Latest posts by Paul Schnackenburg (see all)
- Project Honolulu - A new way to manage Windows Server - Wed, Nov 22 2017
- Use Azure Managed Service Identity (MSI) to store passwords in your code securely - Thu, Nov 9 2017
- Azure Data Lake overview - Fri, Sep 22 2017
Paul Schnackenburg - Ben Armstrong
PS: So the first and most important question I’ve got for you is that for first time ever you actually beat Jeff Woolsey in the speaker scores at TechEd US 2012.
BA: Yes I did, very happy.
PS: Now according to him this was apparently because you bribed people?
BA: Absolutely, I mean –at the beginning of my session I made an appeal to the people, I said I’ve been competing with Jeff for years now and the show that we’d done prior he beat me by 0.02. So I made an appeal to the people, the bribing did happen but it was completely accidental, we had some spare Hyper-V posters and actually one of my – one of the people from my documentation team brought them in, so I had posters to give away too – but I did beat him so I was very happy.
PS: I asked him this morning so that I could corroborate that this was the true story. So some more serious questions and you actually alluded to this in the first breakout session that you did today; the scale up features of Windows 2012 Hyper-V, 64 Virtual CPUs per VM, 1 TB of RAM per VM. They compare similarly to vSphere 5.1, they do 64 virtual CPUs, 1 TB of memory.
BA: Match or better.
PS: Match or better?
BA: Where we support more physical processors.
PS: In the hosts?
BA: We do more logical processors in the host than they do.
PS: You do 320 and they do…
BA: They do 256.
PS: 256 – but the reality is that unless you’re a very big company and you have very special needs most people do not have hosts that big.
BA: I would agree with that but since you were at my session the big focus for us, there’s no workload that you can’t virtualize.
PS: Is it also a forward looking thing?
PS: Probably not a lot of people have the money for those sorts of servers, but four years down the track or five years down the track.
BA: Absolutely, and I mean today you go out and you get a nice 8 Physical Processor box today you’re looking at 96 threads, because at 6-Core hyper threaded, 96 logical processes, and it probably came with 256 GB of memory. Three years ago that would have been in fantasyland. That’s the – the rate of progress in hardware at the moment is just staggering, and the really fun thing for me is – yeah, you can go down to HP and it’s like an 8 Processor box is like yeah we’re getting on the high end but that’s not unreasonable. But even with that thing these days the only workload that’s going to make that box sweat is virtualization, that’s the thing that I just love. These days any hardware at scale – I can tell you the number one thing on the hardware vendor’s mind is how do I make this rock for virtualization.
PS: That makes sense. So you were talking about the gain scheduling in VMware and how you guys made Windows 2008 aware that it was being virtualized and thus could manage multiple virtual CPUs in an efficient fashion. Do you know whether VMware takes advantage of that?
BA: I’m actually not allowed to answer that question.
PS: Okay, because we’re talking about competition?
BA: I have to be very careful here, fact of life with Microsoft, VMware is both a competitor and a partner, and I do want to be really clear on this, obviously we compete with them in the virtualization space, but we also view them as a partner, they’re a platform that Windows runs on, and we have – we’ve published all the specification for how to have your hypervisor enable these features under the open specification promise that we have. We also have – are you familiar with the SVVP program?
BA: So we have VMware coverage under that, and interestingly enough and I won’t say more than this – I’m not allowed to talk about it, not because they’re a competitor but because they’re a partner, those conversations are covered under NDA.
PS: Okay that makes sense. How come you went from 4,000 VMs per cluster (the limit in Windows Server 2012 RC) to 8,000 at RTM, when did that happen?
BA: Because we tested it, honestly the vast majority of the time what stops us is just the physical challenge of testing at these scales. So for 4,000 to 8,000 it was just getting the hardware together, getting the test suites together, getting the testing done. I talked about this in my session– we don’t sign off on fictitious numbers, and I can tell you there are many parts of the system when we know in theory it could go higher but we need to make sure that we’ve tested it. This has been a common pattern for us – is that really between the beta and the RTM we’re just pushing our test limits as much as we can, and where we end up saying is supported is what we got tested by RTM.
