ProLiant Servers (ML,DL,SL)
1753854 Members
7740 Online
108808 Solutions
New Discussion юеВ

Re: 5 disks in Radi 5 slower then 1 disks?

 
Jerome_15
Occasional Advisor

5 disks in Radi 5 slower then 1 disks?

Hi,

after I have added a disk in my raid 5, I have executed some tests.
Specially with the SQLIO tool.

The result are:
Raid 5 / 5 * 36Gb disks=
Read: 320 io/sec / 2.5MB/s
Write: 120 io/sec / 1MB/s

Non raid disk 72 Gb=
Read: 520 io/sec / 4.1MB/s
Write: 150 io/sec / 1.2MB/s

My tests are done with 8KB IOs access.

so.... why 5 disks in Raid 5 are slower then 1 disk???

how I have to setup my strip size for the raid 5 and the partition format size?
this partition will be dedicated to file sharing.

thanks.

Jerome.
3 REPLIES 3
Terry Hutchings
Honored Contributor

Re: 5 disks in Radi 5 slower then 1 disks?

What is the stripe size? How was the array originally configured (through ACU or ORCA)? Are the drives the same except for the size (all U3 or U320, 10k rpm, etc.)?
The truth is out there, but I forgot the URL..
Jerome_15
Occasional Advisor

Re: 5 disks in Radi 5 slower then 1 disks?

I have tested variious strip size (16, 32, 64)
my NTFS format has been tested with 16,32& 64Kb.
I'm using the ACU web utility to setup my Raid.
My controller is the smart array 641 (without the write cache memory, only 64Mb read cache)

the drives in my raid 5:
4 drives are identitycal (same model) and 1 has a different model. All disks support the U320.
my first 4 disks use the rom version HPB6 while the latest drive use the HPB7
all are 10Krpm

My 72Gb drive has the HPB1 code and is also an U320.

the same benchmark on another server with 4 disks in Raid 5, but 128Mb of cache (50% read 50% write)
results:
Read: 9000 IO/s / 70MB/s
Write: 5700 IO/s / 44MB/s

so there is a huge difference!!!
Ryan Goff
Valued Contributor

Re: 5 disks in Radi 5 slower then 1 disks?

Please see this white paper:

ftp://ftp.compaq.com/pub/products/servers/proliantstorage/arraycontrollers/battery-backed.pdf

The best analogy I can give you is that no write cache is like relying on a page/swap file instead of using real memory.

Having a BBWC on your 641 controller to enable write cache will allow the OS to transfer data much quicker. Instead of the OS having to wait for data to be written and read from the drives it can keep a continuous flow of data to the controller which speeds things up drastically.

There are other threads in this forum with performance in the title you may also want to refer to as other people have asked this same question as you and we have provided answers.

The part numbers for 641 for cache modules are:
128-MB BBWC (Battery-Backed Write Cache) Enabler 351580-B21
64-MB BBWC (Battery-Backed Write Cache) Enabler 291969-B21

Also as you mentioned you compared your data transfer results to a server with a controller with 128mb cache and that is why you see such a speed increase.

Also, you might want to realize that Raid 1 and Raid 5 will always be slower than Raid 0 because the controller has to caculate the parity information and stripe the data across the two drives. With raid 0 there are no calculations so the speed will be much faster with the disadvantage of no redundancy.