VMFS Isn’t the Problem Here

Good response, converter would be a good way to address the issue.

But, above all issues working directly with VMDK files is not a good practice. Same goes for .VHD files on Hyper-V to be fair.

I’m just wondering if this company’s DR plan is to sift through the burning rubble to find their cheap SATA drives…And then rely on an “integrator” that doesn’t know how to mount a USB disk from a service console…
Dave

I came across that article… The guy is dumb and frustrated… He should have postponed his customer DR test, attend VMworld 2009 and learn something instead of posting craps like this…

Scott you were right on. Lets all agree that this “DR Plan” was/is not a good one. I was still confused that someone with so much experience as an integrator was not able to come up with a solution to this minor problem. After I was done readying his “article” I was so mad at myself for even clicking on the link to give him the web hits.

FastSCP, anyone?

Lot of variables here that Jason did not go into regarding his recovery environment. Assuming that his recovery ESX/ESXi box had enough spindles. He could have attached each portable drive to a nbk and let it fly. Not the best DR plan but it mirrors what people are trying to do when faced with financial, logistical and technical constraints.

Rick can you clarify your comment about working directly with VMDKs?

I thought point of encapsulating a server into a VM was so it would be portable? At some point in time you need to archive the VM’s. Normally this would be VCB or partner product.

Are there better methods to move VM’s between disconnected systems?

Quite obviously a Microsoft-sponsored hit piece that wasn’t even fact checked.

FAT32 has a 2GB filesize limitation and VI3.5 virtual disks are preallocated with no option for 2GB splitting, so the configuration he mentions is not even possible unless he had a terabyte of VMs that were all less than 2GBs in size.

USB does in fact work in the Service Console, if you know how to use the mount command. SMB/CIFS also works if you know how to use the smbclient command. The entire task could have been completed in less than an hour had he used a simple NFS mount from within the Service Console out to the NFS server he setup - the comment by Scott is spot on as well about using VMware Converter 4.

VMFS has been a convenient target for Citrix and Microsoft to attack over the years, until they figured out that carving a LUN for each VM going into deployment is inconceivable for any virtualization project greater than a handful of VMs. Now Microsoft is building their own clustering filesystem for Hyper-V of course…

It’s a common problem that one component gets blamed when another one is the real bottleneck. In a poorly functioning team, that blamed component could get upgraded or even replaced. Thanks to your blog, teams all over the world can reference your article and right wrong-thinking.

Thanks, Rick.

Man I had a lot of harsh things to say about this guy and then thought better of it. I’m glad I’m not the only one.

For a guy with “decades of experience in heterogeneous environments” he doesn’t seem to understand a lot of basic approaches. It’s basically a non problem and goes to show that his customer is likely using the wrong vendor to help support their virtualization effort.

The situation itself is a bit contrived and in the best case scenario he’s still going to have to re-export every virtual disk anyway (since Fat32 is just an awful filesystem for storing large files.) you’re better off just using Converter.

I’m quite disappointed in that guy.

@Andrew Storrs; “Or he could have used Veeam’s (free) FastSCP”
And that would have been unbearably slow. Even with screaming-fast hard drives.
http://www.veeam.com/forums/viewtopic.php?f=4&t=985
Upload to ESXi: 5-10MB/s
“Why ESXi upload is so slow ?
For unknown reasons, upload through file management API is capped on ESXi.
If the author of the article was using ESXi, this is probably the predominant reason it was slow, not their SATA hard drives.

If he had used VMware ESX and FastSCP, it would be much faster.
It also would have been a lot faster, if he used VMware converter to perform the upload.

Or just attached the NFS volumes to the DR VM host, registered the VMs, and used SVMotion or cloning in the DR vCenter to migrate them to the VMFS datastore.

As vCenter does not seem subject to the “upload speed CAP” ESXi imposes on the datastore browser.

