This post is more than 5 years old
4 Posts
0
2783
May 14th, 2012 01:00
RAID configuration
What is the best RAID configuration for a VMWare VMFS datastore that will be holding backups for various SQL: servers. We have around 10 x 1 TB SATA drives. Would it be a RAID 10, or RAID 5. I guess backups would be more write intensive than read.
No Events found!
Storagesavvy
474 Posts
1
May 16th, 2012 18:00
Actually, VMFS IO size is not related to the VMFS volume block size. IO size from VMFS is the same as that issued by the Guest OS. The VMFS block size only relates to the minimum/maximum size of files in the VMFS and the VMFS itself.
Yes, 1MB is a large IO, in fact, by default CLARiiON bypasses write cache for IO’s 1MB or larger (this is configurable). There is actually little or no benefit from write cache when dealing with IO’s that large and generally the host and disk can outperform the cache so cache becomes a bottleneck.
As far as your other questions, if the IO size was 1MB, it would fill 4 stripes and would still be good sequential workload. For a 4+1 RG, the 1MB IO would write out about 1.2MB of data in total ((256KB stripe + 64KB parity) * 4) to the back end disks. The 1MB host IO would end up being 4 back end IO’s to disk but that’s not really an issue in this case. You could create an 81 RAID group which would give you a 512KB stripe in which case the 1MB IO would be 2 IOs. But in reality, 41 is a sweet spot for Clariion.
By comparison, a 1MB write to a RAID10 4+4 group would write out 2MB to the backend disks.
A 1MB write to a RAID6 4+2 group would write out about 1.4MB of data ((256KB + 128KB) * 4)
Richard J Anderson
Anirudh_Banerje
59 Posts
0
May 14th, 2012 02:00
Hi
Well, considering write penalty of RAID-5 Vs RAID-10...RAID-10 would be a better choice for handling database application IOs..
Please refer to the below performance and availability guide for more information about good choice of RAID types..
http://www.emc.com/collateral/hardware/white-papers/h5773-clariion-best-practices-performance-availability-wp.pdf
Thanks
Anirudh
rupal1
433 Posts
0
May 14th, 2012 04:00
RAID 10 vs RAID 5 configuration for high-load database, Vmware VMFS;
500GB x 4 disks
500GB x 3 disks
You can clearly see RAID 10 outperforms RAID 5 at fraction of cost in terms of read and write operations.
RAID 5 read is a little slower than RAID 10 because the data are not always in the same disks. But again, it depends on your application that disk/OS buffer might play an important role.
So for you Raid 1/0 is the go if the cost is not a priority on performance.
etaljic81
1K Posts
0
May 14th, 2012 05:00
It's a backup so stick with R5
Storagesavvy
474 Posts
1
May 14th, 2012 14:00
CLARiiON performs full write stripes in RAID5/RAID6 if the workload is sequential. So for example, if the RAID group is 4+1 RAID5 and you are writing backup dumps from SQL, the stripe will fill in cache and then dump to disk. The RAID5 write penalty doesn’t apply in this case.
For high-bandwidth, sequential read and/or write workloads, RAID5/6 will outperform RAID10 most of the time. If the workload is mixed read/write small-block, etc, then RAID10 will outperform usually.
So, if the VMFS you are creating is just for SQL backup dump files and the disks are 10K/15K FC/SAS then RAID5 is a good way to go. If the disks are large SATA/NLSAS (ie: 2TB) then RAID6 would provide better protection levels.
You will want to make sure your backup BLOCKSIZE is 64KB (65535 bytes) or a multiple of 64KB since CLARiiON uses 64KB element sizes. In the case of RAID5 4+1, the stripe size is 256KB so a single 256KB write IO fills the whole stripe and no RAID5 penalty is incurred. 1MB IO would fill 4 stripes at a time, 64KB IO would coalesce into a single 256KB stripe, assuming that the IO was sequential as backups normally are.
Richard J Anderson
imranvcppatel
4 Posts
0
May 15th, 2012 01:00
Hi Guys
Thanks for all the replies.
So to conclude, for a VMFS volume holding SQL backups using SATA disks, a RAID 6 configuration would give “overall “ good performance, and will not have an impact on the clariion, in respect to forced flushed cache, and high SP utilisation.
rupal1
433 Posts
1
May 16th, 2012 05:00
Yes you can conclude on Raid 6 .The only flaw with RAID 10 is that it uses a lot more disk space than RAID 6.
If the Question is answered please dont forget to mark the thread as answered.
Thanks
Rupal Rajwar
Storagesavvy
474 Posts
0
May 16th, 2012 08:00
RAID10 is awesome for many workloads, but in the case of sequential writes (like backups) where you can achieve full stripe writes for RAID5/6, RAID10 is actually inferior because it will ALWAYS write TWO IO's per host write IO. A full stripe write with RAID5/6 will outperform RAID10 assuming the same number of spindles. Again, this is specific to truly sequential workloads.
zhouzengchao
4 Operator
•
1.4K Posts
0
May 16th, 2012 18:00
(1) I have one questions. Due to the file system block size = 1MB, each I/O from VMFS would be at least 1MB, right? That's a relatively big size....
(2) For backend clariion, a (4+1)R5 RG provides 256KB stripe size, so in this case, would the VMFS 1MB I/O benifit from Morely Write (512KB on I/O operation on backend)??
(3) If so, does it mean only two I/O from cache or frontend port will come to the backend? Of course, these two I/O will incure more disk I/O then.
Please correct me if anything was wrong.
zhouzengchao
4 Operator
•
1.4K Posts
0
May 16th, 2012 19:00
that is make sense, thank you, Richard!