSI purposes and have
>> some questions about EVMS please. I've got a box of Raptor 150 ADFD drives
>> and a 3Ware 9650SE board which will provide the basic back-end storage.
>> Now I'm looking at options for the partitioning to support the server
>> opsys, the NAS storage pool, and the multiple iSCSI volumes. It seems that
>> the easiest thing to do would be to just make a single massive RAID-6
>> volume out of the drives, expose that to EVMS, and then chop it into
>> segments for logical drives. Assuming that's the case:
> 
> If you're planning on using iscsi for high I/O applications putting
> all the drives in a single raid6 is a bad idea. Try not to have more
> than two active LUNs on a single array. If you use one large RAID6
> array and then export a bunch of LUNs you'll notice horrible I/O when
> multiple initiators begin to write/read from the disk, since each
> initiator is reading from significantly different areas of the disk.
> If you are exporting only one LUN and using the rest for a fileserver,
> NFS/Samba, then disregard what I just said.

That's interesting. I guess it depends on what is considered "high I/O".
I'm just planning to boot VMware guest opsys and application files (and
maybe some other systems too but not much), so I don't think this is a
factor here, but it's still an interesting point. Thanks.

>> I understand that EVMS will let me dynamically resize segments, but how
>> transparent is this? I assume it will work for the storage and system root
>> XFS partitions, and I also assume that it will not work for the iSCSI
>> targets (since the partition and filesystem information for those will be
>> managed by the requester, not the server).
> 
> EVMS will let you resize the segments of LUNS as well. The trick is to
> export LVM/LVM2 regions rather than full fledged partitions. I use a
> windows initiator and when I run out of space on my exported LUN, I
> stop the target software, go in EVMS and expand that region, bring the
> target software back up and use the dynamic disk feature of windows.

Nice. So assuming that my iSCSI disks are going to be 20-80 GB (depending
on guest OS and application needs at the time), with an estimation of
around 40 GB average use... How should I structure the disk and LVM layout
for this to avoid problems later, after there has been a lot of resizing?
This seems to be where the idea of having oodles of small segments comes
into play, but I would prefer to avoid that.

> If you look around on the iscsi-target mail archive there is alot of
> information: http://www.nabble.com/iSCSI-Enterprise-Target-f4401.html
> about similar issues along with possible performance.

Thanks. I looked at some of the discussion there and elsewhere, and it
seemed that the best performance comes from having a very large RAID with
the largest number of active spindles. That may be due to the way that
people are benchmarking (single-user, effectively), but I will keep this
and the other points in mind.



-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Evms-devel mailing list
[email protected]
To subscribe/unsubscribe, please visit:
https://lists.sourceforge.net/lists/listinfo/evms-devel
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.