PS: So we think that there might be in the future you guys doing more testing, and we might see figures change as we progress?
BA: Well I can’t talk about it but if you historically look at Hyper-V we’ve actually done that. We have in the past done a release and then six months later come out and say actually we support more than this. Our classic one would be virtual processor to logical processor ratio, where at Windows Server 2008 R2 RTM we said we support four to one; six months later we came out and said you know what, eight to one works. It was all just about testing.
PS: Okay, if you were designing a private cloud for a medium to large business would you make one big cluster or would you make several smaller clusters? Now that you have flexibility you’re not locked down to 16 nodes.
BA: You know that’s actually a really hard question….
PS: That’s why I’m asking you…
BA: I’ve talked to customers who have made the decision either way – let me talk about what are the factors that are going to influence your decision – the biggest factors to consider are how you’re going to manage your environment, what sort of infrastructure do you have in place? So we now support 64 node clusters, with 8,000 virtual machines, so reasonably you could put all your systems in a single cluster. Now you’re going to need some pretty phenomenal storage infrastructure to make that come together, and that would make sense in an organization where you have a single IT management team. The two biggest reasons why I see people moving towards multiple small clusters– actually three reasons, the first reason is because it lines up with the hardware acquisition process. The most common reason I hear for multiple clusters is it’s not a planned big project, but rather they’re requiring four or five servers at a time, and standing up the cluster and deploying it and then managing it. The second reason is they’re not willing to invest in the network and storage infrastructure to have a unified storage, and instead they’re going with more commodity solutions, and buildings, smaller deployments and stitching those together.
The third reason, which does come up, is you have warring political factions, and this department wants to manage their service, and that department wants to manage their service.
PS: That makes sense. Again I’ve got a few VMware questions in here so we’ll see how that goes. So now we have Hyper-V Replica, a feature that as I’m running my own business servicing other small businesses, is a fantastic feature I think, and it seems pretty complete out of the box. There’s compression, it’s there, encryption is there, and you can set it up, you don’t have to buy anything else. Whereas VMware has SRM, and they have just announced their VMware Replica, but it seems more like a technology rather than a complete solution, would that match what you see?
BA: I have to say it’s been – and I would love it if you quoted me on this, it’s been just fascinating to watch the vSphere 5.1 release, and anyone who’s not blind can look at that and go boy, VMware is trying to catch up to Microsoft, isn’t that lovely to see.
PS: That was my follow up question.
BA: It’s great to be in that position, it’s great to be leading and to be watching them trying to catch up. The announcements that they’ve made around SRM, and Replica, and changing their licensing, it’s obviously a response to what we’re doing. One of the things that I love about Replica as compared to SRM is the simplicity of it. To set up Replica, you need two Hyper-V servers and some form of connectivity between them. To set up SRM – my goodness, if you look at the architectural diagrams and you need dedicated SRM servers – all this, and when – and I’ll talk about this tomorrow in my Replica session, but when we started working on Replica we were looking at the disaster recovery space. One of the big stories that we heard, and one of the things we’ve really focused in on, was that in the DR market today there are all these great options but they all required significant investment and infrastructure, and so much so that small and medium businesses were looking at it and going - we won’t have a DR solution. These things are so far out of our ball park we don’t have a DR solution, and we really built Replica to target the small and medium business.
Part of that comes through in no hardware dependencies and so on, but the other part of it was we really wanted it to be simple because we really wanted this to be the two scenarios we really focused in on: We want medium business with an IT generalist to be able to go in and turn this on, and not be a big deal. The other thing that we wanted was we wanted this to be for the small businesses where they don’t have multiple sites, we wanted this to be something that regional hosters would go yeah, I can provide that as a service, it’s easy enough for me to set up, I can see the value in this and that’s what we designed it for.
