5

Currently I'm using ESX (v3 and v4) to test a scripted OS (Windows 2003) and application install DVD. The DVD ISO (8GB) is mounted on a 1Gbps NFS datastore and the VMDK's (20GB) are on an SSD mounted via NFS over a 10Gbps link.

It still takes a lot longer than I'd really like for to run through a test iteration and I'm wondering if mounting the virtual disks and ISO on a RAM disk on the same server as the hypervisor is running on would be worth my while. I can dedicate a server to this VM and 32GB of RAM in the system should be adequate to do the trick I'd guess. (1GB hypervisor OS, 28GB RAM disk and 2GB for the VM is < the 32GB available to me)

Since hosting a RAM disk within ESX does not seem possible I'm open to trying KVM/Xen/Hyper-V. KVM would probably be my first choice of these three.

Anyone out there tried this? Bear in mind this is purely for a test run of the installer, the VM will be discarded as soon as the test is completed so I'm not worried about losing data from the remote possibility of a power failure.

Improve this question
3
  • This can definitely be achieved, and can be effective. The above answers are very direct. But... I can see that you're using fibrechannel access to the SSD storage hosting the ISO. Have you considered limiting the latency by utilising local SSD storage?
    – LincM
    Nov 12, 2014 at 22:58
  • Look at the date on the posts, four and half years ago SSDs cost an arm and a leg.
    – Ausmith1
    Nov 12, 2014 at 23:06
  • Sorry, I should have checked that. But Gbit fibre channel would have also back then. Anyway, clearly didn't check the date. Sorry.
    – LincM
    Nov 12, 2014 at 23:17

2 Answers 2

Reset to default
2

I'm doing it on my 64GB ram server ;) Hyper V + VSuite RamDsik

Improve this answer
3
  • Awesome! I'm going to give that a try this week.
    – Ausmith1
    Apr 13, 2010 at 0:12
  • OK, so due to other priorities it took me a lot longer than a week to get back to this but I finally did. I used the VSuite RAMdisk software listed in this answer which is why I'm giving this answer the points. Unfortunatly my testing didn't give me the results I was hoping for. Using our stock Windows 2003 R2 install ISO it took over an hour to get to the the dialog to input the license key. This really isn't any better than an ESX VM on a FusionIO SSD linked via 10GBps NFS. Installing Windows 2008 R2 on a RAM drive is a completely story. Using a RAM drive it took 6.5 minutes (cont.)
    – Ausmith1
    Oct 18, 2010 at 21:11
  • whereas on a fairly fast RAID array it took over 20 minutes. I'd have to theorize that the Windows 2008 R2 media has built-in enlightened drivers for Hyper-V and that may be causing part of the speedup.
    – Ausmith1
    Oct 18, 2010 at 21:15
4

I've done it with Hyper-V.

The result will be that your storage latency will drop to near zero, which will very much accelerate your scenario. Your total CPU usage will increase some, as the CPUs will be copying data around RAM, rather than telling a storage controller to do DMA.

It's not clear to me from your description whether you intend to clean-install the OS as part of your test or whether the OS can be pre-installed on the virtual hard disk. This will greatly affect the speed of the test, as the Hyper-V storage drivers won't be installed in Server 2003 until late in OS installation.

As a final note, Hyper-V will need a little bit more RAM. So you'll have to use sparse disks that don't actually eat up 20GB. I suspect your test will run even faster if you give Server 2003 another 1GB, too.

Improve this answer
3
  • I'll be installing a clean copy of the OS but I can slipstream in any drivers that are necessary so assuming that the Hyper-V optimized SCSI Storport drivers for Windows 2003 are fairly standard it should be easy to do that. Realistic use of the disk in sparse mode under ESX is 14GB minimum and about 21GB if every possible option is installed. Of course the DVD itself is not needed after the base OS is installed so that could be dispensed with at that point... What RAM disk software did you use?
    – Ausmith1
    Mar 5, 2010 at 20:34
  • The Hyper-V storage drivers depend on other drivers, so I'm not sure how easy the slipstream is. Why don't you just use a differencing VHD that points at a clean OS installation? Then you can throw away the differencing disk after the test and you're back to your starting point? Then your test doesn't have to go through an OS install. I don't remember which RAM disk I used. Somebody pointed me to a web site. Sorry.
    – Jake Oshins
    Mar 6, 2010 at 6:45
  • Unfortunately part of the test is to install the OS from scratch. I can't get around that requirement.
    – Ausmith1
    Mar 8, 2010 at 15:40

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .