Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

6469

May 19th, 2011 00:00

Possible to alter hyper size (VMAX)?

Hi

Is it possible to change the hyper size on a running system? If yes, is there some documenation around this ?

Thanks

2.1K Posts

May 31st, 2011 13:00

This is one of those situations where you are both right, but looking at it from different perspectives. You absolutely can't change the size of the hypers that make up a logical device. The device has to be destroyed and recreated. But if you have an array that is configured as you describe and it is causing potential problems (performance) then there are a few options to take advantage of larger hypers.

Options 1: If you are just trying to head off potential performance problems for the future, and you cannot afford the time and effort required for option 2 then you could simply define a new configuration for all new drives installed in the array from that point forward. So if you are currently configuring your drives with RAID5(3+1) and 30GB devices (so 4 x 10GB hypers per device), you could just draw a line in the sand and start configuring the drives with 90GB devices (4 x 30GB hypers). All new storage allocations (and any Metas) would be based on the new configuration.

Option 2: If you have more time & money and/or the performance issue is more prominent you may want to migrate your entire configuration to the new configuration. This would require a significant investment in disk so that you could build the required capacity (using the new configuration) to migrate all the existing data from the old drives. Once the old drives are clear you would destroy the devices configured on them and rebuild them all using the new configuration. This requires the money for a new set of drives and the resources to handle host based migrations of all the data from the old devices to the new. In the end you will have lot's of capacity for growth though :-)

It's a shame there is no virtual LUN Migration feature in the Symmetrix like there is in the CLARiiON/VNX family. That would make this kind of reconfiguration much less painful. Technically you could probably leverage some host based tools (either built into the OS or using PowerPath) for at least some of the hosts, but it is still more challenging than it would be on a FLARE based platform.

BTW - Please note that I'm just pulling numbers out of the air. I'm not taking into account any effort to maximize the effeciency of the hyper layout on the drives. I'm also not accounting for different drive sizes. Just keeping it simple for this example.

11 Posts

May 23rd, 2011 23:00

Hi,

Thanks for your responses. It is not a logical volume, I am talking about the actual hyper. The reason we want to increase it is that currently we have a lot of objects in our VMAX (object = everything you can create, vdev, datadev, snapdev...). Have had a few discussions with performace persons within EMC and if you want to squeeze the last performance out of the box, you should have larger hyper size to have fewer devs in a meta.

E.g. if hypersize is 10GB, you should create devices at at least 60GB if you have raid 6 6+2, but the larger the better, then use theese to form a meta. When you have e.g. 500 meta devs of 16 devs each, they use more resources than 500 meta devs of 4 devs each. Performance for the storage user is not an issue since theese devs are connected to a striped pool.

11 Posts

May 24th, 2011 06:00

No, we have discussion with EMC right now. If I get an clear answer I will see if can provide it.

180 Posts

May 24th, 2011 06:00

Hi, Have you tried asking this question in the Support Forums Community? You may find some additional insight there.

-Stephanie

11 Posts

May 24th, 2011 08:00

I've taken the Symmetrix Configuration Management class, thank you.

Let say you have 100 physical disks, each have 100 hypers. Thats 100 000 hypers in total, 100k objects which VMAX needs to keep track of.

If you tripple the size of the hypers, each disk would have 33 hypers, 33 000 hypers in total. This would consume less resources, since VMAX would have less objects to handle.

11 Posts

June 2nd, 2011 23:00

Of course! The hyper size is not a set fixed value, it is defined when you create a Ldev. E.g. RAID 6 6+2 is actually using only 8 physical disks. I've always thought that a ldev at 240GB, RAID 6 6+2, could have way more than 6+2 hypers, it could have 32 ((6+2)*4) to. But that is not true, of course.

Thanks everyone

No Events found!

Top