PS: Yeah well I think Microsoft hit the nail on the head there, and that sort of leads into the overall scene I suppose: that for the first time that you and I are sitting here we’re looking at Hyper-V compared to VMware. I know there are other hypervisors on the market but Xen Server really doesn’t have the same market share as VMware. VMware certainly has been the leader, they were the pioneer in many ways when it comes to server virtualization, and this is the first time – with the release in 2012 this is the first time that you’re either at parity or you’re better than they are.
BA: Yeah, I’m so excited to be getting this release out.
PS: You must be feeling good.
BA: The story that I’ve been telling people is with 2008 R2 I would meet with the CEO of CIO who is using VMware and they would say Ben, why should I care about Hyper-V? I would say look VMware has some advanced features that we don’t have, but we have the basics covered and we’re a darn sight cheaper, and are you getting value. With the CEO of CIO – they’d think about that, they’d scratch their head and be like it’s a good point, we should do the analysis. With 2012 I’m sitting here going - we match or exceed and we’re still cheaper. The story I love– I did Tech Ed US–and we had a Hyper-V booth. I had a number of people come up to me, what they basically said – and I heard this a couple of times, it was the IT guy, the implementer, and he was saying we’re a VMware house, I like VMware, it’s what I’m used to, if there’s nothing wrong with VMware frankly I would be happy if we stayed with VMware. However I look at what you’re doing in 2012 and I know that if I don’t go and do the due diligence and look at this, that I’m going to get in trouble.
BA: I was just grinning.
PS: Well it makes sense doesn’t it? Would you say it’d be true to say that the advantage that Microsoft has over VMware when it comes to the whole private cloud virtualization space is that you guys make money out of all the layers because you make Windows, and 75% at least of all the VMs that are running are Windows. So whichever hypervisor you’re running it on you’re paying Microsoft for the license of Windows. You guys run most of the applications, Exchange, file servers, Share Point and everything else, you make money out of that because that’s what people run on top of. VMware however has to make money out of that one layer…
BA: Well partially, and let me put a different spin on this, and this – I think a lot of people miss this. The first simple statement, which should come as no surprise to anyone, is Microsoft is about Windows, it’s what we do, it’s our bread and butter, and it’s what we care about.
The thing you have to understand is Microsoft didn’t get into virtualization to compete with VMware, Microsoft didn’t get into virtualization to try and expand into a new market space. Microsoft got into virtualization because they said hey, Windows users want to virtualize.
Virtualization has never been viewed by Microsoft as a money source, that’s never been the strategy, the focus has always been we want Windows server to be the premier server platform, we want people who are deciding what they run on a server to choose Windows. That’s very much the mentality. One thing I find fascinating is how virtualization has changed the way we sell a Windows server because really for the first time when you look at server 2012 we have the free Microsoft Hyper-V server, we have Windows Server Standard, we have Windows Server Data Center, all with the same features, there is no feature differentiation between any of those. The only difference is how many virtual machine licenses and how many guest OS licenses you get with that.
It’s funny, just about a week ago I saw an e-mail from the lead PM for clustering, which blew my mind because there are times when I fall back into the old server admin mentality, but I saw a comment – and because they were discussing licensing and they were basically saying look, the simple way to understand Window server 2012 licensing: is it a Hyper-V host, you want Data Center, if there’s anything else buy Standard. The lead PM for clustering said that high end SQL cluster, if it’s running on hardware, use Windows Server Standard.
PS: Because clustering is built into Standard, and that’s always been the one feature that has been different between Windows Server versions. Where do you see the scenario for being able to run VMs from SMB shares, VMware does this with NFS, how do you see that?
BA: I got to do a storage session at New Zealand and it didn’t make the cut here, and I wish that it did. I do customer briefings all the time and I have this little sound bite that I like to say about storage. Whether you’re like a mum and pop shop with a server under the stairs, or whether you’re like a Fortune 100 with huge data centers, there are two things, two facts, that hold true about storage no matter what size. The first fact is whatever your storage needs there is a perfect solution out there for you.
