VMware started the Why Choose VMware portion of our website in 2nd half 2008 as more and more vendors were coming on the scene, all claiming to offer products that do what VMware’s solutions do. We felt it was necessary to tell our story, and to back up our claims with complete, academic evaluations of competing products. As such, The Why Choose VMware site shares six key reasons why we see VMware as offering a better solution compared to what others market. There’s quite a lot of content there, but we tried to keep it as factually oriented as possible. For instance, the product comparison tables are lab-validated, based on our technical evaluations and comparisons of the products; they are not derived from just a cursory glance at vendor marketing literature.
Then this past week, we started getting inquiries about a Microsoft video that purports to bust the top ten “myths” on Why Choose VMware. Others in the blogosphere have already responded.
- Gabe’s Virtual World: When myth busting goes wrong and becomes a myth to bust it self
- vinternals: Microsoft Hyper-V 2.0 – NO Memory Overcommit!
- vCritical: Would you buy a hypervisor from these guys?
Ideally, we wouldn’t have to pay much attention to this Microsoft video, but because we stand behind what we post on Why Choose VMware, we felt it was important for our customers and other companies looking to deploy VMware to hear directly from us. Again, we don’t claim to be perfect and cannot say that we’ll never have any errors on the site, but we will attempt to base everything we claim on a technical evaluation of a currently available product. Microsoft’s answers to what it sees as “myths” don’t really even address factual errors – it’s just more marketing rhetoric. Feel free to take a look at the video for yourself (click on the screen shot) and form your own opinion. Then below, we’ll provide our response to each so-called “myth.”
When reviewing Microsoft’s video, please also make sure to also check out the user comments – they are pretty informative in regards to the ‘value’ the video provided to customers (and Microsoft partners). Two examples:
- “My VAR business is fully baked in MSFT success. So, I’m looking forward to a day when MSFT can compete with XEN and VMWare. That said, this video is hogwash! It’s embarrassing. Quit producing this kind of stuff. you’re slowing my sales cycle!”
- “Wow, if this is the Microsoft Virtualization Strategy team, the Microsoft Partners should be worried because Microsoft just doesn’t get it. As a long time Gold Partner, I find the embarassing. Come on guys, if you are going to release something like this, at least wait until you have a real released solution.”
And if you’ve never done so, read Why Choose VMware for yourself. If you do happen to find something that you feel is a factual error, leave us a comment here and we will recheck the facts and address it.
Responses for Each “Myth” in the Microsoft Video
Myth 1: Live Migration
Microsoft video (paraphrase): We’ll soon have live migration so VMware’s website is misleading.
VMware’s website very clearly compares to Microsoft’s current shipping product – Windows Server 2008 – which does not have live migration. Microsoft likes to play the “futures” card and promote features that are not yet GA. Beta products (Windows Server 2008 R2 Beta and Hyper-V Server R2 Beta) don’t count because no one buys or deploys a beta product in production. Hyper-V Server R2 is rumored to ship (GA) in 2H’09 but most companies really won’t consider that product seriously – they’ll wait and evaluate Windows Server 2008 R2 instead, which Microsoft’s own website says is slated for 2010.
Even once Microsoft ships its implementation of live migration, it will be an unproven, version 1.0 Microsoft technology that lacks real-world reliability testing. In contrast, VMware VMotion is already in use by 70% of VMware VI3 customers in production environments. VMotion is well tested in production and customers trust it on their mission critical applications.
p.s. This is an example of typical Microsoft marketing. When it lacks a feature, it downplays it and says that no one needs it (see live migration example here). Once Microsoft is somewhat close to having that feature, then Microsoft starts talking about why that feature is a core requirement and claims that they’ve now caught up to VMware – even though Microsoft hasn’t yet shipped the feature. (We’ll see the same thing happen with Memory Overcommit – wait and see!)
Myth 2: Clustered File Systems
Microsoft video (paraphrase): Cluster Shared Volumes (CSV) is coming and it will address any deficiencies with Hyper-V’s lack of a clustered file system. VMware’s website is misleading in that it claims Microsoft lacks a true clustered file system.
Again, the Microsoft video is promoting a future technology that has yet to GA – Cluster Shared Volumes (CSV). The Microsoft video also side-steps the issue that Cluster Shared Volumes is not a true clustered file system and therefore does not deliver all the benefits of a technology like VMware VMFS. CSV is just a layer built on top of NTFS that can only be used for Hyper-V.
