LeftHand StoreVirtual as Datastore performance?

N

Nate7311

2[H]4U
Joined
Jan 11, 2001
Messages
3,320
Anyone here using any physical LeftHand stuff as iSCSI datastores? Seeing a peculiar issue as we are asked to build and integrate a fresh 5.1 environment into a LeftHand SAN consisting of 4 nodes, a pair of older P4500G2s and a pair newer P4530s. 10GbE connectivity all around between the nodes and to the Hosts through a pair of Cisco 4500x-32s. Of course SAN traffic is on it's own VLAN, and each Host has 2 x 10GbE paths set as Round Robin.

Looking at Performance tool the LeftHand management console, Disk Queues fly up to 40+ during the write phases of each test. And on in vCenter, Latency rises above 100-140 during write phases on the SAN volumes. Of course Local storage latency stays under 6.

I know there's a few LeftHand admins around here, any ideas? HP Engineers installed the new nodes and tied in the older nodes into the cluster, and calls have been made to them as well.

A QUICK test of Crystal DiskMark in a test Windows 2008 R2 VM yields the following numbers:
Mind you that these are EXTREMELY early tests to verify performance. IOMeter testing is running now.

Left Hand RAID 10 Volume
musuB3K.png


Left Hand RAID 0 Volume
OEK7HKM.jpg


LOCAL STORAGE - RAID1 - 500GB NLSAS
6gyHZYD.png
 
Last edited:
Hmm That write speed is wrong. Shouldn't be that pathetic. I get higher writes on my ancient qnap iscsi store. There is something definitely wrong.
 
Right, normally I wouldn't put that much stock in a basic test like this in a fresh environment, but it's a quick verification that at least the basics are happy. I've got a few rounds of IOMeter running now. Being 10GbE all over, I was expecting more like 400+MB/s Seq. for both Read/Write considering all nodes have 600GB 15k Disks.

Drilling into the Troubleshooting reports on the nodes, it looks like the older ones have a P410/512mb RAID card optimized at 50%read/50%write. Still digging into the new nodes. Where is caching setup in the LeftHand Management console?
 
I have no idea - never used lefthand. Just a generic iscsi write slowness thought...
 
I take it you have the latest DSM installed on your hosts..etc?
 
There's a DSM for vSphere? We were told by the HP engineers that there wasn't a specific plugin for the hosts. My research only showed a DSM for Windows. My testing was inside a VM, not a direct iSCSI connection.
 
Found some more info. Looks like the new nodes had the famous HP Raid card write caching issue where all cache is dedicated to read ops only. Multiple patches were required to update this setting. Correction to 50% read/50% write returned a much better write numbers. Even running multiple tests simultaneously. Hopefully someone else can benefit from this.
 
You must log in or register to reply here.
Back
Top