The second fact is it’s more than you’re willing to pay. No matter what your scale point is it’s always more than you’re willing to pay, and this is something that we’re hearing over and over again. So there’s been a really huge investment in storage in server 2012, the investment kind of goes along two angles. The first angle is we want to open up more cost effective storage options to people. We know that there are people who have been going out and buying iSCSI SANs for small workloads running on Hyper-V. By supporting SMB they can now just use another Windows box and that just became a much cheaper deployment for them.
On the other hand though we’ve done a lot of investments in stuff like ODX, Trim, Unmap, virtual Fibre Channel, so on the high end which are focused around, if you spent money on that SAN let us help you get more value out of it, and that’s really being – as I said the two focuses, opening up more cost effective options but across the board making sure that whatever you’re using – let’s get more value out of that storage.
PS: Okay, so server 2012 can be an iSCSI target, you’ve built the free download that you used to have into the platform, but a lot of the functionality of merging your SAN from Windows server 2012 comes from supporting SMI-S. Does the iSCSI target in server support SMI-S?
BA: Honestly I don’t know…
PS: Well, I don’t think it does. I haven’t seen it – doing that anywhere, I haven’t seen that come up, and it’s also when you talk about System Centre Virtual Machine Manager – it also wants SMI-S.
BA: I am worried about saying this to a reporter because if I get quoted on this I know of at least one PM who will hunt me down at Microsoft, but I will say it none the less. Honestly I’ve yet to figure out with Window server 2012 why you would chose iSCSI over SMB3. I will say – so Microsoft PMs, we tend to be very territorial, we believe our feature is the best. I have been getting a lot of flak from my peers on the Hyper-V TM because I am constantly singing the praises of SMB 3.0; it is a phenomenal –it is revolutionary. So I did the Hyper-V storage session in New Zealand and one of the things I said to the people – I’m the Hyper-V guy but if the only thing I achieve today is that you go away and look at SMB 3.0 seriously I will be happy, because it’s got multi- channel, hardware offload, encryption, the transparent failover in the cluster, that’s amazing. There’s just so much stuff in there it is such a huge step forward.
PS: And you say that because you can easily build a scale out file cluster and then you have the high availability basically in all the layers.
BA: Absolutely, and you know the SMB 3.0 is completely enterprise grade, I trust it to the hilt for being able to run in virtual machines. We’ve spent a lot of time working with the SMB team and they’ve delivered so much functionality. There’s stuff that I can point to in SMB 3.0 that outclasses iSCSI. As I said I haven’t figured it out, like I get if you have an iSCSI target, great, we still support it, we’ve done work on iSCSI. But if you had a Windows server on the back end and you were just running Windows on the front end, I can’t figure out why you would choose iSCSI over SMB.
PS: The other thing is that if you want to have a HA SAN, an iSCSI SAN that is highly available, you need two of them and they come with a pretty big price tag.
PS: Whereas you can spin up several file servers and…
BA: Very cheap…
PS: With a much more…
BA: Actually not supposed to have used the word cheap, don’t say I used the…
PS: Cost effective?
BA: Yes, I’ve gotten this far without using the word cheap, the marketing people always get mad at me when I use the word cheap.
PS: Now network virtualization is an awesome new feature as well in Hyper-V, which has a great future obviously, and it’s…
BA: Networking is just an amazing area in this release.
PS: Yeah, and it seems to me that software defined networking is just now coming up on the radar, and just now I saw VMware doing an acquisition in that space whereas you guys seem to have delivered it now, right there built into Windows server.
BA: Yes it is. So a couple of things, one; yeah we’re really leading the way here, it’s going to take them a while to catch up with us. Two – and I’m just so happy about this because for the last 12 months I’ve been predicting this and it’s come true. It’s never fun to be in the catching up position, it’s never fun to do the release where you’re like hey, you know that thing our competitor has, we now have it – fine. What I do like however is when I can say okay, we may be catching up but let me show you how we did it better. With the extensible virtual switch we did it better, and where that’s already showing up is VMware has had their virtual switch stuff out there for ages, the day we released Windows Server 2012 we had twice as many partners. That number is just going to take off, and I can tell you six months from now we’re going to have ten times as many partners as they do. Well the fact of the matter is – and I’ve had VMware people talk about this, VMware has two partners, I can’t remember who one of them is, the main one is Cisco. I’ve talked to VMware people and I’m like hey are you using VSwitch, no I don’t like Cisco, and if they don’t like Cisco they can’t use VSwitch.