Microsoft built CSV to do only one thing: to address a major limitation in Hyper-V 1.0, namely the requirement to have 1 VM per LUN in order to do VM-independent HA and quick migrations. But from Microsoft’s architectural descriptions of CSV, it appears that they’ve made trade-offs in performance by requiring multiple ‘coordination nodes’ that significant amounts of meta-data must route through.
Again, the VMware website clearly compares shipping VMware products to shipping Microsoft products. Microsoft needs to ship/GA CSV first before it earns the right to claim that they’ve got a capability.
Myth 3: Hyper-V is Version 1.0
Microsoft video (paraphrase): Hyper-V is a good version 1.0 Microsoft product
No much to say here… Customers will decide for themselves how much of their business they will bet on a version 1.0 product from Microsoft.
Myth 4: Low Performance
Microsoft video (paraphrase): Hyper-V 1.0 performs well
Microsoft likes to point to tests that usually run a low number of VMs per host. This type of test fails to address the fact that consolidation ratios are increasing in line with the increases in x86 hardware computing power (ex. Intel’s Nehalem generation of server processors have 8 cores per socket). Recent tests by the Taneja Group (commissioned by VMware) clearly show how Hyper-V can’t keep up with ESX as the number of concurrently running workloads increase – see report here. Hyper-V lack of scalability directly impacts a customer’s bottom line – see Myth 9 below.
Myth 5: Disk Footprint
Microsoft video (paraphrase): The code size of the hypervisor is not important
The Microsoft video says that VMware should compare in-memory usage instead of disk footprint. The problem is that in-memory usage is very dependent on what’s running on the host at any given time, is not deterministic, and does not show the size of the trusted computing base.
Therefore, disk footprint is the better comparison. First, it’s an apples-to-apples comparison because it does not try to measure things while operations are running and memory usage is continually changing. Second, it more accurately shows the size of the trusted computing base. The premise is simple: the larger the trusted computing base, the more code there is to protect and maintain. VMware designed ESXi to reduce the amount of code that needs to be protected and maintained. Leading analysts have highlighted how Microsoft’s use of Windows in the parent partition is concerning because it represents a large potential attack point.
Myth 6: Broad Hardware Support
Microsoft video (paraphrase): Hyper-V supports a broad set of hardware.
Yes, Microsoft’s Hyper-V architectural allows it to run the same devices as Windows Server 2008 because it uses the same generic, third-party device drivers. But that may not be a good thing. Microsoft has publicly stated (ex. at TechEd 2006) that third party devices drivers are the biggest culprit when it comes to Windows instability. So those are the same drivers that would be running a company’s virtualization deployment where reliability, stability, and uptime are paramount??
Myth 7: End-to-end System Center Management
Microsoft video (paraphrase): VMware’s website should compare to the entire Microsoft System Center suite
The entire Microsoft System Center Suite (VMM, OM, CM) still does not come anywhere close to the virtualization management solutions found in VMware’s vCenter Suite. The full System Center does not have:
- an automated workflow and failover solution for Disaster Recovery (like VMware SRM),
- a self-service, multi-tier, lab automation solution (like VMware Lab Manager),
- an integrated solution to patch off-line VMs (like VMware Update Manager),
- a non-disruptive maintenance mode for host updates (built into vCenter Server),
- a VM lifecycle management solution (like VMware Lifecycle Manager),
- and the gap will grow larger once VMware releases its upcoming management offerings later on in 2009.
Myth 8: Memory Oversubscription
Microsoft video (paraphrase): Memory oversubscription is not as important as VMware makes it out to be. It can also have an impact on performance.
The Microsoft video does not deny that memory oversubscription is useful. Instead, they question the 2:1 ratio that is seen in some of VMware’s prior material. 2:1 was used because VMware clearly saw customers who were able to achieve this ratio in their environment. VMware summarizes the findings from a customer survey on memory overcommit in a blog for those who are interested in how much memory oversubscription is used by VMware customers.
