StoreVirtual Storage
1748256 Members
3955 Online
108760 Solutions
New Discussion юеВ

Re: VSA on Hyper-V Performance Question

 
Drew Dubber
Occasional Advisor

VSA on Hyper-V Performance Question

Hi

 

Just created a 2 node VSA setup at home as a test.  No massive datastores or top end kit therefore - 2 x M5014 + BBU in different hardware hosts, both with 4 x 3TB 7200rpm disks in a RAID-10.  Both hosts have Windows 2012 + Hyper-V with only one VM so far - a VSA.  Both nodes have a 10GBe card installed with 6144MB RAM allocated to the VSA.  I've hooked up a copy of ESXi to the datastore I created, also with a 10GBe card.

 

Problem for me is I can get 30MBps performance to the datstore tops.  Looking at the performance figures nothing seems to stand out as an issue.

 

If I do a SMB copy of a bunch of data between the hosts (so I guess I remove ESXi + Hyper-V+ VSA from the equation), I get about 200MBps data transfer sustained.

 

So I dont expect the 200MBps since there is the replication and other overheads but is the maximum data transfer of the setup I described above really as low as 30MBps?  

 

Thanks for reading!

 

10 REPLIES 10
oikjn
Honored Contributor

Re: VSA on Hyper-V Performance Question

30MB is rather low, but if you have your cluster sizes setup funky or something else going on you could see those kind of speeds.  What does windows resource manager say you are doing when you are running these tests? 

 

Might want to check flow control and also make sure yoru drives are set to write-back and not writethrough for raid...  writethrough w/ 7200RPM random IO would be pretty poor.

Manfri
Frequent Advisor

Re: VSA on Hyper-V Performance Question

First test the performance of disk vith IOMETER

SMB do lots of caching with windows 2012 so it may disrupt your test

Second test from the Hyperv Hosts the performance of VSA with IOMETER on a LeftHand Volume

Third

 

check the networking side: you have 3 machine right?

10GB nic aren't cheap so thery must work well with hyperv... you  you have tested SMB with NIC on hyperv switch you use for ISCSI to verify that there aren't interference from Hyperv Networking? have also tested with iperf?

 

ISCSI is also finicky about jumbo packet, flow control and so on

 

the nic use jumbo packet and or flow control?

the switch use jumbo packet and flow control?

i do not remember if VSA support jumbo packet.. so verify

do you use any link aggregation/teaming?

 

ping -f 8000 is your friend...

 

Drew Dubber
Occasional Advisor

Re: VSA on Hyper-V Performance Question

Thanks to you both for replying. I'll do some more tests over the ISCSI network and report back. I've run CrystalMark / Atto on the disks - they are well in excess of 200MB/s. Caching I don't think has anything to do with it on SMB - I'm using a 40GB image to test so I don't think it can cache that much. I'll download iperf and I have a look at that. Windows resource manager says nothing going on, its having a holiday :) RAID cache setting is fine; both cards have BBU. As using 10GBe no teaming / aggregation. I haven't used jumbo packets yet but I didn't think it would make *that* much difference.
Will come back after more tests - cheers for the advice
oikjn
Honored Contributor

Re: VSA on Hyper-V Performance Question

jumbo frames definitely helps with 10gbe, but that is generally because there is a physical packet rate limit and if you are using packets too small you will hit that rate limit before you hit the bandwidth limit of 10gbe. I don't think that's your issue here, but turning on jumbo frames through your 10gbe iscsi network should help. to troubleshoot you really need to find out what your bottleneck is and generally on VSAs its either disk latency/IO limitation or CPU.
Drew Dubber
Occasional Advisor

Re: VSA on Hyper-V Performance Question

To be fair I have 2 boxes but they arent the same config.  One is a supermicro based build with 12GB RAM and 2 x e5607 CPU's.  The other is a HP Microserver G7 with 10GB of RAM.  CPU usage is nothing according to resource manager on both servers, as it the network utilisation.  The disk setup looks good to me but need to test it out more as per the other email (plan to do this tomrorow!) 

oikjn
Honored Contributor

Re: VSA on Hyper-V Performance Question

think of each VSA node like you would a disk in a traditional raid array.  The raid group (here the VSA cluster) will only be as fast as the SLOWEST node in the cluster and only use as much capacity per node as the SMALLEST node in the cluster.  So the last thing you want to do is take a small SSD VSA and pair it with another VSA that uses huge "green" 5400RPM SATA drives.

 

If the cuslter was recently configured, there may be some activity going on in the background of the cluster, but to check things out, I would watch CMC stats on the cluster as well as perfmon w/ the Host to see where the bottleneck is exactly or at least which node is holding up the other. 

Drew Dubber
Occasional Advisor

Re: VSA on Hyper-V Performance Question

As per my original post both hosts have 1 x M5014 + BBU , both with 4 x 3TB 7200rpm disks in a RAID-10. They both most pretty much the same disk score in CrystalMark etc. Will have to get around to doing more tests as soon as I have more time but it isn't something obvious as this unfortunately.
oikjn
Honored Contributor

Re: VSA on Hyper-V Performance Question

eventually you will find one stat that is obvious and the holdup. 

 

you don't mention the CPU on the HP system.  Single threaded CPU speed is very important for VSAs.  Its easy to max out their CPU and even through they let you put in two vCPUs since 9.5, CPU can be an IO limiting factor.  The e5607 @2.26ghz is only slightly faster than the minimum recomendation of 2ghz.

 

no matter what, you should get >30MBps, but that is assuming you aren't somehow doing something to the system to overstress it...  Aka, 30MBps @ 512b random IO would be rather amazing w/ your setup, but is crap if its @ 64k sequential.

Drew Dubber
Occasional Advisor

Re: VSA on Hyper-V Performance Question

The CPU on the HP is a simple AMD TurionтДв II Model Neo N54L.  I run resource manager on both Windows 2012 boxes during say a VM suspension and neither of the hosts is stressed & the HP stats within SANiQ or whatever its called idles as well.  Hopefully we get some time this weekend to test and provide some screenshots.  Maybe I am just expecting too much?  At the moment I am really testing it by setting 3 VM's to suspend at the same time.  Its way slower than a test setup I have on ESXi.