Re: Performance not what I expected using SSD (5754 Views)
Reply
Occasional Visitor
Cruice24
Posts: 1
Registered: ‎12-17-2013
Message 1 of 4 (5,814 Views)

Performance not what I expected using SSD

First time diving into SSD, here is my setup

DL380p Gen8

Embedded Smart Array P420i

and a Smart Array P420 in slot 2

 

P420i has:

 

  1. SAS Array A
  • 2 logical Drives (Mirrored 146gb SAS Drives)
  • 50GB / 86GB

2. SAS Array B (RAID 5 6 x 600GB SAS Drives)

  • 2 logical Drives
  • 1.4 TB Each

P420 has:

 

1. Solid State SATA Array A (RAID 5 3 x 600 GB SSD SATA drives)

  • 2 logical drives
  • 400GB / 714GB (4.1GB unused space)

2. SAS Array B (RAID 5 5 x 600 SAS Drives)

  • 2 logical Drives
  • each 700GB (1TB unused space)

Caching has been disabled for the 2 SSD logical drives

 

Backed up about 90GB worth of database data (backing up locally TO one of the SSD logical drives) took about 2 hours.  The restore of the same data took 3 hours 15 minutes, again SSD to SSD.  A second restore test of the same data to a NON-SSD logical drive ("spinning to spinning") took 2 hours 45 minutes and the backup took less time as well (1 hour 30 min).

 

A third test was moving all the data to the embedded controller, and the backup time was 1 hour 30 min, and the restore time was 15 minutes faster (1 hour 15 min).

 

Straight disk to disk copy (using robocopy) SSD to SSD we were getting about 18GB per minute.  Regardless if Cache was enabled or not...made very little difference

 

Straight spinning disk to spinning disk copy (non SSD) we were getting about 22.5 GB per minute.

 

So in both cases typical spinning disk were out perfroming the SSDs and I don't know why.  As I said, this is our first dive into SSD and I was suspecting a much better performance. 

 

What am I missing?

What am I doing wrong? 

Suggestions?

Would I get better performace using SAS SSD instead of SATA SSD?

Frequent Advisor
VBO
Posts: 33
Registered: ‎04-15-2012
Message 2 of 4 (5,789 Views)

Re: Performance not what I expected using SSD

Hello,

 

Need more informations like :

OS Version

Controller : Firmware version, cache size (FBWC), strip size, raid mode

SSD : brand, model and version

 

Best Regards

Vincent

Trusted Contributor
ernst limbrunner
Posts: 143
Registered: ‎03-18-2004
Message 3 of 4 (5,766 Views)

Re: Performance not what I expected using SSD

hi

 

first of all it is not a good idea to have a backup on the same physical array as the original data.

 

in your case all reads and writes during the backup go/came from the same disks.

this will always deteriorate performance.

 

raid 5 performance depends heavily on the controller. so you should have a write cache for

the backup logical drive. (parity generation and extra writes for raid 5)

 

for raid 0 a write/read cache for ssds does not make much difference.

 

you might also use the most current driver and firmawre and activate FAST PATH in

smar array administrator. this will circumvent the smart array firmware for raid 5 reads

(but not for raid 5 writes)

 

regards

ernsto - germany

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Regular Advisor
a_o
Posts: 69
Registered: ‎02-22-2012
Message 4 of 4 (5,754 Views)

Re: Performance not what I expected using SSD

[ Edited ]

SATA performace < SAS performance.
I've seen this in real world scenarios.

What's the performance rating of the SSD drives?
I've tested SAS SSD drives (Pliant LB806M) that can push 12 gigabytes writes and 18 gigabytes reads per minute on a p420.
I found that SATA SSD drives - especially consumer grade ones, don't come close to those numbers in the real world.

Also, realize that the nature of your data - i.e. random vs sequential, 4K vs 8K vs 32k vs 64K blocks will greatly impact your performance.

My take is that SSD drives will generally yield a 30%-50% improvement over HDD drives given a mix of I/O patterns.

Lastly, I would suggest trying to create a single drive SSD array, and re-run your tests. you might be surprised at what your may find.

The opinions expressed above are the personal opinions of the authors, not of HP. By using this site, you accept the Terms of Use and Rules of Participation.