VM performance is seems very slow

[Rody]

hi guys.

I have been playing with an all in one server for a while now, first I followed GEA's guide esxi, pfSense, Openindian +nappit, and a few other vm's for plex server, vent server, and others for fun, and learning.
Motherboard: ASUS P8C ws
CPU: Intel Xeon E3-1245V2
RAM: 32gb G.SKILL Ares Series (non ecc)
HBA: IBM 1015 itmode
DRIVES: 3X 3WD green 3tb drives.
1 segate 80 gb storage drive
1 crucial m500 120gb ssd.

When i started I did not have an HBA so i passed all my disks through with RDM to the vm's and this worked pretty good. My windows 8 vm had a 36gb raptor drive all to it self passed through with rdm and this worked very good.

last week I did a small upgrade by adding an ibm1015 in it mode and a small 120gb ssd, both passed through to a freenas vm. After some experimenting I decided to ditch OI +nappit vm for freenas because freenass seemed to be substantially faster on my hardware.

I installed pfsense and freenas on the segate 80gb data store, setup raidz on the 3x3TB drives and raidz (i don't remember the proper name for zfs on a single drive) on the single ssd drive, I then created an nfs share on a folder on the ssd so i could use it as a datastore on esxi and store my few os's on this super fast ssd.

First I installed windows 8.1 and it took a long time to install. I tried moving some files around to and from this vm and it seems to read files very fast (100+mb a sec), but writes very slow and fluctuates a lot any where from 5mbs to 30mbs. The whole virtual machine seems very laggy. The vm has 4 gb of ram and 4 threads, I really thought this would be very fast, but it was much faster when I had the 36gb raptor drive passed through with rdm.

is there a guide some where that can tell me how to make vm's that perform well or can someone please tell me what I did wrong? my google foo is broken or maybe I am just not surching for the right terms.

thanks in advance.
rody

 

[lopoetve]

disable sync writes on freenas.

 

[Vader]

I would also note that you incur additional risk when you do that...since you are essentially enable async writes, this goes for Metadata as well..and can lead to data integrity problems from system hard shutdowns..etc. I would also recommend a UPS to try to avoid that if possible.

 

[danswartz]

For sure. I run sync=disabled on my nfs datastore, but I also do hourly snapshots, and have a heavy duty UPS...

 

[Rody]

thanks a bunch, that really helped, of course it opened a new can of worms... should i turn my ssd into a zil/slog.

scratch that the crucial m500 120 is way to slow, but for now the sync disable works pretty good

thanks.

 

[danswartz]

well, honestly, if you have a good ups, and you do hourly snapshots, if it crashes and a VM gets corrupted, you can recover it in minutes from the most recent snap.

 

[Rody]

I was thinking that with a slog It might speed up transfers everywhere not just on the one nfs share that I have sync disabled, though I was pretty surprised to be getting 80+mbs from WD green drives.

 

[danswartz]

Not sure what you mean? In general, if you have no SLOG, it will put a ZIL on the pool zdevs, so not fast. Not that many if not most clients will use standard mode, so flushes don't slow things down. ESXi causes issues, since it forces sync mode even if the nfs server advertises async, hence setting sync=disabled (or having a good slog ssd.) Note if you are going to do this, get one with a super-capacitor.