Symmetrix VMAX performance troubleshooting

1) Key metrics include FA/DA/Disk Utilization and IOPs.  Also the most important is host response time

2) I thought there were definitions in the help file.

3) VLUN can move a volume from one pool to another without disruption.

4) FAST movements can change the drive speed the data is on, so it can impact the response times.

Thanks for your help Quincy56.

I had few questions...

1) for a Meta TDEV when FAST VP is in place, is it better to have striped or concat and why ?? Performance wise which is better considering VP is striping every thing any way.  Does LUN sizes affect the decision ?

2) For a given Disk Group with 160 disks , how can we calculate IOPS of the disk group  available to the applications.??

FAST VP really doesn't change the performance for a VP device.  A striped meta will still have higher maximum performance than a non-striped. Size doesn't really matter, but there usually is a relationship with size to IOPs.  More than 8 members is likely to have no or very limited improvements in performance.

Depends on the drive speed.  You can assume the following for planning purposes (with good response times)

150 IOPs per 15k

120 IOPs per 10k

70 IOPs per 7.2k

1000 IOPs per EFD

Remember front-end writes will turn into multiple disk IO depending on the raid protection, 2 for mirrored, 4 for RAID5, and 6 for RAID6.

Ok, what I should have said from a front end point of view.

Obviously the backend performance will change if we move the device to a different drive technology.

Yes, the device is already wide striped, but even 100% cache operations can be significantly different between having one Symm device active on a LUN, vs. having multiple Symm devices active for a LUN.

The IOPs per drive is about what the drives will do for small block reads with 1 IO queued. Simple physics.

Quincy56,

FAST VP really doesn't change the performance for a VP device.

Above contradicting to what you said earlier as i thought FAST moves between tiers and change perofmance of the device.

My question was a concat thin meta ...every thing is distributed on each disk of the pool in a small 7680 chunks extents ...so for a 2TB  thin lun it will spread on all disk of that group .

Now as this is on all the 160 spindles, it will give the performance , regardless of concat meta or striped meta .

What do you reckon ?

2) Where did you got the IOPs from for the drives ?

Ok, so for a Disk Group with 160 SATA drives it will be times 70 IOPS = 11200 IOPS theoretically.

If you invert the single threaded IOPs, you get the response time.  So 70 IOPs is saying that the drive will get about 14.2ms response times, which is in the range of the expected performance of a SATA drive.

quincy,

makes sense now from fast VP side.

For metas...i think we might have to convert ..will it affect the host data access ?

For, IOPs sorry i still don't get it...how can we say all SATA drives with 7200rpm will have 70IOPS..

Sorry not been best with physics ...

Thanks quicy makes sense...i will have to see a least possible host impact way to convert concat -> striped..

Kavan,

It'd be good to plan the conversion in non peak hours. The host data access will not be affected but there is a chance the performance goes bad during the duration of conversion. And remember that it is a time consuming process, not a quick one.(could take 30 min to 1hr depending upon the size).  I have seen resopnse times going higher during the conversion process but got better after the conversion is done.

Santosh,

Take a scenairo:

2TB Concat LUN : A123 needs converting to striped with data protected. I know we need BCV devices but what configuration BCV device is needed and i guess same size as source LUN. Will it have to be RAID1 ?

Do i also need Target lun with say 2TB striped meta?? Can the target LUN,if needed be smaller, i guess not.

Also what about host LUN access to target LUN will it needs usual map/mask and how they will change this at application end, transfer from source -> tgt.

Santosh,

For conversion...is i the case that only source CONCAT meta LUN is there and BCV is used for data protection...whilst source get converted to striped....and the requests will be served by BCV. Thus, this process can affect performance.??

Trying to find a proper proecedure for this.

Kavan,

Another tgt device is not necessary for conversion as you mentioned in your first post. Your second post is true.

All that you need to have is a BCV meta that matches in size, config and member count with the source device. The BCV meta is used to protect the data during the conversion. (If you query the source device or BCV device during the conversion, you can see the Source-BCV pairs established and the source tracks protected).

The Source LUN can be accessed without any interruption all through the process.

The performance might get poor only during the conversion process.

The symconfigure command for the conversion looks like this:

convert meta , config=striped, stripe_size=1920, protect_data=true, bcv_meta_head=

Refer to the Symmetrix Array Controls CLI for more details about the conversion and expansion of striped meta.