Posted by: MikeLaverick
when relevant content is
added and updated.
Well, I know VMworld ended last Thursday – however, I never got time to blog about the last day. I met up a friend of mine after the conference and ended up getting very drunk with her up in Haight-Ashbury… I went up to see my friend who lives in Menlo. We went for a walk on the beach (Half-Moon Bay?) and then on seafood resturant there, and then drinks in the evening. The next day went for another walk up in the hills, and out for a second nights worth of drinks. It seems I’ve been burning that old candle at both ends because on top of my hangover and jetlag – I also came down with a cold… still I’ve only myself to blame!
So anyway, today has been first real opportunity to rumage through the detritus of the conference and read my notes from those last couple of sessions.
Technology Preview: Storage VMotion
I tried to get to all of the “Technology Previews” – as this was really only places where “new stuff” from VMware was on show. That said for those in the know “storage VMotion” is not really a new technology. You might know it by its more popular VMware Community name “DMotion”. Basically, we talking about moving a VMs files from one VMFS volume to another whilst the VM is running. So its a storage management feature. Since ESX 3.0.1 we have been able to have an ESX2 and ESX3 host in the same VirtualCenter – and move both the VM and its file from VMFS2/ESX2 to VMFS3/ESX3 as part of a migration strategy. You can carry on doing this until the old ESX 2 server has no VMs left. It can then be then trashed and rebuilt. Basically, this functionality VMware have now labelled “Upgrade Storage VMotion”. In future releases were going to have this feature available for ESX3 data moves.
How does it work. Basically, before the virtual disks are “moved” (the actual process is more like copy, and delete the old files) a snapshot is engaged on the VM. This checks out the disk lease (read: unlocks the VMDK files in the VMFS volumes) so it can be copied elsewhere. Once the VMDK files (and all the smaller files) have been relocated, the snapshot is “committed” to the virtual disk. Simple.
There are some caveats and restrictions in what we were shown. Firstly, currently this can only be triggered via the CLI. This is a bit odd given how much ESX3i (with no Service Console) was trumpted, and the fact that VirtualCenter has an option to handle this kind of action for cold migrations and “upgrade” storage VMotion (its called “Move virtual machine configuration files and virtual disks”. We were promised VirtualCenter integration once the feature is release. Secondly, the “Storage VMotion” feature will not work with RDMs. So you won’t be able to relocate your storage held in RDMs. The feature won’t be incompatiable with RDMs it just will ignore them. Thirdly, you would think by the term “Storage VMotion” would allow for the VMotion AND the storage relocation of the VM to another ESX host. Think again: Right now, the VM stays on the SAME ESX host, and just allows you to move the files about. Again, it was said that this may well be adjusted before the feature is released to VMware Customers.
The caveats are pretty obivious if you are already familiar with VMware. Firstly, plenty of free space for snapshots on the VMFS volumes. Secondly, plenty of temporary RAM and CPU. During the “Storage VMotion” process affectively a “second” VM is created with the properties of the first. This temporary VM is used to assist the “Storage VMotion”, like an VM this requires an allocation of resources to function.
Sun Virtual Desktop Solutions in Theory & in Real World Deployments:
My main reason to attend this session was to say hi and offer moral support to a former student of mine, Chris Saul. Chris works at Sun in the EMEA region and covers a huge number of countries especially from what it sounds like in the Gulf Area. I believe he’s based out in Dubai right now. So it was interesting to learn from someone who’s been through the certified training – how they have got on in the real world. As the title of this breakout sessions states Chris’s primary work is in the field of VDI using Sun “SunRay” devices and services to act as the broker between the end-users terminal and thier Windows XP Desktop. I managed to catch up with Chris after the session and have lunch with him, where we chewed the fat about VDI deployments.
As a ex-Citrix guy I guess you could say I am skeptical about VDI. But Chris was able to flag up the key points that his customers have. Firstly, there are some very unhappy Citrix Admins and Users out there – who find Citrix expensive, and difficult to manage especially in a region where technical skills are slim, but there are plenty of guys who can install Office to Windows XP. Secondly, out where it is very hot and aircon is expensive, and power sometimes unreliable, being able to scale up very rapidly the number of seats without the need of lots of expensive power and space is valuable.
It still seems there are number of bug-bears with VDI that have yet to be surmounted. Firstly, 10-15GB virtual disk out our SANs is not “cheap” from a per MB perspective. Secondly, Microsoft’s RDP on the desktop side isn’t everything its cracked up to be. Thirdly, bolting down and install software on to 100+ virtual desktops is noway as difficult as doing it with 100+ physical desktops – but is still a management headache. Ironically, one of the big selling points of VDI is a fully-flavoured XP desktop – which we then bolt down with AD policies. Hopefully I’m not the only one hear who sees a contradiction there. Lastly, each on of these VDI desktops is going to need around 512MB of RAM for each Windows XP instance for good usage – that isn’t for free either. The bottom line is that this will reduce the concurrency. In the world of terminal services (albeit VDI, Citrix of MSTS) concurrency is king.
I think some of these challenges could be addressed with technology. The Lab Manager product from VMware already allows for one virtual disk with many snapshots. Imagine each one of your VDI users gets snapshot not the virtual disk. You would need plenty of free space for the snapshot – but no way near the 10-15GB per user that conventional VDI offers. Additionally, with just one virtual disk – patching, and software upgrades would be simplier. It would be like Citrix MFPS box, without Citrix. I’m not sure how would deal with the downtime required for this – perhaps there’s a technology solution to be had there. It could be done by simply having two Windows XP “Images” and toggling the users between them?
As for the concurrency issue – what we really need is copy of Vista “designed for” virtual machines with all the *rap removed that a VM doesn’t need. Microsoft could couple this with a very friendly license model to make Vista VMs much cheaper to run. In away they have already done part of this (see http://www.rtfm-ed.co.uk/?p=353)
Advanced VMFS Configurations and Troubleshooting:
This wasn’t a bad session. Unfortunately, I heard much of this before at the TSX in Nice this year, and partly at VMworld 2006. In fact this was a criticism I heard from fellow delegates, that there was an element of rehashing of previous conferrence material. In away I don’t mind this – after all who has time to attend ALL the breakout sessions in any conferrence, and sometimes hearing something for the second helps it “stick”.
Virtualization – The Alchemist for Grid Computing:
Mmm, I walked out this session half-way through. That was partly because I was hearing vanilla “benefits of virtualization” material. The reason I attended this session is I know zip about grid computing – and think this interesting because I know their is IBM hardware out there that is 4U that can be rigged together to make a signal computer. Is that grid based computing? I dunno? I found the session odd because the kind of people who run grid based applications are not the kind of people who would have that programme written for Windows or any other guest operatating systems supported currently by ESX. Is that a sweeping statement. I guess it is. I could be wrong. After all what do I know about grid based computing. Nothing, and certainly my knowledge wasn’t enhanced by this session! Sorry….