Microsoft's view on free Hyper-V 2012 vs. free ESXi 5.0

[kewnev]

Don't forget the other feature that Hyper-V supports that (correct me if I'm wrong) VMWare doesn't : Dynamic Memory.

[J1mbo]

Sounds more like it's translating them (virtual block to physical block). But if it is that simple, it seems amazing VMware aren't doing the same.

Also there is remoteFX - not available with terminal servers running on VMware, IIRC.

[dellock6]

Again, I do not want to be the VMware's defender, they do not need me, but usually MS announces new releases and their features way before they are available on the market,while VMware (and also Veeam by the way) keeps everything well hidden until the new release is out.
So we cannot say what it's going to come out from VMware in their next release, and if it's going to be out even before MS Server 2012. I'm sure anyway they will not stay still looking at MS...
Luca.

[J1mbo]

TBH the very fact we're having this discussion speaks volumes in itself.

[dkvello]

OK, There's no such thing as a free lunch. And, MS is offering You a full night out, a 7 course gourmet meal with good wine and drinks, together with a very hot date... for free.
That's enough to make me a skeptic.

"Hyper-V also support "vMotion" (live migration) of VMs between 2 host without using a shared SAN storage!"
Now, the plain physics of this will make it more or less useless as anything other than a "tick-box" feature. Bumping around a VM with one 10-20 GB vDisk could be doable, much less a 1TB file-server/DB server etc.

If TRIM/UNMAP, the way it's explained here, had been implemented in any current OS, i'm pretty sure VMware would already have supported it. But, alas, this demands a whole new file-system (W2K12 ReFS) and is a feature of the OS/FS and not of HyperV itself. We wouldn't need sdelete if NTFS allready could do this by itself.

When this is a feature available in W2K12+ReFS, then it will work with vSphere 5.1 (Ican only imagine )
Anyways, I use SVC/Storwize V7000, and the new Compressed Volumes (as opposed to only thin-prov) will handle this itself, wether VMware can support this from the start or not. No negative performance impact whatsover.

[J1mbo]

It's not free. Sure Hyper-V server is free, but that's it. SCOM is priced per CPU, but it's a lot less than vSphere by virtue of the fact that MS are the underdog and secondly because VMware have gotten greedy (IMO).

[tkrajewski]

IMHO "share nothing Live Migration" will be killer in SMB space...

Tomasz

[J1mbo]

There's lots of videos worth watching, like this one on replication.

[dellock6]

Sure? sorry I can't imagine people keep moving VM around only for the fun of it..
Is there really a use case for this, other than "manual" migration or "scheduled" server disposal? I should prefer replicating via Veeam one server with local storage towards a twin of it, as I already do today with really small environments scheduled, automatic, failover-proof. Otherwise, sorry I will go with a shared storage, or a VSA appliance that can replicate data between hosts in real time.

Killer would be "scheduled" replication between hosts without shared storage....oh wait, that's what I can already do with Veeam

[Gostev]

"Hyper-V also support "vMotion" (live migration) of VMs between 2 host without using a shared SAN storage!"
Now, the plain physics of this will make it more or less useless as anything other than a "tick-box" feature. Bumping around a VM with one 10-20 GB vDisk could be doable, much less a 1TB file-server/DB server etc.

I disagree. If that is the case, VMware Storage VMotion is also a "tick-box" feature? And it also works just fine without any shared storage, just over the network, and on VM of any size. Well, I would not say it is a "tick-box". I've seen live polls at a few VMworlds in past, and it is consistently voted as one of the most useful VMware features.

If TRIM/UNMAP, the way it's explained here, had been implemented in any current OS, i'm pretty sure VMware would already have supported it. But, alas, this demands a whole new file-system (W2K12 ReFS) and is a feature of the OS/FS and not of HyperV itself. We wouldn't need sdelete if NTFS allready could do this by itself.

That is not correct. You clearly missed today's weekly forum digest, because I have already explained all of this there
1. TRIM is already implemented in all modern operating and file systems. So, this feature does not require ReFS to work. My hope PC fully leverages TRIM!
2. This is a feature of Hyper-V itself. Hyper-V intercepts TRIM commands within virtualization storage I/O stack, and makes a good use of it.

