msa1500 configuration problem

[Volodymyr Romaniuk]

Hello!
We experience some problems with your products MSA 1500cs + msa20. The structure of our system is following:

2 sets:
417453-421 HP DL380G5 5110 1G (1 piece)
264007-B21 Slimline DVD-ROM Drive (8X/24X) Option Kit (1 piece)
375861-B21 HP 72GB 3G SAS 10K SFF SP HDD (2 pieces)
351580-B21 128-MB BBWC (Battery-Backed Write Cache) Enabler (1 piece)
399771-B21 Redundant Power Supply 350/370/380 G5 Worldwide Kit (1 piece)
A8002A HP FC2142SR 4GB PCI-e HBA (2 pieces)
394795-B21 HP NC380T PCI Express Dual Port Multifunction Gigabit Server Adapter (1 piece)
452141-B21 Integrated Lights-Out Advanced Pack, No Media ,1 Server License (1 piece)

2 sets Storage:
AA986A MSA1500 controller shelf (1 piece)
218231-B22 HP MSA1000/1500 Controller 256 Cache All (1 piece)
AA987A MSA 1500 Fibre Channel I/O Module (1 piece)
254786-B21 256 MB Cache Upgrade (2 pieces)
335921-B21 HP StorageWorks MSA20 Storage Enclosure (1 piece)
432341-B21 HP 750GB 7.2k HP SATA 1yr Service (12 pieces)
221692-B22 5m SW LC/LC FC Cable ALL (8 pieces)
A7984A HP StorageWorks 4/8 Base SAN Switch (1 piece)
A7446B HP 4GB SW Single Pack SFP Transceiver (12 pieces)

The system uses raid 5 with 11 HDD. The problem is when we read a big file the speed is about 90MB/s, when we write the same file speed is about 25 MB/s. When we read and write files at the same time the indices of speed decrease immediately down to 10-20 MB/s per 1 user! Which, I guess, is abnormal. In my opinion, the problem is in settings of msa1500. The scheme of our system is attached. If you have any ideas about the determination of our problem, please reply to us as soon as possible. thanks a lot!

[James Muell]

your attached zip file was empty. Looking at what you wrote. Don't expect alot of performance from the MSA platform. It looks like you only have 1 MSA controller so ALL I/O is running through 1 controller, which tops out at 160mb/p/sec, so if you have multiple hosts connected and trying to push I/o it will be slow.
-- What OS?
-- Do you have the latest driver and firmware on the hbas?

[Pieter 't Hart]

if you want writing speed do not use raid5!

yes more disks are working for you, but one write operation from the OS results in 11 writes to + 11 reads from disk by the controller.
this will be 10 data writes + one calculated "parity" write + 11 reads, the parity from the read blocks is calculated again and compared to the read parity.
so to confirm the data on disk is ok.

this can be improved by configuring the controller to use more cache for write operations. or better adding more write cache.

[Uwe Zessin]

Ahem...
Any half-intelligent RAID-5 implementation (and I don't think the MSA is t-h-a-t bad) will do it like this for small I/Os:

- 2 reads (old_data + old_parity)
- calculating in-memory:
-- remove the old_data data from the old_parity giving temporary_parity
-- add new_data to temporary_parity giving new_parity
- 2 writes (new_data + new_parity)

[Pieter 't Hart]

Uwe thanks for the addition.

Volodymyr states he has a raid-5 set of 11 disks.
As the host-io is only finished when all the separate disks-operations are finished,
I think you must take into account these 11 oerations at the disk level (writing one big file + shared use).
Yes, as these operations will be paralellized it will not take the same time as 11 separate disk-io's, but it will take more than just two.
Also the SATA disks used here will behave different from FC or SCSI disks.

[Volodymyr Romaniuk]

Thank you for help!
Our system is built on a base of Windows 2003 Server. Also we have the
last edition of Firmware and drivers. We use MSA1500 which is connected to one MSA20 with which one user works only.

