This post is more than 5 years old
60 Posts
1
1598
hard disk selection in clariion
Hi,
planing to buy a new clariion cx4 .we are uing oracle10g database,like to have good planing for the disks to get the maximum performance of the box.
Have few doubts, can we go for low capacity disks for the first 5 disks( flare disks) and the rest high capacity disks?
We don't want to use the flare disks.
is it a best practice to go with more disks of less capacity?
Thanks
Raju
RRR
1 Rookie
1 Rookie
•
5.7K Posts
1
April 27th, 2010 03:00
The 73GB disks are supported by CX4, but not as the so called Vault disks, so the lowest capacity drives you can order are the 146GB drives. This is exactly what we do for each CX4 we install.
For best performance on your application please work with your application experts. They can (should) usually tell you what they need for transaction log performance and database performance.
RAID10 is usually very good, but also very expensive as all data is available twice. But on the other hand on RAID10 random reads are very fast, since data can come from 1 of 2 sources. For random writes RAID10 is also very good as your write penalty is only 2, where on RAID5 this penalty is 4, so you need more disks to be able to handle the IOps.
10k drives deliver about 130 IOps.
15k drives can deliver about 180 IOps.
These are rule of thumbs are they are usually very pessimistic. Depending on the IO size IOps can be higher or lower, but the values I named are mostly the ones we use for calculations.
For example:
If you know your application needs 1000 IOps or so on the log disks (100% random writes) and 2400 IOps or so on the database disks (50% writes and 50% reads), the math would be:
1000 random write IOps for the logs would need 1000/180 x 2 (write penalty)= 6 disks x 2 = 12 disks on RAID10 using 15k drives
using 10k drives this is: 1000/150 x 2 = 14 drives in RAID10
If you would use RAID 5 for this:
using 15k drives this is: 1000/180 x 4 = 24 drives
using 10k drives this is: 1000/150 x 4 = 28 drives
These numbers are the minimum number of req'd drives. If you use RAID10 the number you'll eventually need to use is an even number. If you use RAID5 you could think of 6 Raid Groups with 4 drives each (3+1) in the 24 drive configuration or 7 Raid Groups with 4 drives each (3+1) in the 28 drive configuration.
Finally you need to take the needed amount of space into account. The higher number of the 2 is the one you need to go with, but for the logs usually simply take the smallest available drives since space isn't an issue.
For the database the math is somewhat similar:
2400 IOps where 50% is writes ans 50% is reads
The total amount of IOps needed on RAID10 = (2 x (2400 x 1/2)) + (2400x 1/2) = 3600
The total amount of IOps needed on RAID5 = (4 x (2400 x 1/2)) + (2400x 1/2) = 6000
If you would use RAID10 for this:
using 15k drives this is: 3600/180 = 20 drives
using 10k drives this is: 3600/150 = 24 drives
If you would use RAID 5 for this:
using 15k drives this is: 6000/180 = 34 drives
using 10k drives this is: 6000/150 = 40 drives
These numbers are the minimum number of req'd drives. If you use RAID10 the number you'll eventually need to use is an even number. If you use RAID5 you could think of 7 Raid Groups with 5 drives each (4+1) in the 34 drive configuration. Another thing you need to consider whether or not you want to share all drives for all LUN's you need. To do this you create a number of equally sized Raid Group, let's say 5 for example. Suppose you need 6 LUN's for your database. Now you create 6 striped META LUN's, each of the size you need where each META is devided in 5 equally sized LUN's each of which residing on a RG. This way each META uses all disks.
Does this make any sence ?
Raju_auh
60 Posts
0
April 27th, 2010 01:00
Thanks RRR,
We like to size on performance.it's an oracle OLTP data base.what's the smallest capacity drives available to use for flare disks in CX4 120/240 boxes?
previously we used have 78GB.What are the main aspects has to be taken care to get the best performance?
RRR
1 Rookie
1 Rookie
•
5.7K Posts
1
April 27th, 2010 01:00
Sure, order the smallest possible drives for the first 5 drives and the rest that actually suits your needs. Do you size on Gigabytes or performance (IOps) ? Sizing on IOps mostly asks for smaller drives, but that's all up to you to decide. We can help you with your decision though.
kelleg
4.5K Posts
0
April 27th, 2010 08:00
Everyone should copy this post for future reference - very good and detailed explanation.
glen
Raju_auh
60 Posts
0
April 28th, 2010 02:00
Thanks RRR,
could you pls tell me how to find out the IOps for the log files and the IOps for the database?so mean to say, we should create separate raid levels according to the IOps.
SKT2
1.3K Posts
0
April 28th, 2010 04:00
you need to work with your application/DB expert to identify the performance/IOPS and IO(RD/WR) details. Once you have those use RRR's formulas to determine the RAID level and the number of disks to be balanced for the LUNs you create.
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
April 28th, 2010 06:00
On Windows for example you could use performance monitor (perfmon) to determine IOps and Read/Write ratios as well as block sizes.
On UNIX I don't know (SAR perhaps ?)
1 Consideration though: if there are no bottlenecks this method works fine, but if for example a LUN on the old CX700 is stressed and a severe bottleneck, the values you'll get from measuring the IOps aren't optimal.
SKT2
1.3K Posts
0
April 28th, 2010 07:00
Extract utility on HP-UX can be used to generate IO/S. Normally generate this for chekcig the IO load. SOme times i see my systemreporting more than 22000 in a GDW environment.
Raju_auh
60 Posts
0
April 29th, 2010 00:00
If I implement ASM in oracle10g , do I really need to worry about the disks?
SKT2
1.3K Posts
0
April 29th, 2010 01:00
which OS? There is a mapping between device file to the respective ASM logical(we use mknod to create ASM devicename with same same minor number in HP-UX) device name. You need to make sure the they stay intact.
Raju_auh
60 Posts
0
April 29th, 2010 01:00
we are using windows2008.finally decided to go with ASM disk groups.As per my understanding ASM disk group needs disks(lun) of equal size.
read somewhere to use small luns as members in the ASM group.planing to create RAID groups(RAID10) with 4/6 disks and then luns with 100GB.is there any other better options?
YoBoy
10 Posts
0
April 29th, 2010 02:00
Hi All,
Please rectify one typing mistake as per my concern.
i.e. 10k drives deliver about 130 IOps. is wrong and 10k drives deliver about 150 IOps. is the right one.
Regards,
Kamal
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
April 29th, 2010 07:00
Kamal,
EMC's rule of thumb is 130 for 10k and 170 to 180 for 15k drives.
But like I said that depends on the actual I/O size, but if you need some quick math done, use 130 and 180.
Cheers,
RRR
cvasanth27
1 Message
0
April 24th, 2012 05:00
good one..