By the way, if I were Microsoft, I would patent this feature, tsk-tsk-tsk... pretty creative implementation anyway, so far I wow'ed everyone I explained it to. Microsoft folks get full credit here, you definitely gotta be very out-of-box thinker to see how this very SSD-specific feature can be repurposed to solve one of the biggest virtualization storage problems!

[J1mbo]

And if I were Microsoft, I would patent this feature, tsk-tsk-tsk... pretty creative implementation anyway, so far I wow'ed everyone I explained it to Microsoft folks get full credit here, you definitely gotta be very out-of-box thinker to see how this very SSD-specific feature can be repurposed to solve one of the biggest virtualization storage problems!

I dunno about that, it's been painfully obvious that TRIM would work for this, given that the crux of the issue has been from the start that SCSI didn't have a delete command (since it was designed in the 1980s for 8-bit spinning drives).

[Gostev]

Yep, sounds so very obvious for me too... now that I learned how it works but all that matters for patent is whether prior art exists. And I don't believe any hypervisor has this feature today - or even in the works.

[Gostev]

It's not free. Sure Hyper-V server is free, but that's it. SCOM is priced per CPU, but it's a lot less than vSphere by virtue of the fact that MS are the underdog and secondly because VMware have gotten greedy (IMO).

And you don't need SCOM to run Hyper-V anyway, unless you are big shop. As some people already mentioned, smaller customer may as well just use Veeam ONE for management. I must say, I really like that suggestion!

[tkrajewski]

Sure? sorry I can't imagine people keep moving VM around only for the fun of it..

Im my neck of the woods I see demand for "biological" DRS so to speak. Small shops, 2-3 servers/hosts max, manually locating VMs on hosts, some trial/error in it (less of it with Veeam ONE ), patching parent partition, etc. Moving VMs will become natural for those admins of small environments, they don't have to think about storage connections, sharing it (althogh it would become also trivial with SMB shares). From my experience small shops don't schedule and are not proactive. They are reactive and fight fire when they see one. And those features are mainly aimed at such environments. At least that's how I see it...

Tomasz

[averylarry]

Sure? sorry I can't imagine people keep moving VM around only for the fun of it..
Is there really a use case for this, other than "manual" migration or "scheduled" server disposal? I should prefer replicating via Veeam one server with local storage towards a twin of it, as I already do today with really small environments scheduled, automatic, failover-proof. Otherwise, sorry I will go with a shared storage, or a VSA appliance that can replicate data between hosts in real time.

Killer would be "scheduled" replication between hosts without shared storage....oh wait, that's what I can already do with Veeam

I wish I had storage vMotion. As a small business I buy small business hardware, which means I can't just load up on storage. So I often get to a point where I buy new storage (instead of expanding existing storage). In that case, I want to move my data from storage A to storage B.

Unfortunately, in the small business world, I still have terabytes of data and I can't have hours to days of downtime just to implement new storage. With storage vMotion, I don't particularly care if it takes days -- I can still have my servers up and running. Somtimes I end up doing an OS based migration using robocopy instead.

As a small business, I use the free ESXi as well as an Essentials Plus bundle. I'm pretty good for many years probably on the vRam. But I worry about the 2 processor, 6 core limitation. The price for features I'm interested in as a small business makes me think I might consider Hyper-V. My biggest fear is that Microsoft hasn't matured their Hyper-V product enough yet.

[alexrogi]

Will this make you return to the drawing board and rethink your hypervisor strategy? I'd love to hear your opinions.

Yes, we already move to Hyper V 2008R2. We have been running HYPER V clusters since December with no problems at all.

We were using ESXi 4.1 as hosts to VM, but we decided to move from VMware to Hyper V, since ESXi changed the licenses for the memory tax.

Using ESXi 5 (free) with a limit of 32 gb is simply not acceptable and the money to do the upgrade to a full ESXi was too much.

We still have ESXi 4.1 hosts, but we are moving VM's to Hyper V slowly. It's a great product, but too expensive.

