VMware Cloud Community
usao
Enthusiast
Enthusiast

Managing VMFS datastores.

Im running into a issue with VMFS datastores.

I am creating the datastore from 8 luns presented from a Violin flash storage array.

For maximum performance (in throughput), we need to stripe accross 8 luns. However, the datastore appears to be concatenating the luns together and thus im not getting the full speed out of the Violin storage array.

Is it possible to get VMFS to stripe accross the underlying luns rather than concatenate?

Our goal is to get around 10 GByte/sec from the VMFS datastore.

When we presented the 8 luns to a Linux host and created a striped filesystem, we were able to achieve this level, but with the existing VMFS datastore, I seem to be capping out around 2.5GByte/sec which is about 4 times slower than desired.

2 Replies
MKguy
Virtuoso
Virtuoso

This is expected and by design. ESXi doesn't support any kind of RAID or striping of VMFS volumes across multiple physical block storage devices (the only exception is VSAN, which does something similar).

This job should be done by the storage array in the backend and not by attached end-hosts, what's holding you back from creating one large LUN from all the disks on the storage array?

-- http://alpacapowered.wordpress.com
0 Kudos
rcporto
Leadership
Leadership

Like already stated on previous post, this is the expected behavior, and you can found a explanation here: VMFS Extents - Are they bad, or simply misunderstood? - VMware vSphere Blog - VMware Blogs

Here is just a brief about the first misunderstood:

Misconception #1 – Extents are like RAID stripes

This is one of the common misconceptions. I’ve seen some folks believe that the Virtual Machines deployed on a VMFS volume (with extents) are striped (or the file blocks/clusters allocated to the VMs are striped) across the different extents.

This is not correct. Extents are not like stripes. If anything, extents are more akin to concatenations than stripes. They do not rotate Virtual Machine assignments or even VM block or cluster allocation assignments across different extents in the datastore.

---

Richardson Porto
Senior Infrastructure Specialist
LinkedIn: http://linkedin.com/in/richardsonporto
0 Kudos