Category Archives: Hypervisor

Some Feedback On How to defrag a Hyper-V R2 Cluster Shared Volume

Posted on by
3

Hans Vredevoort posted a nice blog entry recently on the defragmentation of Clustered Shared Volumes and asked for some feedback & experiences on this subject. He describes the process used and steps taken to defrag your CSV storage and notes that there may be third party products that can handle this automatically. Well yes, there are. Two of the most know defragmentation products support Cluster Shared Volumes and automate the process described by Hans in his blog.  Calvin made a very useful suggestion to use Redirected Access instead of Maintenance mode. This is what the commercial tools like Raxco PerfectDisk and Diskeeper also do.

As the defragmentation of Cluster Shared Volumes requires them to be put into Redirected Access you should not have “always on” defragmentation running in a clustered Hyper-V node. Sure the software will take care of it all for you but the performance hit is there and is considerable. I might just use this point here as yet another plug for 10 Gbps networks for CSV. Also note that the defragmentation has to run on the current owner or coordinator node. Intelligent defragmentation software should know what node to run the defrag on, move the ownership to the desired node that is running the defragmentation or just runs it on all nodes and skips the CSVs storage it isn’t the coordinator for. The latter isn’t that intelligent. John Savill did a great blog post on this before Windows 2008 R2 went RTM for Windows IT Pro Magazine where he also uses PowerShell scripts to move the ownership of the storage to the node where he’ll perform the defragmentation and retrieves the GUID of the disk to use with the  defrag command. You can read his blog post here and see how our lives have improved with the commands he mentions would be available in the RTM version of W2K8R2 (Repair-ClusterSharedVolume  with –defrag option).

For more information on Raxco PerfectDisk you can take a look at the Raxco support article, but the information is rather limited. You can also find some more information from Diskeeper on this subject here.  I would like to add that you should use defragmentation intelligently and not blindly. Do it with a purpose and in a well thought out manner to reap the benefits. Don’t just do it out of habit because you used to do it in DOS back in the day.

To conclude I’ll leave you with some screenshots from my lab, take during the defragmentation of a Hyper-V cluster node.

As you can see the CSV storage is put into redirected access:

And our machines remain online and available:

This is because we started to defrag it on the Hyper-V cluster node:

Here you can see that the guest files are indeed being defragmented, in this case, the VHD for the guest server Columbia (red circle at the bottom):

Hyper-V, KEMP LoadMaster & DFS Replication Provide FTP Solutions For Surveyors Network

Posted on by
Reply

Remember the blog entry about A Hardware Load Balancing Exercise With A Kemp Loadmaster 2200 KEMP Loadmaster to provide redundancy for a surveyor’s GPS network? Well, we got commissioned to come up with a redundant FTP solution for their needs last month and this blog is about what we came up with. The aim was to make due with what is already available.

FTP 7.5 in Windows 2008 R2

We use the FTP Server available in Windows 2008 R2 which provides us with all functionality we need: User Isolation and FTP over SSL.

The data from all the GPS stations is sent to the FTP server for safekeeping and is to be used to overcome certain issues customers might have with missing data from surveying solutions. This data is not being made available to customers by default, it’s only for special cases & purposes. So we need to collect the data in its own folder named after its account so we can configure user isolation. This also prevents GPS Stations from writing in locations where it shouldn’t.

As every GPS Station slogs in with the “Station” account it ends up in the “Station” folder as root FTP folder and can’t read or write out of that folder. The survey solution service desk can FTP into that folder and access any data they want.

The data that’s being provided by the software solution (LanSurvey01 and lanSurvey02) is to be sent to its own folder “Data” that is also set up with user Isolation to prevent the application from reading or writing anywhere else on the file system.

The data from should be publicly available to the customers and for this, we created a separate FTP site called “Public” that is configured for anonymous access to the same Data folder but with read permissions only. This way the customers can get all the data they need but only have read access to the required data and nothing more.

For more information on setting up FTP 7.5 and using FTP over SSL you might take a look here http://learn.iis.net/page.aspx/304/using-ftp-over-ssl/ and read my blog on FTP over SSL Pollution of the Gene Pool a Real Life “FTP over SSL” Story

High Availability

In the section above we’ve taken care of the FTP needs. Now we still need redundancy. We could use Windows NLB but since this network already uses a KEMP Loadmaster due to the fact that the surveyor’s software has some limitations in its configuration capabilities that don’t allow Windows Network Load Balancing being used.