We already have Cisco signed up but we have so many partners, and the big difference here is – and this is – there are many different strengths that Microsoft has but something that Microsoft does well is knowing how to build an ecosystem.
PS: I was just going to say that.
BA: Yeah, that’s one of our real core skills. I don’t know if this is what happened but if I look from the outside and I crystal ball gaze, I look at what VMware getting it kind of looks like Cisco approached them and they built something for VMware. Whereas we spent a lot of time talking to the whole industry making sure we built something that the whole industry could use. Let me give you the quick technical detail, so the VMware V-switch solution – what that basically said is hey if you want to do deep network integration you can throw out our virtual switch and build everything yourself. That’s a huge burden, and frankly only a company like Cisco is going to step up for that. We talked to all the various partners and basically said: hey what do you want to build; they came back and they told us that they wanted to build on top of an existing solution. We said okay, we’ll take our virtual switch, and our virtual switch is always going to be the virtual switch but we’ll put in extension points where people can plug into our switch and they can do things like intrusion detection, they can do things like firewalls. Now the really neat trick that we did was as much as possible – obviously we couldn’t do this 100%, but as much as possible we actually made the plug in interfaces for our virtual switch similar to the Windows NDIS interfaces.
So if you’re a vendor and you’ve written network management for Windows already it’s really very easy for you to plug this into the virtual switch and provide this as a service on Hyper-V and that’s what we’re seeing. So that’s an example where I go yes, we may have been playing catch up but we did it right, and I can tell you in six months’ time you’re going to be looking at our partner ecosystem versus their partner ecosystem, the difference is going to be huge.
PS: Well it has some benefits to be second as well because you can learn from the mistakes from the ones who came before you.
PS: You got this question in your session this morning so I suspect I know what your answer is going to be but if you look across the entire solutions– this is what VMware offers today, this is what Hyper-V offers today, and in server 2012, the one standout feature that is missing is Fault Tolerance (FT). Having two VMs that are literally mirrored, and the idea is to protect against the failure of the underlying hardware, and have the machine keep on running rather than rebooting on another host in the cluster, you guys have nothing like that today and you won’t tell me whether you’re going to be working on something, but.
BA: I won’t tell you whether we’re going to be working on something like that. I will – and I’m going to be very careful in my answer because there are really two sides to this answer. The first side to this answer is I don’t want to be one of those jerks – and I mean this is the classic Microsoft thing, I don’t want to spend the next couple of years going FT sucks, you never want to use FT, and then we turn around and release something like it and go like yes it’s the best thing since sliced bread, I hate it when people do that.
So upfront I’m going to acknowledge – yes there is value provided there, and certainly it’s – you can easily imagine that it’s something that we’ve discussed at length and whenever we’re working on planning for new releases it’s something that’s on the list to be considered. That said let me step you through some of the reasons why – for instance why we didn’t do it in this release, and why we chose the features that we chose instead.
The first thing is that VMware’s FT Technology as it exists today – and I fully acknowledge that this could change, but FT Technology as it exists today we believe comes with too many limitations.
You can only do it for Uniprocessor virtual machines, that’s a big one, I’m sorry, I said this in my architecture, I’m old school server admin, you could not pay me enough money to run a uniprocessor server because that just needs one hung user mode process and your server is toast. It comes with a big latency impact; it really slows down the latency. It actually provides no protection against any sort of software failure, and the hardware protection is even a little bit questionable itself.
As I said we’ve spent a lot of time looking at this and there’s this interesting scenario to look at with FT because if you have an FT solution and your box goes down fine, it fails over. But what if your box doesn’t go down, what if a bit in memory goes bad? Well if a bit in memory goes bad then the paths are going to diverge. Now FT doesn’t know which one is good, it’s going to assume that the primary just went bad and fell over to the secondary but if something went non-fatally wrong on the secondary you’ve actually just lowered your overall availability.
