Disk Enclosures
1752637 Members
5799 Online
108788 Solutions
New Discussion юеВ

Performance issue with Continuous Access

 
Alain Sen├йcal
Occasional Advisor

Performance issue with Continuous Access

Hi,

We have two EVA8100 with 96 x 450GB/15k FC drives each setup using only one Disk Group. So each EVA8100 has one Disk Group of 96 disks.

We have a 4GB ISL between the local and remote site switches. The distance from local to remote site is about 5km.

I've setup a CA synchronous replication for a 500GB Vdisk between the two EVA8100.

As soon as the full copy started, we noticed a major I/O performance issue on other Vdisks (which are not replicated with CA) on the source EVA.

Why is it so and how to prevent this is the future?

Thanks

3 REPLIES 3
Greybeard
Esteemed Contributor

Re: Performance issue with Continuous Access

your 500GB vdisk will span the physical spindles of the drives in the group, as will ALL of your remaining vdisks, the I/O is therefore split between the vdisks, if you wanted to combat this then put some disks in another group for the CA volume (leaving a good overhead ) and run it on it's own. For performance and resiliance I would also create a minimum disk group (with overhead) for the system's LUN0 to live on too.
_________________________________________________
How to assign points on this new forum? Click the Kudos Star!
Amar_Joshi
Honored Contributor

Re: Performance issue with Continuous Access

Possibly there could be two reasons either disks spindles are being bottleneck or the controller is busy.

1. If you just created this 500GB Vdisk and running the full sync, there is a possibility of diskgroup releveling and also read intensive sync running which will compete with other hosts (specially if they are also read intensive and worse if they have random IO pattern).

2. CA replication may overload the HSV controllers hence you need to check for HSV load and also if they are balanced.

Use EVAPerf and see which resources are overloaded, you may hit bingo on disks or controller.

Uwe Zessin
Honored Contributor

Re: Performance issue with Continuous Access

> for the system's LUN0 to live on too.

The EVA's LUN_Z is a controller [type 12(10)] LUN. It does not use a virtual disk. Even if it did - creating a separate disk group would waste^H^H^H^H^Hrequire a minimum of 8(!) disk drives.

On the other hand, there is at least one disk drive per disk group (called the 'quorum disk') which holds the EVA's meta-data. This is fully automatic with no control by the user.
.