We want both the GPS stations and the surveyor’s application servers to be able to send FTP data when one of the receiving FTF servers is down for some reason (updates, upgrades, maintenance, or failure). What we did is set up a VIP for use with FTP on the Kemp Loadmaster. This VIP is what is used by the GPS Stations and the application to write and by the customers to read the FTP data.

DFS-R to complete the solution

But up until now, we’ve been ignoring an issue. When using NLB to push data to hosts we need to ensure that all the data will be available on all the nodes all of the time. You could opt to only have the users access the FTP service via an NLB VIP address and push the data to both nodes without using NLB. The latter might be done at the source but then you have twice the amount of data to push out. It also means extra work to configure and maintain the solution. We could copy the data to one FTP node and copy it from there. That works but leaves you very vulnerable to a service outage when the node that gets the original copy is down. No new data will be available. Another issue is the fact that you need a rock-solid way to copy the data and have it done it a timely manner, even after downtime of one or more of the nodes.

As you read above we provide an NLB VIP as a target for the surveyor’s application and the GPS Stations to send their data to. This means the data will be sent to the FTP NLB array even if one of the nodes is down for some reason. To get the data that arrives from 2 application servers and from 40 GPS Stations synchronized and up to date on both the NLB nodes we use the Data File System – Replication (DFS-R) built into Windows 2008 R2. We have no need for a DFS-Namespace here, so we only use the replication feature. This is easy and fast to set up (add the DFS service from the File Server Role) and it doesn’t require any service downtime (no reboot required). The fact that both the FTP nodes are members of a Windows 2008 R2 domain does help with making this easy. To make sure we have replication in all direction we opt to set it up as a full and the replication schedule is 24/7, no days off J Since we chose to replicate the FTP root folder we have both the Data and the Stations folders covered as well as the folder structure needed to have FTP user Isolation function.

This solution was built fast and easily using Windows 2008 R2 out of the box functionality: FTP(S) with User Isolation and DFS-R. The servers are running as hyper-V guests in a Hyper-V cluster providing high availability through Live Migration.

A VDI Reality Check @ BriForum 2011 For Resource Hungry Desktops In A Demanding Environment

Posted on by
Reply

So what did we notice? VDI generates enough interest from various angles that is for sure. Both on the demand side as on the (re)seller & integrator side. Most storage vendors are bullish enough to claim that they can handle whatever IOPS required to get the most bang for the buck but only the smaller or newest players were present and engaged in interaction with the attendees. One thing is for sure VDI has some serious potential but it has to be prepared well and implemented thoroughly. Don’t do it over the weekend and see if it works out for all your users.

The amount of tools & tactics for VDI on both the storage side and the configuration/management side is both more complex and diverse than with server virtualization.  The possible variations on how to tackle a VDI project are almost automatically more numerous as well. This is due to the fact that desktops are often a lot more complex and heterogenic in nature than server-side apps. On top of that, the IO on a desktop can be quite high. Some of it can be blamed on the client OS but lots of that has to do with the applications and utilities used on desktops.  I think that developers had so many resources at their disposal that there wasn’t to much pressure on optimization there. The age of multi-cores and x64 bit will help in thinking more about how and application uses CPY cycles but virtualization might very well help in abstracting that away. When a PC has one vCPU and the host has 4*8 cores, how good is that hypervisor at using all that pCPU power to address the needs of that one vCPU?  But I digress. All in all, it takes more effort and complexity to do VDI than server virtualization. So there is a higher cost or at least the APEX isn’t such a convincing clear cut story as it is with server virtualization. If you’re not doing the latter today when and where you can you are missing out of a major number of benefits that are just to good to ignore. I wouldn’t dare say that for VDI. Treating VDI just like server virtualization is said to be one of the main reasons for VDI failing or being put on hold or being limited to a smaller segment of the desktop population.

