This post is more than 5 years old
265 Posts
0
1892
VMAX3 , No meta Volume , what about WP
so VMAX3 have no meta volume !!
how can i achieve good performance without meta ,
what about Write Pending Limit WP , how is it calculated ,
does a big TB single volume going to have same WP% as any small volume
Quincy561
1.3K Posts
0
April 8th, 2015 12:00
The WP limit for a device should still be 5% of total writable cache, but since the cache is large, and there is no thick support, it should not be an issue. All volumes should have the same value, unless the device is smaller than the 5%, then the limit will be the size of the device.
We have performed lots of tests on a single device and compared to a single device on 76, or even a 64 way meta on 76, the performance of a single device is MUCH better on V3.
Mosa3lyan
265 Posts
0
April 8th, 2015 12:00
Thanks John ,
do we still have wide stripping on VP ,
if i connect a hosts to more than one port on same director , where both ports handled by same group of cpu cores will i gain anything / more cpu power
Quincy561
1.3K Posts
1
April 8th, 2015 12:00
No need for metas, trust me.
All CPUs on the director can handle all the ports.
You can use host IO limits to control performance for any storage groups.
Mosa3lyan
265 Posts
0
April 8th, 2015 12:00
it would be even better if we have meta , right ,
what about we dont have static mapping between ports and CPUs
so all cpu cores for FEs will server all FE ports and i cant isolate hosts on cpus based on performance
Mosa3lyan
265 Posts
0
April 8th, 2015 13:00
Thanks again
Quincy561
1.3K Posts
1
April 8th, 2015 13:00
Yes, optimized will try and give the best performance possible with the available resources.
Setting other SLOs will promote until the SLO is met, then it will stop promoting, where optimized has no limit.
Quincy561
1.3K Posts
1
April 8th, 2015 13:00
Yes, and TDATs are done automatically, the user can no longer configure the backend, so no way not to have it configured using best practices.
Using more than one port on the same director will add more throughput for large block IO, since each port will be limited by the port speed (800MB/sec for 8gb).
For small block, better to spread over multiple directors, also for availability.
Mosa3lyan
265 Posts
0
April 8th, 2015 13:00
optimized look like 100/100/100 Fast policy
Mosa3lyan
265 Posts
0
April 8th, 2015 13:00
thanks
Mosa3lyan
265 Posts
0
April 8th, 2015 13:00
is it recommended to use optimized SLO for all devs
rawstorage
419 Posts
0
April 9th, 2015 02:00
We don't currently have DCP on VMAX3, however the noisy neighbour issues should be a thing of the past. VMAX3 has a pool of CPU cores that can be applied to any Front end port on a director so bursty workloads can be dealt with quicker and more efficiently than before. This helps make it easier to balance resources. We call this feature dynamic core allocation and it's all part of the new dynamic virtual matrix.
I'd advise checking out the VMAX3 New Features WhitePaper on EMC.com there's a lot of good information in here describing the changes.
https://www.emc.com/collateral/technical-documentation/h13578-vmax3-family-new-features.pdf
Mosa3lyan
265 Posts
0
April 9th, 2015 02:00
do we still Have DCP ,
do we need it as we can isolate noisy neighbors by SLOs