topic Re: SAN vs Local Disk in HPE EVA Storage

SAN vs Local Disk

George Selleck — Tue, 11 Dec 2001 17:01:12 GMT

We ran a series of read/write tests on an N class machine - HP-UX 11.0, TachLite FC HBA model A5158A, not fabric,Brocade switch
- and found that local disk (7200 rpm) r/w a 1 GIG file faster than the SAN.

Did we miss configure something ? Fiber card? Switch? Wait for Fabric?

Re: SAN vs Local Disk

Michael Tully — Tue, 11 Dec 2001 22:00:31 GMT

Hi,

What method of recording was done to gather
this information. If you are using sar -d
it will not give an accurate reading because
of the caching in the SAN.
One thing to remember is that fabric is better than loop.

-Michael

Re: SAN vs Local Disk

Edward Borst_2 — Tue, 11 Dec 2001 22:01:16 GMT

how does your SAN configuration look like??
how is the lun where you did the test configured??

Re: SAN vs Local Disk

Duncan Edmonstone — Wed, 12 Dec 2001 09:57:44 GMT

Did you wrie to raw disk, or to a file system?

Anything to a file system would most likely be skewed by the buffer cache...

HTH

Duncan

Re: SAN vs Local Disk

Deshpande Prashant — Wed, 12 Dec 2001 16:04:05 GMT

HI
With SAN you envolve more devices (switches/bridges).
The advantage of using SAN being more flexibility and growth. If you have more numbers of disks locally attached against going via SAN , SAN might show performance improvement against just single disk.

Thanks.
Prashant.

Re: SAN vs Local Disk

George Selleck — Wed, 12 Dec 2001 17:10:13 GMT

1) The SAN was setup with RAID5 (3 disks) - RAID0 was even slower

2) File system not raw, loop is used right now - fabric later

3) Tests done
write test
timex cat gig1 > /san/gig1
read test
timex cat /san/gig1 > /dev/null

Re: SAN vs Local Disk

Duncan Edmonstone — Wed, 12 Dec 2001 23:00:27 GMT

As I said previously, any test on a file system is fairly meaningless, as you are really just exercising the buffer cache after the first run... A better test would be to use dd to copy from an lvol on the SAN to /dev/null, and compare this with dd from a local lvol (same size) to /dev/null - don't do it the other way round if you have anything in your lvols - to make the test 'meaningful' use a blocksize the same as one of your filesystems (usually 1k for VxFS file sytstems)

Other points worth remembering:
- Are you comparing like with like? What are the disks attached via the SAN?
- These tests are only giving you results for sequential reads/writes - many applications do not have this kind of profile anyway.
- If you are on a FCAL loop, then remember you are sharing the 100MB/s bandwidth with other devices on the same loop.

HTH

Duncan