The final caveat about FT is you need fantastic data interconnection to make it all work. So we looked at all of these things and basically went you know what, all those things combined, this is niche, let’s go focus on the big picture. Now I will say kudos to VMware, I love having them as a competitor. Something that I have seen them do multiple times, and I think FT is a great example of this, is they build cool technology, they don’t always understand why the world wants this but they build cool technology. FT is cool technology but – and honestly I get people asking me about FT and my number one response is - are you using it?
PS: Well statistically I would also add if you look ten years ago the main reason that servers failed were hardware problems.
PS: I think today if you look statistically at server hardware today we have learnt a lot in the last ten years of how to build servers. The main reason that servers fail today is not hardware, even if I look at my small business clients…
BA: Honestly the main reason why servers fail today is unfortunately buggy drivers; Microsoft has come out and said that multiple times.
PS: Yes and the thing is that if the server crashes because there’s a buggy driver here it’s going to crash over here in FT as well, so that’s not going to help you.
BA: A side observation about that, and this is actually something that we’ve seen, is because of the leading cause of server crashes being buggy drivers we actually see better uptime in virtual machines because there are fewer third party drivers.
PS: That makes sense. What about the free Hyper-V server 2012, where do you see that being used, in what scenarios does it fit?
BA: This has actually been really interesting, I’ll be honest, and I’ve been surprised that the marketing team has been as gung ho on this as they have been. Hyper-V server 2012, the free edition, has always been kind of a dark horse in the Microsoft world, but it’s something that we’ve viewed as important. With Server 2012 we’ve come out and said, it comes with the full feature set, the only difference is guest license. The first one that – the one that we’ve always said, which has never really made that much sense, is like well if you’re going to run lots of Linux VMs you could use this, and it’s like let’s be honest, if you’re going to run lots of Linux are you really going to use Hyper-V server – no, you’re probably going to go use KVM.
But the other scenario, which we’ve been actively pushing, is if you’re doing VDI go use the free Hyper-V server because the guest OS licensing that comes with standard and data centre is for server instances. If you’re doing VDI we have the VDI licensing and it doesn’t matter which Server OS you’re using. Once again what’s Microsoft about – we’re about Windows, so the license, the copy of Windows through the VDI licensing and use the free Hyper-V server, it’s a great solution.
PS: And again VMware limits their free offering to 32 gigs of ram.
PS: And they didn’t change that in 5.1…
BA: Actually they did change it because – and you should fact check this but I believe this is the case, in vSphere 4.X it was limited to 32 GB of RAM, in vSphere 5.0 they dropped it to 8 GB of RAM, and you can’t do anything with 8 GB, and in 5.1 they put it back up to 32 GB.
PS: Wow, how does your implementation of SR-IOV that you described in the session today – how does it differ to VMware’s?
BA: Honestly they only just announced the support for SR-IOV at the VM World that just happened. So I don’t know all the gory details, the one thing that I do know is we’re able to do live migration. I stepped through that but it’s something that we’re really adamant about. The first one is we want all our features to work together, we hate that game of like you can use feature X or feature Y but you can’t use them both together, we do a lot of work for that. The second one is – I’ve said this - a number of times today and in different venues, I don’t like working on features that no one uses. The fact of the matter is with private cloud, virtual machine mobility, live migration is so crucial that if you developed a feature and said but it doesn’t work with live migration, great – no one is going to use it. So we’ve worked really hard and we do SR-IOV with live migration which VMware doesn’t do, even in 5.1. With our virtual Fibre Channel we support that with live migration. VMware has Fibre Channel virtualization, but they don’t support it with live migration.
PS: I didn’t know that. You speak about Hyper-V and VMware with more confidence now than you said did two years ago, I think this might be because you guys did what Jeff Woolsey was describing for Windows server 2012, you went out and talked to customers. Microsoft has always been reasonably good at listening to customers – well if you look ten years ago it was a bit more like the Borg but nowadays, in the last five years you’ve been pretty good at listening to customers, but it really seems to me with server 2012 that you’ve taken it even one step further.