[Volodymyr Romaniuk]

Iâ ve made some tests with IOMeter and Iâ ve got next results:

Test 1:
MSA1500cs+MSA20 connect to DL380
Cashe 512MD DDR, Read/Write 50/50
Raid 5 stripe size 64K (12hdd 750GB)
512kb, 100%Read , 10 Outstandins IO, 100 Sequential- 120 MB/s
512kb, 100%Write, 10 Outstandins IO, 100 Sequential- 70 MB/s
512kb, 50%Write 50%Read, 10 Outstandins IO, 100 Sequential- 35/35 MB/s

Test 2:
I took HP Smart Array 642 controller and connected MSA 20 directly to this controller, without MSA 1500cs. But I donâ t have a battery for this controller. So Iâ ve set 64MB cashe to 100% Read.
Raid 5 stripe size 64K (12hdd 750GB)
512kb, 100%Read , 10 Outstandins IO, 100 Sequential- 191 MB/s
512kb, 100%Write, 10 Outstandins IO, 100 Sequential- 97 MB/s
512kb, 50%Write 50%Read, 10 Outstandins IO, 100 Sequential- 47/47 MB/s

When I used cheap controller I had better performance than MSA 1500csâ ¦. Over 60% for Read, and 30% for write (without cache!!!)
Suppose thereâ s something wrong with my MSA1500cs, or maybe my MSA1500cs has a bad productivity with MSA20 or with sequential stream.

[Uwe Zessin]

Unfortunately, the MSA1500cs and the MSA20 do not play well together.

You should also note that the MSA20 operates differently whether it is connected to a PCI backplane RAID-controller or the MSA1500cs.

In the first case, the MSA20-embedded SmartArray 6 controller does the RAID function.

In the second case, each SATA disk drive in the MSA20 is encapsulated into a JBOD (although the SmartArray terminology calls it a RAID-0) and the SmartArray-5-based MSA1000 controller module in the MSA1500cs does the RAID function.

[Oleg Fedoryshyn · Accepted solution]

As I understand from this forum if I want have good performance with sequential thread i must buy p600/800 controller and msa20? But what for msa1500cs? For people that want spend a lot of money for build array with pure performance...
I find some link about msa1500cs and msa20,30 performance
http://www.itpro.co.uk/storage/reviews/97186/hp-msa1500-san-scsi-starter-kit/page2.html
So maybe someone from hp support can say something about this problem? That is HP forum and I think HP specialist has got some test with msa1500cs + msa20 with 750gb HDD.

[Pieter 't Hart]

maybe this links to the problem :
CLI-2> show other_controller
Controller 1 (right controller):
MSA1500(c) Hewlett-Packard PB9840TX3UR05E

Host Port_1:
REPORTED PORT_ID 500508B3-0094BF51
PORT_1_TOPOLOGY = Link Down

in combination with :
CLI-2> show auto_path_change

Implicit (automatic load-based) LUN ownership changes are currently ENABLED.

when load is generated, the host tries a path-change for some lun wich doesn't succeed?

CLI-2> show eventlog
TIME CLASS SUB DETL MESSAGE
0d 00:06:12.2 0006 0000 0000 Redundant operation state change.
0d 00:06:30.1 0006 0000 0000 Redundant operation state change.
0d 00:06:36.1 0006 0000 0000 Redundant operation state change.
0d 00:07:03.6 0006 0000 0000 Redundant operation state change.
0d 00:07:48.1 0006 0000 0000 Redundant operation state change.
0d 00:09:41.7 0006 0000 0000 Redundant operation state change.
0d 00:10:17.7 0006 0000 0000 Redundant operation state change.
0d 00:10:23.7 0006 0000 0000 Redundant operation state change.
0d 00:24:05.7 0006 0000 0000 Redundant operation state change.
0d 11:53:53.4 0006 0000 0000 Redundant operation state change.
0d 11:54:06.1 0006 0000 0000 Redundant operation state change.
0d 11:54:12.1 0006 0000 0000 Redundant operation state change.
0d 11:54:18.1 0006 0000 0000 Redundant operation state change.
0d 11:54:24.1 0006 0000 0000 Redundant operation state change.
0d 15:07:28.3 0006 0000 0000 Redundant operation state change.
0d 15:07:40.9 0006 0000 0000 Redundant operation state change.
0d 15:07:46.8 0006 0000 0000 Redundant operation state change.
0d 15:07:52.8 0006 0000 0000 Redundant operation state change.
0d 15:07:58.8 0006 0000 0000 Redundant operation state change.