On Tuesday, March 31, Microsoft's Edwin Yuen will be hosting a live web chat 11am-3pm EST. Edwin is a sr. technical product manager. Edwin came to Microsoft with the acquisition of Softricity (and the SoftGrid application virtualization technology). He now also covers Hyper-V and System Center VMM. Sign up here.
Patrick
I’d like to start this post by saying thank you to our friends at Microsoft for running my blog post here. As AMD is mid-way through Virtualization Ecosystem Month (VEM), it has been exciting to see the partner support around the topic of virtualization and I encourage you to check out Microsoft’s guest post on my blog.
Both AMD and Microsoft are continually asked if virtualization can really become a sustainable technology in the mainstream computing market. Skeptics are quick to point out that the technology world is littered with products that have never made it beyond the early adopter edge of the market. Names like Apple Newton, Betamax, Sega Dreamcast, and OS/2 come up in various articles and discussions about technologies that didn’t become widespread market successes. Will virtualization suffer this same fate? Is it destined to stay on the fringe?
In my opinion, all indications are that virtualization is jumping the technology chasm between early adopters and the mainstream market. Recent survey results from Forrester demonstrate that firms (both large and small) are in the midst of rethinking and overhauling IT infrastructure and client systems. Virtualization is one of technologies at the center of these efforts with Forrester reporting that 54% of the enterprises and 53% of the SMBs surveyed have either implemented x86 server virtualization or plan to do so within the next 12 months—which is significant in today’s current economy.
It is no secret that Microsoft’s virtualization solutions play a critical role in virtualization’s march into the mainstream. In my view, virtualization for Microsoft means helping businesses of all sizes maximize cost savings and improve business continuity with products based on familiar Windows interfaces and well-known Windows-based technologies.
AMD and Microsoft have an on-going technology collaboration designed to integrate our respective hardware and software products to better handle the rigors of memory-intensive virtualization environments. The AMD Opteron processor provides the underlying foundation with AMD-V hardware-assisted virtualization technology, along with a broad range of power management capabilities. Windows Server 2008 Hyper-V and Microsoft Hyper-V Server 2008 utilize these virtualization and power management technologies to help deliver well performing and highly efficient virtualization solutions. Together, Microsoft and AMD products provide flexible management, optimal user performance, and enable data centers to conserve power through workload consolidation—a big win for any company looking for ways to cut costs and optimize their IT infrastructure.
AMD and Microsoft have already clocked some impressive successes with mid-market companies who are turning virtualization into a business advantage. For example, ServiceU used AMD-based Dell server running Hyper-V to reduce the number of physical servers in its three data centers by 35 percent, and lower the energy utilization at one data center by up to 60 percent—yes, you read that correctly, 60 percent! While WASSER used AMD-based HP blades and Hyper-V to create a virtualized environment that reduced the number of physical servers by more than half without a major sacrifice in performance. With the next versions of Microsoft’s highly anticipated server virtualization products, Windows Server 2008 R2 Hyper-V and Microsoft Hyper-V Server 2008 R2, there will be added support for AMD-V Rapid Virtualization Indexing (RVI) technology. In case you aren’t familiar, RVI enables virtual machines to directly manage memory utilizing hardware resources rather than software resources. This process can help to accelerate the performance of demanding virtualized applications, such as database and Web serving, by reducing hypervisor cycles and the associated performance penalty that is commonly associated with virtualization.
In addition to the virtualization technology coming out of AMD and our technology partners, we’re also seeing a lot of conversation on community sites such as Spiceworks, which I personally think is very indicative to a maturing (and thriving) technology and definitely not one that could ever be in the same category as Pets.com. In fact, I have a feeling if virtualization continues to be the go-to technology for businesses looking to maximize efficiency and streamline their data center, we might start seeing virtualization fanboys start popping up—now wouldn’t that be cool?
Margaret Lewis, Product Marketing Director at AMD
Exciting news for Windows Server Hyper-V, System Center, and VMWare customers. The beta of System Center Virtual Machine Manager 2008 R2 is now available on the Microsoft Connect site for download. If you don't already have access to VMM on the Connect site you can get to it here by simply filling out a quick survey, which will help us better understand your virtualization needs. The new beta continues to extend the data center management capabilities of the System Center suite of offerings to build on the all up physical and virtual IT asset management features you already know, including the integration of Operations Manager and Virtual Machine Manager through the PRO functionality.
You can find current, full details on the VMM 2008 R2 beta in a post by Rakesh Malhotra at this location.
