Just realised that ESXI is compatible with almost every MOBO I could lay hands on in my office. Installation basically depends on 2 things:
a. Intel Pro 1000 GTS NIC
b. Disabling C1E Support (http://james.com.sg/2011/04/09/installing-esxi-on-m4a89td-prousb3)
Got it working on GA-880GM-USB3….
Hmmm.. apparently I have been erring on the kiasu side. I have always went with thick provisioning with the concept that storage is cheap … and that I should always go for the performance.
Apparently, thick and thin provisioning has little impact on performance. It is only the eager-thick provisioning that has some impact for the first time writes.
http://www.vmware.com/pdf/vsp_4_thinprov_perf.pdf
If only I had known earlier… I would have gone thin… all the way…
Tried installing on an old motherboard of mine… Seems like it has no problems… all we need is the good old Intel Pro1000 GT NIC.
 |
 |
 |
 |
 |
Strange thing is… even the Intel CPU E5200 works.
But if you were to check up on the CPU,
http://ark.intel.com/Product.aspx?id=37212
You will find that it has none of the VT technology. Always thought we needed those to run ESXI
Ah… so this is the difference. When I tried installing a 64 bit application, I hit this warning message. Apparently VMWare can run 32 bit VMs without VT .. but not so for 64 bit VMs.
http://communities.vmware.com/thread/197291
Anyhow, stubborn me continued to try installing Ubuntu (64-bit) on it through my SAN (from Nexenta)… and seems to be doing just fine…
Same kind of performance I get for my other Load Test based on just 1 cable of I Gbps network cable.
Well, if you think you are getting good performance just by setting up your ISCSI initiators to point to the right VMKernels with multiple NICs in ESXI, you are mistaken.
Apparently, the default IOPS for LUNS in VMWare is set to 1000. (This controls how many IOS are sent down a given path before vSphere starts to route it down another path).
So the recommended best practise is to set it to 1 for every LUN.
Script:
for i in `ls /vmfs/devices/disks/ | grep naa.600` ;
do esxcli nmp roundrobin setconfig --type "iops" --iops=1 --device $i ;done
Compliments from:
Just realised that I have too many stupid disks that are just too large, i.e. 100 GB. I need to change the disk to thin provision.
http://virtualbox-provenpractice.blogspot.com/2009/04/thick-disk-to-thin-disk.html
Sort of like slimming down all my ESXI hard disks. Sample code to do conversion of thick disks to thin disks:
vmkfstools -i "DaVinci – Dynamic IP_1.vmdk" newfile3.vmdk -d thin
Was wondering how I could ssh into my VMWare instance until I hit upon this video:
Found an interesting article that allows us to run commands to the ESXI host.
http://virtualfuture.info/2009/01/run-esxtop-on-esxi-from-vima/
One of the main problems with ESXI is that… we are always running out of RAM. Yeah.. too many Virtual Machines…
So I am thinking of getting this :
RAM Monster
Well it seems like it has 32 slots for RAM.. that will work out to over 128GB if we use cheap 4GB DDR3 rams!
ESXI Compatible
And yeap! It is VMWare compatible. So glad that VMWare actually started maintaining their Hardware Compatibility List for server boards…
http://www.vmware.com/resources/compatibility/detail.php?device_cat=server&device_id=18842
Documentation
Well, my Link Aggregation is definitely not working.. which is kind of expected… ‘cos I am currently using a cheapo 35 USD 8 port switch… I mean.. if that works… it will be kind of amazing.
I tried running my benchmarks simultaneously on both VMS, i.e. Picasso and Davinci which are both on a LUN on my ZFS computer. Their bandwidth is like halfed…
(Compared to the blazing fast response at http://james.com.sg/2011/04/10/esxidisk-reads-benchmarking/)
You might be thinking.. maybe its your harddisk that is slow? Well, I actually did a
show performance network
on the nexenta system … and guess what… most of the traffic is still routed through e1000g0. Almost nothing at all goes to e1000g1. 
Guess will have to work on it tonight and try to upgrade it to a much nicer switch… this will have to wait till at night.. when nobody is at the office.. .and yeah! I got the whole office to myself again!
TRENDnet TEG-240WS
Okay. Just swapped in my TRENDnet…
After Trunk Settings
Link Aggregation Results
Hmmm… not sure if this means it is partially working… somehow for ZFS system, it magically dedicates one port to Tx and the other to Rx. So I guess the operation for ZFS is sort of working on the ZFS server.
However, for ESXI, it is definitely not working. Both VMs are still hitting at the same PortID 4.
I really need a switch to handle Link Aggregation. Otherwise, there is really no point in doing all those ZFS stuff when the choke point is at a gigabit internet. So here are some options:
Do not see to find any article on the internet on successful implementation of Link Aggregation for the above switch.
Seems that the above switch supports both static and dynamic link aggregation:
But ESXI only seem to support Static Link Aggregation
TRENDnet TEG-240WS
Well, I do have a WebSmart switch currently… and maybe it might work. I mean.. there is some kind of Trunk Setting in the configuration…