The video also claims that performance can be impacted when using memory oversubscription when applications need all the memory. First, that’s like saying “CPU oversubscription can impact performance if applications require all the CPU.” Remember, physical resource oversubscription is the whole point of server consolidation. Server applications have been found to NOT require all of their physical resources at the same time. So, why virtualize and then lock up resources (such as RAM memory) like was previously done on non-virtualized physical servers? That’s the type of behavior that led to physical server sprawl in the first place. Second, a recent Taneja report (commissioned by VMware) showed an example of how VMware ESX can achieve near-linear scaling even when oversubscribing memory up to a 2:1 ratio (see DBHammer/SQL graph on page 6 of the report).
However, let’s step back for a moment. What a customer cares about is not memory oversubscription in and of itself – they care about the final cost to run the X number of applications that keeps their business going. In a virtual environment, one must understand VM density in order to figure out that cost. Earlier VMware material focused only on memory oversubscription because it was a straight forward way to explain one technology that allowed VMware to achieve higher VM densities.
Since then, VMware has received customer feedback that many other VMware technologies contribute to the higher VM densities and asked why VMware does not talk about those. So here are the top 5 VMware technologies that contribute to higher VM densities (including memory oversubscription):
- Memory Oversubscription – More efficient use of physical RAM by reclaiming unused physical memory and consolidating identical memory pages among VMs on a host.
- Direct Driver Model – VMware ESX can achieve very high I/O throughput and can handle the I/O requirements for more VMs simultaneously requesting hardware resources.
- Support for Large Memory Pages and Nested Page Tables – Optimize memory access and can provide substantial performance benefits for mission critical, memory-intensive applications, can reduce CPU resource consumption by up to 15%.
- DRS with Resource Pools – Dynamically load balance VMs across a cluster so applications get required resources when they need them -a “safety net” that lets administrators run individual servers at higher utilization levels while meeting service level agreements.
- High Performance “Gang” Scheduler – Can account for CPU and I/O needs of virtual machines by dynamically allocating more resources and larger processor timeslices to VMs.
Myth 9: Lower Cost Virtualization Solutions
Microsoft video (paraphrase): Cost-per-application is a theoretical argument.
The Microsoft video claims that cost per application is too “theoretical.” That’s like saying “miles per gallon” on a car is not a relevant metric. As explained on Why Choose VMware, a virtual server represents a many-VMs-to-one-host relationship. So it is imperative to understand how many VMs can run per host as that directly impacts how much software, hardware, and infrastructure are required to virtualize the entire datacenter. It’s quite straightforward: the more VMs one can run on a host, the lower cost it is to the customer. This is one reason why VMware made available its Cost-per-Application Calculator for customers to run different scenarios. The output report includes all of the assumptions and formulas so people can see exactly how the results were calculated.
The Microsoft video also says that people should not compare to VI3 Foundation or other VMware SKUs because they do not match up to Microsoft’s “enterprise” offering. They say that people should only compare to VMware VI3 Enterprise Edition. That’s just not credible since Microsoft’s comparisons always leave out the fact that VI3 Enterprise Edition delivers capabilities that Microsoft Hyper-V lacks. A few examples: VMotion, DRS, Storage VMotion, logical resource pools, DPM, ultra-thin hypervisor offering, off-host backup proxy, …
Myth 10: Need VMware Virtualization?
Microsoft video (paraphrase): VMware wants you to believe that only VMware has the right solution. In reality, Microsoft offers everything you need, and it simplifies the solution stack to 3-layers instead of 4-layers (if you went with VMware).
VMware does believe that it has the better solution – what successful company doesn’t? To support this belief, the assertions and comparisons on the VMware website are based on lab-validated evaluations of the different products. This approach lets the customer decide which offering is best for their requirements.
It is rather ironic that right after criticizing VMware for claiming to have a better solution, the Microsoft spokespeople come right back and say that the Microsoft offering is all that’s needed when it comes to virtualization. The spokespeople essentially say: why bother looking at VMware (or any other third-party software, for that matter) to help you run your business better? Windows is all you need – anything else just adds complexity and cost. In fact, Microsoft’s solution eliminates an extraneous layer from the stack…
VMware prefers to let the market decide on who delivers more value to the customer by having products available today that can meet their requirements.
p.s. In actuality VMware ESXi simplifies that stack. It removes the need for an instance of a general purpose operating systems (ex. Windows in the parent partition) to be running as part of the virtualization layer.