VDI Stage 2

[heybuzzz]

I’ve been testing VDI on some of my existing ESX servers for about three months now and feel like it’s time to deploy a small group of about 20 users for additional testing. Some will be at home and some will be on campus, but they’ll all be connected 1 to 1 no broker.

I plan on adding these to my farm… two DL 380 G4’s, 2 Dual Core, with about 20 gigs of ram on each (SAN attached, V Motion enabled) I do not have HA. I plan on starting with 10 XP sessions on each so I don’t have all my eggs in one basket.

My question is… How would you implement this in terms of initial install for easy DR?

I could just load everything on the local disks and use ESX Ranger to snapshot my VM’s to my SAN.

I could load ESX on the local disk and host my VM’s on the SAN.

Or

I could boot ESX from the SAN and host everything there.

All three choices obviously have their advantages and disadvantages, but what I want is the easiest way to get my user back up in case of a failure to one of my servers. If anyone has any suggestions I would greatly appreciate them. Thanks

[nick_couchman]

I just upgraded from Starter to Standard for my servers when we bought a SAN. I doubt that I will be able to make a case for Enterprise for my VDI stuff by next year - maybe in '09...

[heybuzzz]

VMTINC - If a user royally screws up their session do you just create a new one from template?

I would like to just revert to a "GOLDEN" snapshot of their individual session taken with snapshot manager moments after I finish setting it up for them, but everything I read says don't hold on to snaps shot very long...

[Jim_ODonald]

I might be a little late to this discussion, but I've got 25 VDI's running on a single HP BL25p with (2) AMD Opteron 285's, 16GB of RAM, and boots ESX 3.0.1 from SAN. The VDI's are all running Windows XP, configured with 512MB of RAM and 10GB drive. The end-users are somewhat mobile in their workgroup and tend to move around during the day. Oh, and they all love it claiming that it is faster than the old PC's they had and can easily move between desks without having to restart applications when they move.

Jim

[mreferre]

Jim,

can I ask you :

1) what applications are they using ? (Office ? Notes ? Outlook ? Business apps ?)

2) what are they using as client devices ?

3) are you using a broker ? If not how do you "instruct" your end-users to loging to their own session ?

Thanks.

Massimo.

[Jim_ODonald]

Massimo,

The end-users are running Office 2003 and Outlook (connected to Exchange). Our line-of-business applications (mostly PeopleSoft) run in Internet Explorer, with the exception of a couple of lightly used specialized applications.

I have been looking at various thin-client devices, but haven't settled on any yet. In the meantime, I created my own thin-client with OpenBSD and rdesktop. I chose OpenBSD because of it's small footprint (I can fit the image onto a 512MB compact flash card.) I have a master image that I use to create thin-clients from -- it takes about 3 minutes to create a new thin-client from just about any old computer.

When the thin-client boots, it goes straight into the GUI where the end-users see a desktop of icons with their names. I let them pick their icons, so that makes it a little more personal for them. They double-click their icon and it connects to their virtual desktop (I have a 1:1 ratio for VDI's and end-users).

I'm trying to keep it really simple for the end-users. I don't like the idea of them having to log-on twice (once for the thin-client and once for the VDI) which is what I have seen so far with other thin-clients. I haven't fully grasped the connection broker concepts yet so that prevents me from purchasing one of those solutions. I'll get more information on them at VMWorld this year.

Jim

[mreferre]

Thanks Jim. Very interesting.

As per ...

>I don't like the idea of them having to log-on twice (once for the thin-

>client and once for the VDI) which is what I have seen so far with other

>thin-clients

This is true in fact single-on is a feature of the brokers and not the thin clients themselves....

Also, per you scenario, you might want to have a look at the 2x ThinClient Server product:

http://www.2x.com/thinclientserver/

There are two versions and the free one might do the job for you.

Also .....

http://www.2x.com/solutions/2XSolutionsguide.pdf

You might be interested in Example B (XP virtual desktops) at pag 17.

Thanks.

Massimo.

[Jim_ODonald]

Thanks Massimo.

I'll definitely take a closer look at 2x. It may be just what I've been looking for, and at a price that fits.

Jim

[chouse]

32 HP BL480c c-class blades with Dual Quad Core Intel Xeon processors and 24 GB memory. We aim for 50 VMs per blade, 100 per blade in an HA failover situation. 1300 VMs total.

4 clusters with 5-11 blades in each. Each cluster with HA & DRS.

2 VirtualCenter servers

2 HP EVA8000 arrays with primary and CA storage

Each cluster has 5-9 datastores of 540 GB in size, targeting 50 VM per datastore.

WinXP VMs with 296MB ram & 10 GB drives

Wyse V90 & HP thin clients at the desk with 19 inch and dual-19 inch LCDs.

Custom home-brew clustered broker to facilitate 1-1 connections between VMs, HP CCI blades & desktop PCs (broker used for remote access from off-campus as well)

[kreischl]

chouse,

How are you handling sound & video?

[chouse]

Don't know about sound as we don't have any sound drives in our XP VMs but video is handled by the Wyse video streaming driver I believe. The desktop part of the VDI project is handled by other people on my team so I don't know specifics.

[heybuzzz]

Chouse or anyone else using HA & DRS..