PS: And I think that’s good.
BA: It’s actually been really fun, we’ve – personally for the Hyper-V team we’ve had great engagement with our TAP customers. We have a phenomenal group of Hyper-V MVPs, I love our MVP community and they’re a great resource for us. But also what’s been really fun, and this probably bubbled up a bit in the session that I just did, is more so than any other release we’ve done a huge amount of usability testing with IT admins in this release, and it’s been fascinating. The number of things that – like the whole move VM wizard we have for moving VMs, we built that once and then we’ve rebuilt it from scratch because we built it the first time and we took it through a useability study and no one succeeded. It was embarrassing, like I remember the first wizard we had would ask people to do a shared nothing live migration and no one succeeded, it was horrid, it was the most painful thing to watch. So we went back and we did a whole round of review and investment, and we did the same thing with Hyper-V Replica, we did a lot of focus on – and this is one of the – it’s been a lot of learning for us, and just learning how people actually use the software, and how they react.
PS: So is this the traditional usability studies, like you put them in a room and you’re behind a one way glass mirror, and you’re watching them do the stuff, they’ve got a list of things to do and they’re just not clicking on the button you want them to click on.
BA: It is so painful to be behind the window and to just be watching –like when they’re heading down the wrong path– but I think it’s an important experience for anyone who designs software to go through because it really shows you– it’s a humbling experience as well.
PS: So when you go out to customers you often talk about multi hypervisors that people are using, Xen server they’re using, red hat KVM, they’re using VMware, and you guys come in with System Centre Virtual Machine Manager 2012 which manages the three main ones. Does anybody actually use System Centre Virtual Machine Manager like that, or is it just a feature that you have so that you can tick the box and say Virtual Machine Manager supports all three?
BA: Let me give you a two part answer to that, the first one is I have to acknowledge that we are now in a multi hypervisor world, and I find that fascinating. Three or four years ago I would meet with customers and they would very clearly say - we chose VMware, you’re lucked out, and three or four years ago people very much had this mentality of I’m going to choose my platform and that’s it.
These days it is rare for me to come across someone who is just using one virtualization platform, and I’ve even come across customers who have like three virtualization platforms in there– I find it fascinating. The thing that I’m seeing is what’s happening is people are virtualizing the workload, they’re having a good experience with that, and they’re not going to go away and rip out what they have but when they have their next project, when they want too virtualize the next workload, they are 100% open to going yeah we have to go and do the due diligence and make sure that we figure out what is the best solution for this scenario.
The second part is do I know of anyone using SCVMM to manage all hypervisors in their environment – I do know of a handful of businesses, I wouldn’t say it’s common. What I would say is that multi-hypervisor management is often the foot in the door, for people who have an investment in VMware being able to tell them look, you don’t have to throw that all away, this stuff can work side by side. It’s the foot in the door to make them go OK -we’ll have a look at it. The reality is that – from what I’ve seen most of the time –if they’re doing VMware they go to vCentre and if they’re doing Microsoft stuff they go to SCVMM. The big thing that people really appreciate – the fact that the System Centre can give them an all up view of their entire data center – maybe when they’re doing maintenance they’re going and using vCentre, but the fact that SCVMM, SCOM, Config Manager – they get the all-up view; they have a single place that they can go to for just monitoring the environment.
PS: Hyper-V ties in with VDI, Server 2008 had an inbox VDI solution that was incredibly complex to set up, and get configured, and working. I know that you have improved it a lot in server 2012 but people still go with VMware View, or Citrix etc. or other solutions, is it just the setup experience that is better in Server 2012 now, or are there other things that make VDI based on just a Microsoft solution viable?
BA: I have to say my hats off to the remote desktop team who own our VDI story; they’ve done some really interesting work around VDI. They’ve made the setup experience a lot better, but one of the big points of feedback that we’ve been hearing from people about VDI is – like even more so than with server virtualization, the biggest problem with VDI becomes storage.