In summary, VMM 2008 R2 beta now supports new Windows Server 2008 R2 beta features such as Live Migration. Some additional features include:
A reminder also that the Microsoft Management Summit is April 27 - May 1 in Las Vegas. This is an excellent time to learn more about all of the management offerings from Microsoft including the System Center suite and VMM 2008 R2. Hope to see you there.
Please go download the beta of Virtual Machine Manager 2008 R2 today from the Microsoft Connect site and let us know what you think.
A quick note that a beta of the Hyper-V management pack for System Center Operations Manager (2007, 2007 R2) is available. In combination with SCVMM (for advanced monitoring of VMs across your Hyper-V, ESX and Virtual Server environments), this SCOM management pack enables health/perf monitoring of the Hyper-V host. The management pack includes health diagram view of virtual machines, virtual components roll-up per host, critical Hyper-V Service monitoring, disk space threshold monitoring.
This beta management pack supports the following OS only (for now):
§ Windows Server 2008 Standard Edition (Full Installation with Hyper-V Role enabled)
§ Windows Server 2008 Enterprise Edition (Full Installation with Hyper-V Role enabled)
§ Windows Server 2008 Datacenter Edition (Full Installation with Hyper-V Role enabled)
To request an invitation to access this beta, you can send a message to the following address: MPCC@microsoft.com
The threat of virtualization sprawl. That was a theme my colleagues heard last week at IDC's "Directions" conference in San Jose. And true to IDC's form, they backed up their predictions with some numbers. Here's an excerpt from one article:
Virtualization has often been seen as something of a magic bullet to this problem, promising to consolidate a number of low-utilization servers onto a single piece of hardware. But the average number of virtual machines per server is only five, Bailey noted, with that number going to eight by 2012. So much for the vision of consolidating dozens of servers onto one machine. More important, though, was that IDC found that just going from five virtual machines to eight means there will be 100 million new servers by 2012, and "all of them still need to be managed." That's a problem, she said, since the tools to do this are not keeping pace.
Our customers have referred to this issue as "islands," referring to the need for different management tools, interfaces, etc. to manage their heterogeneous environment. After all, customers and partners tell us, they're trying to manage services, no matter if the applications run on Windows or non-Windows, physical or virtualized.
For those of you in that last camp, like Atlanta Journal Constitution, Mamut and Maxol, you know that Microsoft and some other systems management vendors are creating tools to keep pace with heterogeneous hypervisors and VMs, and as well traditional physical systems and non-virtualized applications. System Center is one such management tool; VMware vCenter isn't (yet, according to Alessandro). To elaborate on this point, check out RakeshM's latest blog post here. Here's an excerpt from his intro:
Put simply, people want to use a single primary console for day to day management of virtual machines across multiple hypervisors so we went after this problem. As a result, multi-hypervisor management via SCVMM 2008 has proven to be enormously popular with customers and partners alike.
Rakesh sheds light on VMware's concocted "demos" (nice touch using Microsoft exec names "muglia" and "ballmer" as host names) and conjecture. Rakesh summarizes his post with the following:
If you are thinking of using SCVMM 2008 to manage VMware because you have a mixed environment (and we have many many customers who are doing just that), keep in mind that SCVMM does not require you to uninstall or remove VMware Virtual Center from your environment. In fact, you have to keep Virtual Center around because VMware does not expose some APIs (like Vmotion) through ESX. We’re a manager of managers so it is nearly risk free to try it out and make up your own mind about how effective we are.
Oh, and by the way, Rakesh also says that the beta of SCVMM 2008 r2 will be coming this month.
I can't wait to see what demos VMware concocts next to counter the physical/virtual management sales and marketing by HP, IBM, CA and BMC. HP's Peter Spielvogel tips his hand here with HP's reaction. I'm sure this reaction will become stronger once VMware crosses the line to be more competitor than partner.
The other day, Virtualization Review published a comparative performance test of three hypervisors: VMware ESX 3.5, Windows Server 2008 Hyper-V and Citrix XenServer. You can see it here.
NOTE - there are few independent, published performance reviews of hypervisors because including ESX in the review without VMware's permission violates the VMware EULA about posting benchmarks. VMark doesn't count as independent. Amongst reviewers, this EULA restriction is well-known and am told serves as a deterrent to try to do performance comparatives. Rick Vanover and his editor, Keith Ward, deserve kudos for securing VMware approval for the performance comparison without jeopardizing journalistic integrity. Way to go!