What do you set you "Isolation Response" to in HA Virtual Machine options. I've read a few threads going back and forth on the subject, but can't decide what to do.

I mean if one of my servers in the cluster can't hear the others heartbeat I understand that it will shut down the VMs and try to move them, then power them up on the other host. But, if the connection is down between both servers, but the connection out to the desktops is still working and the SAN connection is fine it will just shut the VMs down (Can't move them) when my user could still work. So, in that scenario I would like the VMs to stay up.

If I keep the Isolation Response to leave powered on then if there is a failure hardware failure how can it move the VMs? It would see two instances.

From the white paper;

Isolation response determines what a host in a HA cluster should do with running virtual

machines when the host loses its network connectivity (not receiving HA heartbeats and

unable to ping gateway). By default, virtual machines are powered off in case of a host

isolation incident. This releases their shared storage locks, which allows the virtual

machines to be started up on other hosts.

You can change this default behavior for individual virtual machines and choose Leave

running to indicate the virtual machine on isolated hosts should continue running even if

the host can no longer communicate with other hosts in the cluster. If you choose to do

this and it turns out that the original host can't access shared storage, the virtual machine

lock will time out and the virtual machine may be started on a second host (a condition

commonly referred to as split-brain). This condition is more likely to occur with NAS or iSCSI

storage, in the case of network failures, since both methods are TCP/IP based. For these

types of storage, keeping the Isolation Response at Power off (the default) is highly

recommended.

What's recommended for SAN attached???[/b]

[admin]

Hi Chouse,

i would limit the amount of vms per datastore to 20-25 to reduce the possibility of SCSi reservation conflicts.

[chouse]

Yes, we are scaling back and changing our SAN storage but not because of SCSI reservation conflicts. Everyone says to keep the number of VMs per datastore low to reduce such conflicts but to be honest we haven't seen that many because we aren't doing many lock-intensive operations - our VMs are always on, we don't use Snapshots, and VMotions are infrequent.

But either way, our pair of EVA 8000s could not keep up with the massive I/O introduced when all the VMs need to process a new update or application at the same time so we are moving our VDI datastores to new HP XP1024 storage and scaling back to 25 VMs per volume.

As VDI becomes more mainstream, we are starting to see recommendations from HP, VMware, and other sources about the recommended number of VMs per datastore. Prior to that, the recommendation always was up in the air because for most people, it depended on the I/O profile of the VMs - high I/O = low number of VMs and vice versa.

[heybuzzz]

Chouse or anyone else using HA & DRS..

What do you set you "Isolation Response" to in HA Virtual Machine options.

[OM4EVER]

doing a VDI pilot with 2 dell PE 2950s quad proc (dual core) 32 GB - ESX 3.0.2. Currently using Netapp for datastore storage with 21 VMs. We have HA/DRS configured and kept Isolation Response to the default. i really only realized the pros/cons from your post so it may be something to consider. From your article it seems that this is the safest way to proceed.

We're using Wyse ThinOS with provision CB and in stage two of the pilot we have issues with smartcard authentication going from the Wyse device to the CB. We would want our users to get a prompt for a pin. Anyone working on a config like this? using smartcards

[admin]

I can only recommend to stay with the default of "Shut down". If changed to "Keep running" the VMs will have problems being restarted on a different host. It is better to invest in stabilizing the network than to work around issues like spanning tree etc. with changing the node isolation default response.

Btw: we have customers running spanning tree events in under 8 seconds - so finally there is a way to control things like that in the network. And you are then safe because you stay under the limit of 15 seconds.

This is not different between SAN, NAS or iSCSI. The mechanism stay the same. In all IP-based storage systems any ESX host will not be able to write to its storage anyways when he loses network connection.

Message was edited by:

alexanderthoma

[Peter_Ghostine]

We're currently working with Wyse on providing support for smartcard authentication. The Provision broker currently supports smartcard authentication when using Windows PCs and XP embedded thin clients as endpoint devices. Please stay posted and expect this to become available before end of year.

Peter Ghostine

CTO / Co-founder

Provision Networks

[OM4EVER]

does it matter what version of Wyse is being used? We're using the thinOS v10L which I guess is proprietary. The work that's being done, is that on the Wyse side of things or do your developers have to do this for the thinOS. Are there any other devices out there that you are aware of that would do pass-thru authentication between from the thinclient to CB to the VM?

[Peter_Ghostine]

Both Wyse and us will have to make modifications to our respective systems to enable us to leverage the smartcard credentials. When the work is finished, you'll be able to upgrade your WTOS firmware on your s10's and v10's, and the broker software on the back-end.

This is similar to the effort we've undergone earlier where Wyse had to enhance their code to support our on-demand power-on feature, and we had to enhance the broker to provide a Wyse-compliant interface.

Peter Ghostine

CTO / Co-founder

Provision Networks

[OM4EVER]

That's good to know. As long as we're looking at a timeframe before the end of the year that will be good. November, december (no pressure)

We currently have a setup that we want to demo to management and this is a major stumbling block. We will be talking to Wyse tomorrow about that. Isn't there a way as a work around right now with your software and WTOS where we could manually put in some of these entries even if we have enter this info twice or manually without the pass-thru?

thanks Peter