Unsolved
This post is more than 5 years old
2 Posts
0
894
Raid Group to number of Lunes
Is it better to create small raid groups with one lune per group or large raid groups with multiple lunes per group? For example I have a cx3-40 that I have configured two raid 5 groups. One raid group with 5 disk the other raid group with 4 disk. Would I have gotten better performance for both lunes by creating one large raid 5 group with 9 disk and creating to lunes from that? Or if I am mirroring, is it better to create two raid groups with two disk each with one lune per group or combine them into one raid group of 4 with to lunes across the 4? I would think give 4 disk in a raid 1 carved into two lunes would be better?
I ask this because I am configuring a sql 2005 server with vmware and have 10 300gig FC disk available. I need to create some lunes for an esx box as well as lunes for the database itself. I was thinking of taking 4 and configuring in a raid 10, and then carving out different lunes for the different databse, log, and tempdb because this would give me 600gig of available storage to carve up. then taking the other 6 and configuring in a 1.5tb raid 5, and creating lunes for the database as well as an 800gb lune for vmfs partitions for vmware.
pwhyton
12 Posts
1
September 9th, 2010 00:00
it comes down to the IO profile of the application, with the limited number of spindles available i would have probably done the same as you to keep my SQL seperate. If the expectation is that the SQL is going to generate high IO then you have the option to create the 4 drive set as Raid 10 but if it is just providing database functionality for a couple of small apps (vcentre/update manager) then Raid 5 will probably do you.
the dilemma you have is one we face daily, with a limited number of disks how do we get the best seperation? you have 9 drives so 8+1 gives best capacity but then all applications/VMs share the same IO across spindles and you will probably end up with a couple of big LUNs or loads of small ones :-( splitting into 2 Raid Groups is the most logical but will reduce the total available capacity as you now have a greater Raid overhead.
my personel preference would be as you have done and then 2 or 4 LUNs maximum per Raid Group, but again it comes down to the IO profile of what you are putting on there.
Paul
Chris_au
17 Posts
1
September 9th, 2010 06:00
Theres a couple of parts to this one:
How much performance does your database and virtual SQL2005 server each need? If youre only small then just make one big RAID5 and be done with it for the sake of flexiblity.
What are their I/O profiles? Do they write lots as in a transactional online database or read more as in a data warehouse for reporting.
More spindles in a Raid Group will give more performance but having more applications using the same disks causes contention. If it all plays well together then dont worry too much, same goes if you arent needing high leels of performance. Depending on your I/O profiles it might not be as efficient to use RAID5 for a database but look at the additional cost overhead of running RAID 1+0.
In your case a RAID5 of 10 15krpm disks will give you atleast 980 IOPS and 2.4T of space. The RAID10 should do 1300 IOPS but gives only 1.3T of space. Remember that a RAID10 is not very reliable so you should be backing it up frequently.
I'd lean towards one large RAID5 with 10 disks and give the SQL server a RDM for its data and lots of memory (8gb+) and prioritise its disk access in ESX aswell.
jps00
392 Posts
0
September 9th, 2010 08:00
I recommend you read EMC CLARiiON Storage System Fundamentals for Performance and Availability, (Fundies) in particular the 'Logical Storage Objects' section. In addition, the EMC CLARiiON Best Practices for Performance and Availability, FLARE Revision 29 (BP) may also be helpful. The 'LUN Provisioning' section may be particularly helpful. Both of these documents are available on Powerlink.
Having written that, the advice above is sound. However, as mentioned in Fundies and BP, a single large RAID group poses potential availablity problems. All your eggs are in one basket, should you have a performance degrading rebuild. There are also several performance and availability advantages to configuring symmetrically when sharing RAID groups across applications. Two (4+1) RAID groups, if they meet your minimum performance and capacity requirements would be preferrable. This is likewise discussed in Fundies and BP.
earl27
2 Posts
0
September 9th, 2010 09:00
What counters would I need to look at on the existing SQL server to determine what my current I/O trends are? Also how do I determine the transaction sizes?
I currently have this sql server on vmware but it is small approx 50 users and a 10gig db. We are bringing on another agencies in the next 3 months, so I am moving this database from 32bit sql to 64bit and provisioning more memory and CPU, but want to make sure my disk subsystem does not become the bottle neck. We anticipate the user count to rise to 700 users in the next 4 months.