OK, back to Rick Vanover's test. His test objectives:
All the hypervisors offer essentially the same base functionality. In this series of tests, the objective was to put the same workloads on each one and see how they stack up. The types of workloads tested varied, to simulate a typical environment in which some virtual machines (VMs) are stressed, and some aren't. Each platform was subjected to the same test plan parameters, to give a fair accounting of their performance.
Read about the comparison parameters, test environment and caveats. The results will be surprising (in a "man bites dog" sort of way) to many. Keith wrote:
The results, from writer/online columnist Rick Vanover, were startling, to say the least. The Porsche of hypervisors? XenServer. Raise your hand if you saw that coming. It outperformed Hyper-V and ESX in most categories. The pokiest? ESX. Again, not at all what I expected. In fact, even in the few tests ESX came out on top, it barely edged out the competition. Microsoft did well across the board, and is definitely a fine product.
We're pleased to see Hyper-V won 4 of the 11 tests (the others going to XenServer by a less than a horse length). For example, test 2 focused on a large number of heavy workload systems: 1 database server running one midsize database and 12 VMs with a heavy workload of CPU, memory and disk operations. Key takeaways from this test:
· Hyper-V completed SQLjob 52% faster than ESX.
· Hyper-V is 2.3 times faster than VMware ESX in CPU oVirtualization Review's hypervisor testperations.
· Hyper-V is 3 times faster than VMware ESX in test for average RAM operations.
At the end of the article, Rick ran one set of tests with 3 GB overcommit for ESX 3.5. Rick pointed out that this feature is useful to many in the data center, it does come at the expense of performance. The test showed that with ESX overcommit enabled:
· Average CPU Operations per VM were 3x slower
· Average Disk Operations per VM with 4.5x slower
· Average SQLjob Completion Time was 33% slower
As Rick wrote:
Results showed that while you can load more guests onto the host, there's no free lunch. There was a dip in performance and database response time.
Another area of feedback on the Hyper-V VSS Writer is that it does not give much by way of granularity beyond the level of the virtual machine. As a result of that, backup applications are stuck to backing up the entire VHD file, which could easily end up being a rather onerous process, expesially if the administrator is backing up to a separate DR site.
To work around this, backup vendors have been looking into solutions for browsing/indexing the files in the in the VHD file. In general, the user can mount the VHD as another drive on the host and browse its file system. The way to do this is not available via the UI in Windows Server 2008 Hyper-V, but is available via the WMI API. Ben Armstrong has a blog post on doing that via vbscript/powershell over here. Given this avenue, the first approach is to follow the following steps:
1. Create VSS snapshot of that involves the VM
2. Mount the VSS snapshot as a volume on the host
3. Get the VHD residing in the VSS snapshot
4. Mount the VHD as drive on the physical machine
5. Go to town on the files residing in the VHD.
Problem with the steps above is that #4 fails due to the fact that the snapshot volume is a read-only volume and the WMI API for mounting VHDs tries to mount it in RW mode. So, we add a step 3a to the sequence and modify step 4 slightly. Here is the updated sequence:
2. Mount the snapshot as a volume on the host
3a. Create a differencing disk on a physical volume with the VHD in the VSS snapshot as the parent disk
4. Mount the VHD created above as drive on the physical machine
since the differencing disk sits on a RW volume, the Mount API will succeed, while keeping the VHD in the snapshot unaffected. The backup application can now browse/index the files inside volume residing in the VHD at will and allow for a more granular and thus optimized backup experience for the user.
There are some important caveats to keep in mind about this approach:
1. While this approach allows for granular backup of specific files inside a virtual machine, the onus for ensuring the right files are available upon restore rests on the backup application as well. So, the backup application will need to do a similar process of mounting the VHD and copying down the right files during the restore process.
2. If the virtual machine in question has snapshots, this approach is not possible in v1 it is not possible for a user or a third-party application to create a differencing disk off a snapshot VHD file (ie: a .avhd file). If the virtual machine has snapshots, the backup application will need to copy the VHD and avhd files to RW media before mounting the avhd files in the physical machine. Additionally, the backup application will need to know the latest snapshot in the snapshot tree.
3. In Windows Server 2008 R2, VHD files can be mounted in RO mode using the native disk management APIs. Thus, there is no need to create differencing disks anymore and the VHDs (as well as AVHDs) can be mounted directly from the snapshot volume.