My experience with server virtualization is also with rather heterogenic environments where we have VMs with anything between 1 and 4 virtual CPUs, 2 to 12 GB of RAM. And yet I have to admit it has been a great success. Never the less I can’t say that helped me much in my confidence that a large part of our desktop environment can be virtualized successfully and cost-effectively as I think that our desktops are such vicious resource hogs they need another step forward in raw power and functionality versus cost. Let briefly describe the environment. 85% of the workforce at my current gig has dual 24” wide screens, with anything between 4GB to 8 GB of RAM, Quad-Core CPUs and SCSI / SATA 10.000 RPM disks with anything between 250 GB to 1TB local storage in combination with very decent GPUs. Now the employees run Visual Studio, SQL Server, multiple CAD & GIS packages, and various specialized image processing software that gauges image and other files that can be 2GB or even higher. If they aren’t that large than they are still very numerous. On top of that 1Gbps network to the desktop is the only thing we offer anymore. So this is not a common office suite plus a couple of LOB applications order, this is a large and rich menu for a very hard to please audiences. That means that if you ask them what they want, they only answer more, more, more … And I won’t even mention 3D screens & goggles.

Now I know that X amount of time the machines are idle or doing a lot less but in the end that’s just a very nice statistic. When a couple of dozen users start playing around with those tools and throw that data around you still need them and their colleagues to be happy customers. Frankly even with the physical hardware that they have now that can be a challenge. And please don’t start about better, less resource wasting applications and such. You can’t just f* the business and tell them to get or wait for better apps. That flies in the face of reality. You have to be able to deliver the power where and when needed with the software they use. You just can’t control the entire universe.

I heard about integrators achieving 40-60 VMs per host in a VDI project. Some customers can make due with Windows 7 and 1GB of RAM. I’m not one of those. I think the guys & gals of the service desk would need armed escorts if we rolled that out to the employees they care for. One of the things I notice is that a lot of people choose to implement storage just for VDI. I’m not surprised. But until now I’ve not needed to do it. Not even for databases and other resource hogs. Separate clusters, yes, as the pCPU/vCPU ratio and Memory requirements differ a lot from the other servers. The fact that the separate cluster uses other HBA’s en LUNS also helps.

Next to SANs local storage for VDI is another option for both performance and cost. But for recovery, this isn’t quite that good a solution. The idea of having non-persistent disks (in a pool) or a combination of that with persistent disks is not something I can see fly with our users. And frankly, a show of hands at BriForum seems to indicate that this isn’t very widespread. VDI takes really high-performance storage, isolated from your server virtualization to make it a success. On top of that if you need control, rapid provisioning, user virtualization &  workspace management in a layered/abstracted way. Lost of interest there but again, yet more tools to get it done. Then there is also application virtualization, terminal service-based solutions etc. So we get a more involved, divers, and expensive solution compared to server virtualization. Now to offset these costs we need to look at what we can gain. So where do the benefits to be found?

With non-persistent disk you have rapid provisioning of know good machines in a pool but your environment must accept this and I don’ see this flying well in face of the reality of consumerization of ICT. De-duplication and thin provisioning help to get the storage needs under control but the bigger the client-side storage needs and the more diverse these are the fewer gains can be found there. Better control, provisioning, resource sharing, manageability, disaster recovery, it is all possible but it is all so very specific to the environment compared to server virtualization and some solutions contradict gains that might have been secured with other approaches (disaster recovery, business continuity with SAN versus local storage). One of the most interesting possibilities for the environment I described was perhaps doing virtualization on the client. I look at it as booting from VHD in the Windows 7 era but on steroids. If you can save guard the images/disks on a SAN  with de-duplication & thin provisioning you can have high availability & business continuity as losing the desktops is a matter of pushing to VM to other hardware which due to abstraction by virtualization should be a problem. It also deals with the network issues of VDI, a hidden bottleneck as most people focus on the storage. Truth be told, the bandwidth we consume is that big, it could be that VDI might have it best improvements for us on that front.

Somewhat surprising was that Microsoft, whilst being really present at PubForum in Dublin, was nowhere to be seen at BriForum. Citrix was saving it’s best for its own conference (Synergy) I think. Too bad, I mean when talking about VDI in 2011 we’re talking about Windows 7 for the absolute majority of implementations and Citrix has a strong position in VDI really giving VMware a run for their money. Why miss the opportunity? And yesterday at TechEd USA we heard about the HSBC story of a 100.000 seat VDI solution on Hyper-V http://www.microsoft.com/Presspass/press/2011/may11/05-16TechEd11PR.mspx.

