MSA1000, Slow Transfer Speeds ?

[NigelUK]

After finally setting up our MSA system this weekend we are noticing the write speeds on the server to be fairly slow, average of approx 31MBps, should transfers be this slow, or would there be a bottleneck somewhere?

Any help would be appreciated

[CA1064576]

Can you tell us more about your configuration? (hom many disks, RAID level, etc.) to revierw your configuration and make suggestions.

[NigelUK]

Sure,

Server - Dell PE 2850
3.2ghz P4
2Gb Ram
Onboard Perc 4e/di Raid Controller, connected to that are 6 U320 scsi drives in a raid 5 config
MSA 1000 Small business San Kit in a Raid 5 config, with 11 U320 Drives All in 1 virtual disk (1 disk hot spare)
Operating system is RHEL 3, Kernel version 2.4.21-4.ELsmp

Dmesg info MSA

scsi1 : QLogic QLA6312 PCI to Fibre Channel Host Adapter: bus 9 device 4 irq 82
Firmware version: 3.03.01, Driver version 7.01.01-fo

Starting timer : 0 0
blk: queue f7245c18, I/O limit 4294967295Mb (mask 0xffffffffffffffff)
scsi: unknown type 12
Vendor: COMPAQ Model: MSA1000 Rev: 4.32
Type: Unknown ANSI SCSI revision: 04
Starting timer : 0 0
blk: queue f7245a18, I/O limit 4294967295Mb (mask 0xffffffffffffffff)
Vendor: COMPAQ Model: MSA1000 VOLUME Rev: 4.32
Type: Direct-Access ANSI SCSI revision: 04

[CA1334709]

I am afraid I don't have any insight, but I am experiencing the same thing. I sure do hope I have something configured wrong after spending $50k on this dog.

I have 3 new DL380 G4 Performance Servers and got the MSA1000 Starter Kit with the Redudancy kit. I have 3 volumes with from 3 to 7 146k drives in RAID 5. With DAS I see speeds 3 times the SAN. Last night I started copying 400Gb to a new volume and after 8 hours it was only thru 138Gb. My home PC does better.

Something is up. (and it ain't the performance):-)

Please post your solution here if you find one, I will do the same.

Tim

[CA1074530]

Hi,
we also noticed some perfomance degradation after a while. Although we do have some Notes Servers hanging of our's with 700 mailboxes and a few very large application databases.
A couple of items we were asked to check to improve performance were:

1. Spread disks over multiple trays, the Controller shelf has two busses, + the other two shelves. This does however, limit the number of disks you may want to have in a raid 5 group if you wish tray protection.

2. Ensure rebuild and expansion are set to LOW priority. anything else will kill performance.

3. Check your cache settings are appropriate for your data. ie. check the bias for R/W allocation.

I would be interested to know if these helped, or if you have already done these. We have hit a wall with ours, although we may be able to add more cache. (don't expect that to jelp with random i/o though)
I also presume your fabric is all 2GB and there are no HUBS or 1GB ISL's involved?

What type of Zones have you created? (if at all)

Dean.

[CA1334709]

Dean,

We have a Notes server as well with about the same number of users. We're workin' with HP today, they had up updated drivers and install SP2 for SecurePath 4.0c. Still about 30MBs (about what I get from my laptop ),

In response to your questions,

1. Since I have RAID 5 I did not spread them across controllers. I normally would with RAID 0 or 0+1.

2. Yup...Low.

3. 50/50

Your assumptions are correct. We have two 2/8 switches int he back of the MSA1000 and 2 FCA2214s in each server with one leg in each switch.

[CA1074530]

Tim,
I would be interested if HP have a solution for you.
We did notice some improvement by spreading the load. We have districuted our disks so that a RAID5 array has one disk on both channels in the controller shelf, and one disk in each of the other two shelves. If you need more disks, it is relatively safe to put additional disks in the controller shelf as a failure there will most likely render the entire unit inoperable, and less likely to destrory the array. Otherwise, you may need to create more Arrays.
For Notes servers, this is also a good option. I noticed on ours, Disk Queue length was quite high, by distrubuting databases over arrays, we can obviously reduce the disk queue length. You can keep a simple directory structure under the Notes server bu utilising file links, so that the users do not see where the files are physically located compared to the Notes Data directory.

[CA772171]

Unfortunately, the MSA1500 is just as slow. We use ours for backup and we get slightly less than 20 Gb/hr (slower than the equivalent LTO). That translates to 5-6 MB/sec (40-50 Mb/sec).

For your disk rebuild, you will need to weigh in pros and cons. For instance, when a disk failed in our MSA if the priority is set to low the new disk somehow never gets completely rebuilt (even after 3 weeks) as the MSA almost always runs backup. We therefore set ours to 'medium'.