While the Hyper-V VSS writer provides backup admins significant flexibility in their ability to back up VMs in an application-consistent way while running backup applications in the physical machine, it does have a prominent limitation when it comes to VMs with pass-through storage. The VSS Writer excludes pass-through storage from the set of items to be backed up for a VM. Here I am using the term “pass-through” rather loosely, to include local storage which may be configured directly on the virtual machine (which may be iSCSI initiated from the physical machine or FC attached to the physical machine) as well as iSCSI targets that are initiated directly from the guest. Technical issues due to the way we create VSS snapshots as well as challenges while doing the restore operation has kept us from providing this functionality so far. I will not, however, belabor the reader with the details of the technical challenges. Instead, let us look at ways to work around this limitation.
Approach 1: run backup in guest and backup in host
One approach is to run backup applications within the virtual machine along with a backup application in the host. The backup application running in the virtual machine will capture the data residing on the pass through disk while the host side backup application will capture the VHDs and configuration files associated with the VM. The user would then need to run the backup in the virtual machine followed by backup in the host. During the restore process, the reverse process is followed: restore the host-side files and then run the restore inside the virtual machine after booting it up.
There are a two problems with this approach. Firstly, having to orchestrate the backup process within the virtual machine with the backup process in the host and doing something similar during restore is going to be challenging for any management entity; be it an IT admin or a management application. Secondly, there is an additional limitation for pass-through disks which are connected to the host via FC: The value-added storage management functionality provided by the storage manufacturer (for example, hardware VSS snapshots) is filtered out from the VM due to the fact that the Hyper-V storage stack filters out custom CDBs. These two combine to make this workaround rather unattractive.
Approach 2: Fixed size VHDs
At this point, let us take a moment to look at the motivation behind using pass-through disks for virtual machines. The first reason, stated by many is the enhanced performance available through the use of pass-through storage. The second reason is the ability to use the advanced management capabilities provided by the storage providers for backing up LUNs (iSCSI or FC).
Now, in Windows Server 2008 R2, the performance difference between VHD-based storage and pass-through storage has been reduced a lot. This is particularly true for fixed-size VHDs. In our internal perf labs, we have had test runs where the performance of fixed size VHDs actually matched the performance of pass-through disks. Even in v1, the performance of fixed disks was pretty close to that of pass-through disks. So, instead of configuring the LUN as a pass-through device on the virtual machine, the user could make the LUN online on the physical machine. The user could then create a fixed size VHD that occupies the entire space of the LUN and configure the virtual machine to connect to that VHD. Since the LUN is exposed to the physical machine, the management capabilities provided by the storage provider (e.g: hardware VSS snapshots) are available to be used. In this configuration, the user then does not need to run any backup application inside the virtual machine, as none of the volumes in the virtual machine (by virtue of being backed by VHDs) are excluded from the backup. Additionally, the hardware VSS snapshots and LUN management capabilities provided by the storage manufacturer can be leveraged entirely from the host.
Bryon over at the Windows Server Division blog pointed out a certificate error that can occur for those of you running WS08 Hyper-V and connect using vmconnect. The cert error results in the inability to start or connect to VMs running on WS08 Hyper-V or MS Hyper-V Server 2008.
Here's an excerpt:
Symptoms and resolution:
§ You may be unable to start or connect to virtual machines running on Windows Server 2008 or Microsoft Hyper-V Server 2008. This occurs when connecting using vmconnect. Connections made using Remote Desktop won't be affected.
§ KB Article 967902 has been created that details the symptoms and resolution. This KB article provides a direct link to download the quickfix to resolve this error.
Important Notes:
§ Though this error may occur, the Hyper-V service will continue to operate. Neither the Hyper-V host nor the running virtual machines will go offline.
§ It is not expected that this issue can be exploited for malicious purposes.
§ Customers running Windows Server 2008 R2 Hyper-V beta won’t experience this error.
HP has published 2 white papers describing their NIC teaming support for Hyper-V.
First is a 5-page "how to" document specific to HP ProLiant. Here's the abstract:
HP ProLiant Network Teaming Software allows ProLiant systems running Microsoft Windows Server 2008 Hyper-V to take advantage of network controller teaming technology. This paper describes, for system administrators and technicians, the method of software installation for implementing network controller teaming and it identifies known issues with teaming and Hyper-V software.
Second is a 59-page whitepaper that:
provides a high- and low-level discussion of the technology behind HP ProLiant Network Adapter Teaming for HP ProLiant servers running Microsoft Windows. HP ProLiant Network Adapter Teaming is software-based technology used by server administrators and network administrators to increase a server’s network availability and performance. HP ProLiant Network Adapter Teaming provides network adapter,network port, network cable, switch, and communication path fault recovery technology, in addition to, transmit and receive load balancing technology.