Unsolved
This post is more than 5 years old
8 Posts
0
10881
July 18th, 2012 11:00
MD3200 SAS Performance/Configuration Question
Hey, I'm trying to determine how different configuration will affect performance. In the scenario below, will option 1 provide better performance? Why or why not? Thanks for the help.
1. Create one Disk Group: 18x2TB RAID6 and create two Virtual Disks and assign each virtual disk to a separate controller.
2. Create one Disk Group: 18x2TB RAID6 and create one Virtual Disk
No Events found!
Dev Mgr
6 Operator
•
9.3K Posts
0
July 18th, 2012 12:00
Performance always depends on your usage.
What do you know about:
- read/write ratio?
- randomness of your IO?
- nature of the IO? (e.g. file server, virtual machines, Exchange server for 250 users, SQL for 100 users, etc)
- disk space requirements?
- IOPS requirements?
itpro44
8 Posts
0
July 18th, 2012 12:00
Thanks for the quick reply.
Specifically, this will be a file server.
70/30 read/write environment.
Randomness - Unsure (95% jpg files 5% word docs, pdfs, video, etc.)
Disk Space - 25TB
IOPS - best we can get from 18 x 2TB disks
Maybe another question would be, What impact does splitting a Disk Group into multiple virtual disks (as apposed to 1 virtual disk) have? Does is allow for more reads or writes to be queued up?
Dev Mgr
6 Operator
•
9.3K Posts
0
July 19th, 2012 09:00
With a 70/30 read/write ratio, you probably should consider raid 10, and definitely stay away from raid 6. Raid 50 may have been an option, but the MD3200 series does not offer this.
The more drives you put in a parity based raid set (e.g. raid 5 or 6), the slower the writes become. Reads do speed up with more drives though. This is why most raid controller manufacturers recommend about 7 or 8 drives at most in a parity raid set as this is a decent balance between the 2.
At about 30% writes or more (so 70% reads or less), the slowdown from the writes starts to impact overall performance negatively. Raid 6 is probably even more impacted.
Also; when using an 18 drive raid 5 or raid 6, upon a drive failure, expect a fairly substantial performance hit that will last several days (for the hotspare to rebuild).
2TB drives means you have the MD3200 and at least 1 added MD1200 (with 6 drives). I'd recommend to fill the second enclosure up with all 2TB drives. Then I would make it a 24-drive raid 10, yielding 12 x 2TB x ~93% (difference between decimal and binary "gigabyte"), which is about 22.3TB. If you really need the full 25TB, add another MD1200 with at least 4 more drives on top of filling up the existing MD1200 enclosure. Raid 10 doesn't have parity to calculate, so it doesn't take a performance impact on writes and the performance hit on a drive failure is also substantially less (possibly barely noticable).
itpro44
8 Posts
0
July 19th, 2012 10:00
Hey Dev, I totally follow you. If cost were no object, then I would do just that. Unfortunately, I'm not in that scenario. I've asked the question above because that is the scenario I am in. I'd like to determine how the MD3200 handles those two options and which one would deliver better performance and understand some of the reasons why. Additionally, on second thought, our environment is more like 90/10. Reads are definitely more important to us, as we are never in a rush to get large amount of data onto the server. We do however, want to pull it off fast. Most reads come from users browsing large sets of .jpg files. 2-7MB each, or pulling a large set of .jpgs, pdf, word docs, etc down to their system (100GB data sets). Can you comment on the differences of the options above?