Unsolved
This post is more than 5 years old
1 Rookie
•
63 Posts
0
668
Net effect of 100% EFD FAST VP Tier?
We have a request for a test LUN to utilize a 100% EFD lun. We will use FAST VP to achieve this, instead of binding directly to the EFD thin pool.
Question is, what is the net effect of a FVP policy configured to use 100% EFD and 1% FC, if the goal is to have the LUN use EFD 100% of the time? The SA is concerned that the 1% FC allocation may possibly (though not by much I think) skew his performance testing results. We have explained this is the only way to achieve this via FAST VP, as we cannot remove the FC tier since the LUN is originally bound to the FC tier.
Feedback is much appreciated
sauravrohilla
859 Posts
0
April 15th, 2013 19:00
Well, you are right. You cant have just one tier in Fast VP policy so it is required to have 1% in FC I would also suggest to enable VP allocation by policy feature, this will allow your writes to directly go to EFD tiers when the previous write was promoted to EFD thus bypassing FC (parent) tier.
hernster
1 Rookie
1 Rookie
•
63 Posts
0
April 16th, 2013 06:00
Saurabh,
Correct me if I am wrong, but the feature you mention "allocation by policy" is only in 5876 code, correct? We are still on 5875.x and the planned upgrade to 5876 will not happen before this current testing. I should have noted that previously.
Appreciate the feedback!
RobertDudley
448 Posts
0
April 16th, 2013 11:00
if you watch the pool allocations during the testing then you will be able to see if data is being written to the FC pool.
Just as an FYI I did a test with an AIX database team with a 100GB pre-allocated all EFD lun. The AIX admin was able to generate 44,0000 IOPS (random) on this lun with a 24 CPU AIX P6 with 8 FC paths to the host; powerpath was installed. The lun was built on a group containing 24 EFD disks. This was a bit over two years ago on a VMAX20K.
RobertDudley
448 Posts
0
April 16th, 2013 11:00
If the test parameter is to utililze a 100% EFD lun why are you not just creating a lun in EFD for the testing? To answer your question yes the 1% of storage in the FC tier can certianly skew the results if data is written to it.
hernster
1 Rookie
1 Rookie
•
63 Posts
0
April 16th, 2013 11:00
Robert - Fair question. As it stands we are not to create any devices within any thin pools outside of FC thin pool. Given that constraint, this was the only option, for the purposes of this test. I raised the same question, and management here wants all luns to be controlled by an FVP policy, no one-off bindings to other pools, even when testing. Thanks for your response, much appreciated.