DL380 G5 - Poor array performance - P400 controller

[Nick Salty]

Hi There,
I have a 380G5 with the P400 controller, firmware v5.26.

The background:
This server is solely used for backups, using backup exec 12.5. I have recently tried to start doing 'disk backups' from Exchange as a staging area before writing to tape. There are two types of backups one can do: whole database and brick level (known as GRT).
The difference is that 'whole database' backup creates 1 large .img file, while the 'brick level' creates ~750 files, and allows granular restore of mailbox items.

I created an additional RAID0 array using 2 146gb disks (in addition to the 4-disk RAID5 set for the os/apps).

The problem:
If I create a backup job that backs up the 'whole database', then the data is written to the RAID0-drive at ~1600mb/min - great.
However, if I create a backup job for the 'brick level', to the same drive, it only writes at ~150mb/min - bad!

Ok, so you'd think it was the software - so did I- but if I attach a cheap external USB disk, thereby bypassing the P400 controller, and point my 'brick level' backup at that, it writes at 800mb/min!

So at the moment, a cheap external USB disk gives more performance than a RAID0 set.

Does anyone have any suggestions?
I have already upgraded the firmware to the latest version I can find, and ensured the drive is either empty or degragmented when testing my backups.

Many Thanks!

[Matti_Kurkela]

Do you have the Battery-Backed Write Cache (BBWC) module installed in the SmartArray controller?

Writing into one large file is simple enough for good performance without a write cache, but your brick-level backup may require a more complex access pattern which would be seriously hurt from the lack of write caching.

Although the BBWC is technically optional, in practice it is mandatory for most tasks requiring good write performance.

MK

[Mark Balducci1]

I am having the same issue, been working on this server for 4 days now, and banging my head against a wall, 4 hard drives in a raid 5 is so much slower than a single drive, i have tried a p400 with 256mb ram, and en e20 with 128 battery backed up ram. same thing - no difference from the 2 cards, latest firmware and drivers running windows server 2008 x64 - when i boot i get a message saying raid 5/6 performance will be degraded due to a background rebuild, but i left the server on and running for days and it still gives me the same message, also tells me that its doing a background rebuild in the hp array config utility. what to do??? copying data from one drive to another is so slow! (10mbs) please help!!!!!!!

[TTr]

> but if I attach a cheap external USB disk, thereby bypassing the P400 controller, and point my 'brick level' backup at that, it writes at 800mb/min!

What if you try the "whole database" method on the external USB disk? If it gives you proportionaly higher than the 1600mb/min as the "brick level" did then the P400 is consistently slow. The BBWC is very critical for the P400 as Matti pointed out.

[Mark Balducci1]

ive tried with an E200 128mb BBWC and the raid 5 is still painfully slow - no different from the P400 - still copies at about 10mb a sec, but from 2 single drives (no raid) it copies at about 40mb a sec

[TTr]

@Mark Balducci1: This posting was for Nick Salty's issue, you can not highjack the thread for your own problems. Your problem is clearly different caused by array card backgraound activities. Did you kick-off any such activities? RAID migration, disk failures etc. They can take days to complete.
Please start your own new thread and post your problem with details.

[Nick Salty]

Thanks very much for the replies.
No, the server was not purchased with bbwc option.
I have just tried the full database backup to usb disk = 1,322mb/min.

I also tried reconfiguring the second array with a single disk instead of two in a RAID0, and got slightly slower performance than before = 148mb/min.

Even without bbwc I'd expect to get better performance than from a usb disk?
If I copy just (mixed size) files it seems to write at about 1,500mb/min.

During the slow backup type ~700 files are created, totaling ~18GB. During the fast, 'whole db' backup type, a single large file can be created (or 7 3GB files for the fat32/usb disk), which explains why it's faster in general, but better performance from usb still surprises me.

Would/could changing the stripe size help?

[Mark Balducci1]

im so sorry for the bad nettiquite, i honestly thought it was something to do with the raid drivers and would be related to nick problem, sorry nick for butting in.. but nick a word of advise.. are you runing symantec endpoint protection? i am, as well as symantec backup exec 12.5 and system restore 8.5, and the latest version of endpoint protection is causing nothing but greif on this server, slowing down the network and i supect file copies too..

[TTr]

Changing the stripe size allows for some fine tuning but it will not make the difference that you expect. The BBWC is very important just like with cache memory on large standalone array systems. The USB devices have gotten very fast in the recent years. As the memory chips get cheaper, faster chips are used in the drives and they run pretty much at the speed of the USB bus, the chips are no longer a slow down factor. They have become low end SSD disks.