Now that we are seeing the new features of Win 2012 Hyper V, confirmed that we did the right choice:

Cluster Shared Storage
Concurrent Live Migration
CSV 2.0
CSV Block Cache
Dedup
Guest Application Monitoring
Guest Aware NUMA
Hyper-V Replica
Live Storage Migration
SMB 3.0

All this by free. See more on:
http://www.aidanfinn.com/?p=11979

[averylarry]

My other worry is that Microsoft will eventually start charging -- once they have you invested.

[J1mbo]

Part of it for Microsoft is that making Hyper-V free means they sell more Windows licenses, or at least more higher-value Windows licenses. Their customers can try out new features more easily by simply downloading demo VMs etc. Whereas for VMware they don't have that income option.

[J1mbo]

BTW re quick migration and the SME - one pitfall is licensing at the application layer - classic example being SQL Server. I hope the licensing is updated (to be fair, again!) with the next SQL server as the current product requires all CPUs in a cluster to be licensed if using vMotion or its equivalents.

[alexrogi]

My other worry is that Microsoft will eventually start charging -- once they have you invested.

Yeah, that can happen..

Or maybe, just MAYBE, VMware will wake up and make ESXi 5.0 free and unlimited again...

[Gostev]

My other worry is that Microsoft will eventually start charging -- once they have you invested.

I think they should be fine for a while with making money on selling those operating systems by the way, for some reason nobody is talking about the Datacenter Edition of Windows Server, which seems to be yet another creative attack on VMware (due to providing unlimited OS licenses for guests). From the numbers I remember, I believe that it is cheaper to go with Datacenter Edition as long as you have more than 6 VMs per host (and you will always have more than that these days, of course). That is, cheaper than to go with free Hyper-V, and buy Windows OS licenses separately.

My biggest fear is that Microsoft hasn't matured their Hyper-V product enough yet.

Ditto, that's my only worry too. Will see soon enough how well those 4 years of development were spent!

Or maybe, just MAYBE, VMware will wake up and make ESXi 5.0 free and unlimited again...

But then, what do they sell? They are in really, really tough situation right now - that is for sure.

[bhwong]

From above, "Moreover, this does not even require Hyper-V integration services installed on the guest - all you need is OS that implements TRIM (fairly common in these days of SSD invasion)."

Does TRIM means when Windows delete a file, it basically zero-filled the file similar to sdelete so that the storage is returned to SAN automatically?

[J1mbo]

Does TRIM means when Windows delete a file, it basically zero-filled the file similar to sdelete so that the storage is returned to SAN automatically?

Kind-of. SCSI is a protocol designed for spinning disks, so there isn't any concept of block delete since the blocks all exist all the time physically, what's on them is irrelevant. But for flash-based storage, there is a used block write penalty since blocks have to be erased before they are written (erasing a byte is to value $FF incidentally). Anyway the new(ish) command TRIM allows the OS to communicate with the device that the block is now free hence in the background the device can erase the block, so that when a future write happens there erase part of the write is already done. For SSDs it also means the block is a candidate for wear-levelling algorithms, presumably.

Anyway with that in mind it's obvious that this is of use for iSCSI volumes that are thin-provisioned at the storage end since it communicates exactly what we've been missing - that the block is now free. So the storage can then do whatever it needs to with that information. Storage systems tend to work with quite big chunks of data for thin-provisioning, but once a consecutive lump has become free then the storage can (again, presumably) pop that back in the unused pool, hence naturally re-thinning the volume as we go

Re datacentre licensing, that's a great point although the customer needs to consider whether to go with disposable OEM stickered licenses (which can't have SA so no upgrade rights) or buy them through volume licensing. But either way it's a great value proposition for anything over a tiny environment.

[smile_dav]

I told you back in March that Windows Hyper-V will change people's thoughts on VMware...
It's nice that VMware finally has some competition!

[tkrajewski]

And considering total revamp of "Get The Facts" page on VMware website they are watching things closely... There is even vSphere 5 vs Hyper-V 3 beta comparision doc here. Of course it's "what we have that they don't have" and doesn't mention "what they have that we don't have" of course... But I agree - competition is always best for customers.

