Dynamic Memory and RemoteFX coming to Hyper-V R2 with SP1

Windows Virtualization Team Blog : Dynamic Memory Coming To Hyper-V

The Dynamic Memory looks pretty sweet. Its not exactly overcommit, but then again overcommit feels dangerous if you are running mission critical apps. (queue HeroPsycho war

)

Find more posts tagged with

Comments

astorrs

I may not be HeroPsycho but I can't resist the opportunity to toss one over the fence.

Hyper-Me

astorrs wrote: »

I may not be HeroPsycho but I can't resist the opportunity to toss one over the fence.

It just seems like something along the lines of a memory leak could happen and eat up lots of the memory on a host and affect the other guests.

Is it even recommended for use in production scenarios?

EDIT: Nevermind, answered my own question

http://blogs.vmware.com/virtualreality/2008/10/memory-overcomm.html

It appears that Dynamic Memory is basically a different name for overcommit anyway. lol

astorrs

Hyper-Me wrote: »

It just seems like something along the lines of a memory leak could happen and eat up lots of the memory on a host and affect the other guests.

Is it even recommended for use in production scenarios?

Like I said most of the confusion/worry people have over the 3 technologies that get lumped together as "memory overcommit" go away once they learn enough about how they work.

Not any more likely (and I would argue less given the small driver base on ESX) than a memory leak on a Hyper-V box (where I could see a 3rd party network/storage driver leaking memory from the parent partition potentially crippling the entire host).

Almost everyone uses some of them (even if they don't know it, e.g. transparent page sharing). Whether or not people provision more memory to running guests than they have physical seems to be at 50%.

I have no qualms about doing it if you have DRS enabled on the cluster, RAM is expensive (think about it for a second) as you only have so many slots per server and increasing DIMM density drastically increases the price of each server and why pay that if 99% of the time the memory is untouched. Remember overcommit would only be a problem if most of the memory in all the nodes in the cluster was used (otherwise DRS would just move a few VMs off a host that was approaching 90% utilization).

astorrs

Hyper-Me wrote: »

It appears that Dynamic Memory is basically a different name for overcommit anyway.l

Actually until I hear more of the technical details, it appears to only be allowing the hypervisor to swap on behalf of the guests. Unless they've added a balloon driver and something along the lines of TPS, I would be scared s**tless of using it.

Hyper-Me

astorrs wrote: »

Actually until I hear more of the technical details, it appears to only be allowing the hypervisor to swap on behalf of the guests. Unless they've added a balloon driver and something along the lines of TPS, I would be scared s**tless of using it.

The best I could gather is there will be a minimum RAM allocated to the guest, as well as a Maximum that it can go up to. As with anything, it would need to be tested thoroughly before putting anything critical on it.

astorrs

Hyper-Me wrote: »

The best I could gather is there will be a minimum RAM allocated to the guest, as well as a Maximum that it can go up to. As with anything, it would need to be tested thoroughly before putting anything critical on it.

It's worse than that, if your only option is swapping, then getting to 100% memory usage will be disastrous since the hypervisor will be paging indiscriminately vs. pushing it back to and forcing the guest to page through a balloon driver like ESX does (thereby forcing the guest O/S to decide what's actually needed in physical memory).

Hyper-Me

It'll be interesting to see how it plays out.

Perhaps Hyper-V is behind the curve on some performance aspects vs ESX, but its always been totally sock solid dependable. I've never had any kind of weird failure of a hyper-V guest or host out of over 200 guests running 24/7. Not saying ESX isnt reliable, we run most of our inhouse stuff off of it at my new job.

HeroPsycho

Hyper-Me wrote: »

It'll be interesting to see how it plays out.

Perhaps Hyper-V is behind the curve on some performance aspects vs ESX, but its always been totally sock solid dependable. I've never had any kind of weird failure of a hyper-V guest or host out of over 200 guests running 24/7. Not saying ESX isnt reliable, we run most of our inhouse stuff off of it at my new job.

There's no perhaps. Hyper-V R2 SP1 still doesn't have Transparent Page Sharing. Agreed with everything astorrs said. The only issued I've seen with ESX memory overcommit were people who completely misconfigured ESX, and you can't blame VMware for idiotic things people do who don't know what they're doing.

jasonboche

I may have found my long lost brother.

dynamik

jasonboche wrote: »

I may have found my long lost brother.

If you're referring to HeroPsycho, I'm not at all surprised he was put up for adoption.

There's no denying he's got this virtualization stuff down pat though*.

*So does Andrew, but his family loved him...

HeroPsycho

jasonboche wrote: »

I may have found my long lost brother.

Hey, awesome products are awesome products, and idiotic administration is still downright scary. I just saw a Hyper-V deployment for an SMB where the host was only running Hyper-V and a supporting backup app. Physical box had 12GB of RAM. Two VM's, one an Exchange 2010 tri-role server with 4GB of RAM, and a 64-bit W2K8 DC with 1GB of RAM.

Congratulations, newbs, you just starved both VM's to preserve 7GB's of RAM for the host partition that didn't need it.

So weird that the Exchange Management Console intermittently failed due to lack of timely responses from the DC, wonder why!

MentholMoose

The description of Dynamic Memory is pretty vague so who knows what it really does. To be sure, MS will bill it as the best thing since sliced bread and continue lambasting VMware's feature set, that is until more of it is implemented into Hyper-V.

RemoteFX looks good on paper since RDP is long overdue for improvement. I don't know why the blogger specifically mentioned that Citrix will implement it into XenDesktop, considering that a main selling point for XD is ICA, and it sounds like RFX is a competitor to ICA. If RFX is good and RemoteApp and Microsoft VDI finally have a decent presentation protocol, Citrix may have a problem.

astorrs

MentholMoose wrote: »

RemoteFX looks good on paper since RDP is long overdue for improvement. I don't know why the blogger specifically mentioned that Citrix will implement it into XenDesktop, considering that a main selling point for XD is ICA, and it sounds like RFX is a competitor to ICA. If RFX is good and RemoteApp and Microsoft VDI finally have a decent presentation protocol, Citrix may have a problem.

RemoteFX is Calista merged into RDP (Explaining Microsoft RemoteFX).

Citrix will leverage and extend it with HDX RichGraphics with RemoteFX.

MentholMoose

astorrs wrote: »

RemoteFX is Calista merged into RDP (Explaining Microsoft RemoteFX).

Citrix will leverage and extend it with HDX RichGraphics with RemoteFX.

Yeah I saw the first link. A lot of the benefits seem to match what's advertised in ICA/HDX.

I found this article about RFX:
Microsoft announces "RemoteFX," the Calista-based Hyper-V-requiring PC-over-IP competitor. Here's our full analysis.

Apparently if the remote host is a VM then RFX requires Hyper-V, so to even use it with XenDesktop you can't even use Citrix's own virtualization platform. I can see how it can extend XenDesktop, though (some HDX features require a GPU to work, so the remote host can't be a VM to use those features).