This is the rude surprise that causes your VDI deployment to be a lot more costly than you would expect, when put all desktops in a single box and you scope out the storage, you’re going out and paying top dollar for some really high end storage. So the VDI team has actually spent a lot of time focusing on what can we do to make this work with cheaper storage. I used the c-word again - sorry!
There are actually two things that have come out in Server 2012, the first one is if you’re using clustering with VDI we actually now have a read cache built into CSV that is specifically there to address things like the VDI boot storm. I don’t have exact numbers to share right now but the result is pretty phenomenal, if you get a bunch of desktops booting they’re hitting the same blocks, we’re able to cache them and we’re able to get some really fantastic performance there.
The other one that I’ve found fascinating is the VDI team have actually come up with this deployment model that’s kind of a reverse to what people think about. A common concept for VDI is I’m going to use a gold image for VDI, and then I’m going to have a separate VHD that contains the user data, and that way when I need to service the gold image I update the gold image and life is good. People have been doing that for a while but the traditional model that people had used for that is the gold image and the user disc was sitting on a piece of central storage, and that central storage was still at the bottleneck. Well what we’ve done in Server 2012 is we’ve enabled this model where we’ll keep the user disc on a centralized SMB share but we actually push the gold disc down to the local disc on each VDI node.
BA: So now you’re able to use cheap DAS storage, it’s not a huge bottleneck because it’s only used by the virtual machines on that box, and the relatively small user disc is on the central share. That way you’re actually able to use DAS at scale in VDI, that’s a really cool solution.
PS: My final question, with Hyper-V – you don’t want to call it version three but...
BA: Yes, that is a term from the press actually; it’s Hyper-V in Windows Server 2012.
PS: Hyper-V in Windows Server 2012, you basically fired a big broad side at VMware, said OK we can do it as well as you can, we are cheaper, but people don’t throw out investments...
PS: And certainly not at VMware investment, that’s a big price tag. So how is that going to work long term, where do you guys see – because obviously for you personally, or for the Hyper-V team, you would like to have more than a 50% market share rather than the less you’ve got now. How are you going to achieve that, or how is Microsoft going to achieve that?
BA: Keeping in mind I do a lot of meetings with customers who are VMware customers, and they ask me – hey we’ve got this investment with VMware, and I’m reasonable and realistic about this, yes there are scenarios – and we have great case studies out there about customers who have torn out VMware and gone with Hyper-V, and seen various savings through that, fantastic. But that’s not going to be true for everyone.
The two things that I say to people – look, if you’ve got an investment in VMware, and you’re happy with that, and it’s working, great, that’s sunk cost, and for a lot of people that’s how they operate their servers. Once the server is deployed it’s done, I don’t go back and touch it. But what I say to them is one; on your next virtualization project don’t assume VMware, go look at Hyper-V, I’m highly confident that we’re going to win them.
Two – I’m happy about this; VMware does three year licensing, I have lost track of the number of times that I have talked to CEOs who are in the situation where they’re support contract for VMware is coming up for a renewal, that’s an ongoing cost, they already have an Enterprise Agreement (EA) with us because they’re running Windows. They’re looking at that and going should I renew this, and that’s where I say hey, if you have a contract with VMware coming up for a renewal, you already have an EA, look at Hyper-V, do the math. I’ll be very surprised if you don’t find that you can save money when moving to Hyper-V. You can quote me on that too.
PS: This interview should end up on the 4sysops blog, I don’t know whether you’ve seen that?
BA: Yeah, I subscribe to that blog.
PS: It’s got something like 400,000 hits a month (Michael's note: about 500,000 page views per month now), they’re pretty popular I think, so your words will be seen by more than the 12 people in the world who read my blog.
BA: Yeah no worries.
BA: As I said we’re thrilled to get this release out, it’s a huge release for us, but I will be clear, we’re not done yet, no – more cool stuff will come in the future I can you guarantee you of that.
PS: Thank you so much Ben.