Tomasz

[dkvello]

"Hyper-V also support "vMotion" (live migration) of VMs between 2 host without using a shared SAN storage!"
Now, the plain physics of this will make it more or less useless as anything other than a "tick-box" feature. Bumping around a VM with one 10-20 GB vDisk could be doable, much less a 1TB file-server/DB server etc.

I disagree. If that is the case, VMware Storage VMotion is also a "tick-box" feature? And it also works just fine without any shared storage, just over the network, and on VM of any size. Well, I would not say it is a "tick-box". I've seen live polls at a few VMworlds in past, and it is consistently voted as one of the most useful VMware features.

"Works" doesnt mean usable and Storage vMotion is a whole different story ftha vMotion. Networked Storage vMotion is an absolute necessity if You're going to do vMotion without a SAN.

I do my fair share of Storage vMotion, and even with very fast FC storage systems (V7000 with SSD/Easytier) VAAI etc. Moving data of any significant amount takes time. I moved a whole bunch of terabyte vDisks just his weekend, and it takes plenty of hours, even when it was HW-offloaded and between very fast FC-subsystems.

Now, shared nothing vMotion would mean that You do both a vMotion AND a Storage vMotion concurrently over the network. You'd need som very capable 10GBe network switches and fast storage on both ends to make this a tolerable alternative for shared storeage.
Getting vSMBs to buy $10000 worth of 10GB switches + good 10GB NICs + double the disk-space just to get very slow shared nothing vMotion... Wouldn't it be easier and more efficient to just buy a decent SAN/VSA instead ?

As said, it's a nice tick-box feature if You can afford a "vMotion" to spend 10-12 hours migrating that big fileserver from one host to another. As a pure migration feature its OK, but id' rather use Veeam for that.
It certainly wouldn't move any vSMB customers know to HyperV. And larger customers (larger than 5 employees) that I work with has shared storage of some kind. Heck, I'd rather they used some cheap NFS NAS box.

[J1mbo]

Getting vSMBs to buy $10000 worth of 10GB switches + good 10GB NICs + double the disk-space just to get very slow shared nothing vMotion... Wouldn't it be easier and more efficient to just buy a decent SAN/VSA instead ?

Probably worth noting that Hyper-V isn't even released yet and in it's product life cycle 10GbE switches will get much cheaper. I recently bought some 24-port GbE switches - non-blocking, managed and fanless - for £110. That would have been unimaginable only a few years back.

I'd rather they used some cheap NFS NAS box.

NFS shouldn't be viewed as just for cheap storage - it's a great fit for vmware. But if there requirement really is cheap, Debian is all you need and a basic branded server and some care in configuration. Of course you don't get controller redundancy, but cost per GB can be very low (and performance pretty good).

[dkvello]

I mentioned cheap NFS boxes as opposed to expensive NFS ones since this is about "Free" Hypervisors. I'd guess that customers going for Free ESXi or HyperV aren't interested in spending that much money on the other components either

I can get "cheap" 10GBe swithes her in Norway, but those are Fiber-based without necessary SFP's. A 24 Port 10GBe Dell Powerconnect 8024 (Broadcom OEM) is the cheapest Cat6e switch around here at the moment and has a list price of £15000 each. A current bid from Dell gave us two for one (two PC 8024 for £15000).

If Your'e talking simple 1GBe Switches with a couple of 10GBe uplink-ports, then thats a different matter and £110 is possible even here. They don't have much internal bandwith though, and doing a svMotion between two hosts would drain the life out of backbone.

As You said, prices are comming down, but that goes for shared storage as well.

[bhwong]

A fast 10GB switches would not make much of a different if the bottleneck for storage isn't fast enough to utilize the bandwidth right? We can always use NIC Teaming for 1GB switches to improve performance while saving on the price of switches?

btw, what is the preferred RAIDS you guys setup for your SAN storage? Do you prefer performance over protection?

[Gostev]

Correct, Windows Server 2012 supports NIC teaming with dissimilar NICs natively - yet another cool new feature, just not Hyper-V specific (but nevertheless, very useful for Hyper-V of course).