This post is more than 5 years old
9 Posts
1
1726
Vmax performance questions
Hey guys,
Trying to explain the behavior I see in a customer Vmax 10k,
I see SATA disk group very busy and FC and EFD not much at all.
When I drill down I see the disks busy but when I go one level deeper hypers, there is no host IO all BE stuff,
Does this mean it is the BE director destaging from cache to drives? And how to fix it? So FC and EFD are helping more.
Customer said at the first he was binding thin devices to SATA but now knows FC is preferred.
Thanks!
Quincy561
1.3K Posts
0
October 16th, 2014 12:00
As Sean said, check your policies, and also make sure FAST is enabled!
And if you are on the later 76 versions of Enginuity, you should have your FRR (relocation rate) set to 7 or higher. The old default of 5 can be pretty aggressive about moving stuff on the newer releases compared to previous releases.
LionR
9 Posts
0
October 16th, 2014 12:00
Thanks a lot guys!
Sean great blog post. Very valuable info.
Quincy, yes I found busy storage groups bound to SATA not participating in FASTVP.
Current FRR is 5, should I change to 7 this is the version info.
Microcode Version (Number) : 5876 (16F40000)
Microcode Registered Build : 28
Microcode Date : 12.02.2013
Microcode Patch Date : 12.02.2013
Microcode Patch Level : 251
Symmwin Version : 161
Enginuity Build Version : 5876.251.161
seancummins
226 Posts
2
October 16th, 2014 12:00
Hi LionR,
When I see this sort of thing, it's often because of admins binding TDEVs to SATA (as you noted), and/or using policies that trap workloads that they consider less important in the lower tiers. It sounds like they're already planning on rebinding from SATA to FC, so that should help. They might also want to review their FAST Policies, bearing in mind that just because something is less important to the business, doesn't mean it's a light workload. There are some cases where test/dev/UAT workloads can actually be higher than production.
Trapping these heavy test/dev workloads in the lower tiers can end affecting production workloads, because these tiers are used by both production and test/dev. The easiest way to resolve this would be to start by using a 100/100/100 policy for everything. Then if you need to throttle certain workloads, use Host IO Limits, and/or gradually reduce the amount of EFD/FC capacity available to test/dev.
Here are a couple good resources with more info:
FAST VP Best Practices Whitepaper: https://support.emc.com/docu31003_FAST-VP-for-Symmetrix-VMAX-Theory-and-Best-Practices-for-Planning-and-Performance.pdf?language=en_US
FAST VP Best Practices blog post (kind of a summary of that whitepaper): http://blog.scummins.com/?p=87
Thanks,
- Sean
LionR
9 Posts
0
October 16th, 2014 13:00
Will do,
Now you just have to accept my invite in Linkedin
Thanks Guys!
Quincy561
1.3K Posts
0
October 16th, 2014 13:00
Yes, I would recommend changing the FRR to 7 in the long term, but if you change policies to try and get some of this active data off of SATA, a lower FRR might make the movements happen faster.