On a side note, I wish I would/could have gone to PubForum as well. Should have done that. Now, these musings are based upon what I see at my current place of endeavor. VDI has a time and place where it can provide significant operational and usage advantages to make the business case for VDI. Today, I’m not convinced this is the case for our needs at this moment in time. looking at our refresh schedule we’ll probably pass on a VDI solution for the coming one. But booting from VHD as a standard in the future… I’m going to look into that, it will be a step towards the future I think.

To conclude BriForum 2011 was a good experience and the smaller scale of it makes for good and plenty of opportunities for interaction and discussion. A very positive note is that most vendors & companies present were discussing real issues we all face. So it was more than just sales demos. Brian, nice job.

Exchange 2010 SP1 DAG & Unified Messaging Now Supports Host Based High Availability & Live Migration!

Posted on by
Reply

Well due to rather nice virtualization support for Lync and the fact that Denali (SQL Server vNext) does support DAG like functionality with Live Migration and host based clustering, it was about time for Exchange 2010 to catch up. And when we read the white paper  Best Practices for Virtualizing Exchange Server 2010 with Windows Server® 2008 R2 Hyper V™ that moment has finally arrived. I have to thank Michel de Rooij at  for bringing this to our attention http://eightwone.com/2011/05/14/exchange-2010-sp1-live-migration-supported/. So now we have the best features in virtualization at our disposal and that simply rocks. We read:

“Exchange server virtual machines, including Exchange Mailbox virtual machines that are part of a Database Availability Group (DAG), can be combined with host-based failover clustering and migration technology as long as the virtual machines are configured such that they will not save and restore state on disk when moved or taken offline. All failover activity must result in a cold start when the virtual machine is activated on the target node. All planned migration must either result in shut down and a cold start or an online migration that utilizes a technology such as Hyper-V live migration.”

“Microsoft Exchange Server 2010 SP1 supports virtualization of the Unified Messaging role when it is installed on the 64-bit edition of Windows Server 2008 R2. Unified Messaging must be the only Exchange role in the virtual machine. Other Exchange roles (Client Access, Edge Transport, Hub Transport, Mailbox) are not supported on the same virtual machine as Unified Messaging. The virtualized machine configuration running Unified Messaging must have at least 4 CPU cores, and at least 16 GB of memory.”

And it is NOT ONLY for Hyper-V, look at the Exchange Team blog here “The updated support guidance applies to any hardware virtualization vendor participating in the Windows Server Virtualization Validation Program (SVVP).’” Nice!

Anyone who’s at TechEd USA 2011 in Atlanta should attend EXL306 for more details. Huge requirements yes, but the same goes for physical servers. That’s how they get the performance gains needed, it’s done by lowering IO by using large amounts of RAM.

Think about the above statement, we now have support for host clustering with live migration, possibly together with technology like for example Melio (SanBolic) on the software side or Live Volume (Compellent) on the storage side to protect against SAN Failure (local or remote) and combined with DAG high availability for the databases in Exchange 2010 (which can be multi site) this becomes a very resilient package. So to come back to my other post on a brighter future for public folders, if they can sort out this red headed stepchild of the Exchange portfolio they have covered all their bases and have a great platform with the option of making it better, easier and cheaper to implement, operate & use. No one will argue with that.

I know some people will say all this is overkill, to complex, to much or to expensive. I call it having options. When the S* hits the fan and you’re “in the fight of your life” wading your way through one or multiple IT disasters to keep that mail flow up an running it is good to have multiple options. Options mean you can get the job done using creativity and tools. If you have only one tool and one option Murphy will catch up with you. Actually this is one of my most heard shout outs to the team “give me options” when problems arise. But at what cost do these options come? That is up for the business and you to decide. We’re getting very robust options in Exchange that can be leveraged with other technologies for high availability that have become more and more main stream. This means none of all this needs to be bought and implemented just for Exchange. They are already in place. Unless your IT “strategy” the last 10 years was run Windows 2000 & Exchange 2000 until the servers fall apart and we don’t have any more spares available on e-bay before we consider moving along.