Data Storage - MSA 1500 Performance Question

Author

MSA 1500 Performance Question

Apeulus Rex

2006-10-14, 1:13 pm

Hi everyone,

I recently migrated to an MSA1500 based SAN. I have the following config:

MSA1500 with redundant controller, but did not purchase the extra 128MB
cache upgrade (DOH)
2x MSA30's loaded with 72GB 15k U320 drives (all LUNS are 0+1)
1 MSA20 with 7.2k 160GB SATA drives (1 LUN and it is raid5)
1 4 port Brocade 2GB switch
1 Server running 2003 Enterprise (For now)

Now, everything works, but the performance seems really subpar. When I do a
backup from the MSA30 to the MSA20, I get like 7 MB/s. That's truly awful.
I'd be better off backing up to USB at that rate. So I ran some benchmarks
with Iometer, and it claims to be able to do like 100MB/s from both the
MSA30 and MSA20. However, when I do drag n' drop benchmarks in Explorer, I
get the following results:

Cache set to 20% Read / 80% Write
----------------------------------------------------
MSA30 to Local C: 9.8 MB/s
Local C: to MSA30 33 MB/s
MSA20 to Local C: 19.25 MB/s
Local C: to MSA20 12.25 MB/s

MSA30 to MSA20 8.16 MB/s
MSA20 to MSA30 18.58 MB/s

Cache set to 80% Read / 20% Write
---------------------------------------------------
MSA30 to Local C: 29.94 MB/s
Local C: to MSA30 19.96 MB/s
MSA20 to Local C: 59.8 MB/s
Local C: to MSA20 8.29 MB/s

MSA30 to MSA20 6.73 MB/s
MSA20 to MSA30 19.96 MB/s

So it looks like the cache REALLY matters. I am going to put in the cache
upgrade based on this. However, the fact that Iometer reports such
different results than doing drag n' drops, that makes me wonder if there is
something in Windows that I could tweak?

Any thoughts would be appreciated.

charles.morrall@benzler.se

2006-10-16, 1:14 pm

Apeulus Rex skrev:

> Hi everyone,
>
> I recently migrated to an MSA1500 based SAN. I have the following config:
>
> MSA1500 with redundant controller, but did not purchase the extra 128MB
> cache upgrade (DOH)
> 2x MSA30's loaded with 72GB 15k U320 drives (all LUNS are 0+1)
> 1 MSA20 with 7.2k 160GB SATA drives (1 LUN and it is raid5)
> 1 4 port Brocade 2GB switch
> 1 Server running 2003 Enterprise (For now)
>
> Now, everything works, but the performance seems really subpar. When I do a
> backup from the MSA30 to the MSA20, I get like 7 MB/s. That's truly awful.
> I'd be better off backing up to USB at that rate. So I ran some benchmarks
> with Iometer, and it claims to be able to do like 100MB/s from both the
[snip]

What load parameters are you using in IOmeter?

I can probably achieve about 100 MB/s sustained read on a Windows box
connected to pretty much any raid system with a few drives, if I
configured the load for huge sequential reads. However, that doesn't
mean I can get 100 MB/s from that same Windows box and raid system for
any arbitrary application.

Sean Howard

2006-10-16, 1:14 pm

Hi Charles,

That's true, I am not doing any random in the test. Wouldn't backing up the
filesystem be sequential though? That's the thing I'm really concerned
about.

Sean

<charles.morrall@benzler.se> wrote in message
news:1161003457.942094.116510@e3g2000cwe.googlegroups.com...
>
> Apeulus Rex skrev:
>
> [snip]
>
>
> What load parameters are you using in IOmeter?
>
> I can probably achieve about 100 MB/s sustained read on a Windows box
> connected to pretty much any raid system with a few drives, if I
> configured the load for huge sequential reads. However, that doesn't
> mean I can get 100 MB/s from that same Windows box and raid system for
> any arbitrary application.
>

charles.morrall@benzler.se

2006-10-16, 1:14 pm

Sean Howard skrev:

> Hi Charles,
>
> That's true, I am not doing any random in the test. Wouldn't backing up the
> filesystem be sequential though? That's the thing I'm really concerned
> about.
>
Yes, but on an average windows box with user files, the files are
fairly small and numerous. I usually estimate I can get 20-25 MB/s at
best from a filesystem with small (10-100 kB) files. For profiles
directories this drops substantionally, and for database dumps
(hundreds of MB) this figure can double easily.

It might be sequential read on a high level doing backup, but still
every file has to be opened, read and closed. Not as sequential as you
might want.

The advantage of using a disk as a backup device as I see it is that I
can pull multiple file systems from multiple clients in parallel,
without worrying about keeping a physical drive streaming as the file
systems vary in performance quite a lot.

If you only have one filesystem to backup, the bottleneck might not be
in the target disk, instead in the file system itself. You might want
to check for fragmentation and see if that has any effect.