Unsolved
This post is more than 5 years old
96 Posts
0
3045
Pool & Logs
IHAC - VMAX with 8 engines and 8 bay filled with 600GB FC
Pools are on Raid5. each pool has 64 spindles. Presented Data & Log ThinDevice luns from these pools (SQL / SAP application) for windows 2008 server.
Application team has come back to us saying performance is very low when they do microsoft speed test on these luns presented to their servers.
Latency is the issue. IT should be less than 10-20 ms as per their requirement. All allignement etc etc is taken care of at server side.
It is a new setup built about 5 months back.
Any best practice missing here from storage side?
If so can i have a validated document?
any suggestion to improve these?
Thanks in advance
Pras
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
July 25th, 2012 03:00
how many pools do you have and why are they so small (only 64 drives ? )
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
July 25th, 2012 04:00
interesting, have you or EMC looked at pool/disk/FA utilization ? Maybe pools are too small and you are working them too hard and need to buy more spindles to give you more iops.
pmenon5
96 Posts
0
July 25th, 2012 04:00
16 pools and each pool configured with 64 disks as per EMC
Quincy561
1.3K Posts
0
July 25th, 2012 07:00
That seems like more pools than needed for isolation, but I doubt the problem is a backend one, but it could be.
First, are you using striped meta volumes? And how many FA CPUs?
Quincy561
1.3K Posts
0
July 25th, 2012 08:00
Ok, can you post STP data? How many members?
pmenon5
96 Posts
0
July 25th, 2012 08:00
Striped Meta...FA CPU's need to check
pmenon5
96 Posts
0
July 26th, 2012 05:00
Logged case. EMC support investigating the issue.
RobertDudley
448 Posts
0
July 26th, 2012 06:00
What are you using for multi-pathing? if its powerpath what does it show while running the test for pathing metrics?
pmenon5
96 Posts
0
July 26th, 2012 07:00
Powerpath and it is sym_opt
DineshPunyani
54 Posts
0
July 31st, 2012 00:00
@Pmenon5 you have mentioned that you have bound ThinDevice LUNs and have created Striped Meta.. This will cause a performance issue... For better performance Striped Meta should be used for THICK LUNs and Concatenated Meta Should be Used for THIN LUNs .. Backend layout of THIN Luns is different then THICK LUNs , IF you are using STRIPED THIN META then it means you are performing STRIPING on STRIPED LUNS which will give you less performance and in this case available spindles are not used efficiently.
sauravrohilla
859 Posts
0
July 31st, 2012 02:00
Dinesh, its not true. Check the latest FAST VP best practice guide. Stripe is recommended over concat provided you have the newer code of enginuity as that allows to expand the striped tdev online.
regards,
Saurabh Rohilla
Quincy561
1.3K Posts
0
July 31st, 2012 06:00
Yes Dinesh, it is not true. Striped metas on VP (thin) have significantly better performance than concat.
rmueller2
23 Posts
1
July 31st, 2012 21:00
From which point on would one start to use Thin Metas? The maxium TDEV size is about 260GB with 5875. If the requested device (e.g. Exchange 2010 edb drive) is just 100GB, is there any reason to build it of smaller (striped) metas for performance? Thin VP is already striping extends across all spindles in the Thin-Pool as far as I know.
DineshPunyani
54 Posts
0
July 31st, 2012 22:00
Saurabh,
As per my understanding When you are writing data to a THIN LUN actual data is striped across all the DataDevices of THIN Pool to which Thin Lun is bound (which means striping is already happening in backend in case of VP).
Please correct me if i am worng.
sauravrohilla
859 Posts
0
July 31st, 2012 23:00
Hi Dinesh,
You are 100% correct, data is striped on the backend pool in case of Tdev but saying that double striping will cause performance impact and Concat gives more performance benefit than Stripe tdev meta is not fully correct.
regards,
Saurabh