I there’s one good point about VMFS being a black box.. for DR purposes it _is_ nice to have general tools for operating on your filesystem, and reverting a filesystem snapshot of just one VM instead of everything. This is especially useful if there is some sort of damage to the filesystem, you can use general repair tools — and is a reason I like shared NFS storage for VMs; if the VMFS metadata gets corrupt, you’re basically screwed.

If the ext3 filesystem on your Linux NFS server has an issue, you can run fsck if you need to.

I find it funny that Eric Gray (the guy who works for VMware and puts out a lot of anti Microsoft information that he claims is just his ‘personal’ opinion) would accuse someone else of being in the tank for a company. If you check Mr. Perlow’s info you can clearly see he works for IBM, not Microsoft. Eric, I think you should know better and be more careful. It’s obvious that your post had an effect on others as Gregory and Alex then latched onto believing this is a post from Microsoft which simply isn’t true. The guy may personally prefer Microsoft and Hyper-V, but that is far different than being in the tank. Being that you are paid by VMware Eric, I would say it is safe to label you as “in the tank” for VMware. However, if independent individuals prefer Microsoft or VMware it is based upon their own technology preferences.

As to the actual problem, I’m not a VMware guy but it sounds more like an issue of things not working the way he expected or seemed logical to him more than not working perhaps. Though Jimmy’s point about ESXi’s upload cap seems relevant (and odd). It’s nice that there are add-ons that can be used to accomplish the function, but soemthimes if you aren’t in the know about the add-on it can be frustrating trying to figure out how to get something to go that just works on other systems you are used to. I’m not totally defending him because you need to know the different tweaks and ways to accomplish something in all the systems you work with, but you can’t tell me as an IT person everyone commenting here hasn’t at some point been annoyed that they can’t do something in product A as they had been able to do in product B even though product A has a way to do it that is just different. Sometimes until you figure out the trick and the lightbulb goes on it all just seems like black magic.

Scott, If you take a look at the comments section on Eric’s post you will see the calrifications of Eric’s assumptions from much more knowledgable people than myself that do a much better job of explaining the issues. However, some gernal gists of issues. Eric declares CSV a “delicate” feature, inserting the classic FUD that “oh no, can I rely on this to keep working?”. Eric’s sole basis for this claim is that there is a warning pop up that says that CSV is only currently supported for Hyper-V. I’m sorry but that is beyond a leap of logic. Microsoft is simply saying this is what it has been tested and certified for at this point. They’ll probably leverage CSV for other technologies in the future, but it would be irresponsible of them not to give the warning that they haven’t tested such yet. To declare CSV delicate based upon that is FUD. Next Eric claims you can’t backup a VM on a CSV. This is my point about attacking an RC version of a product. The API is currently available for backing up VMs on a CSV, so it is NOT a limitation of CSV as Eric is hoping to imply. R2 isn’t genraly available until the end of October. It’s not fair to expect all of the partners (I consider DPM a partner too because we are talking seperate units of a very large organization) to have backup software ready for the RC. This isn’t the first time I’ve seen Eric attack Hyper-V on something based on using the beta or RC product either. If I attacked VMware based on some partner product not being ready yet for a beta of theirs, I think Eric would call fowl. There are also issues raised in the initial post and in the comments section about a single coordinator node and how IO works in the cluster. Again look to the comments, there is some good clarity from a Microsoft person about how IO is direct, the balancing of coordination, and the redirecting of IO only in the case where all connections to the storage are lost (which I personally think is actually a very cool double safety feature).

Nate, with all due respect — you don’t know what you’re talking about.

I am using the RTM version of Windows 2008 R2, not the RC as you have falsely accused me of.

Have you completely overlooked the Microsoft Virtualization messaging coming from Woolsey/Greshler/Yuen? They claim Hyper-V R2 is available with free live migration — today. They imply that everything is ready to deploy now, in order to stall adoption of VMware.

The fact is, their hypervisor has been released but it’s not ready to be put into production. The GA date in October is a mostly a marketing launch, DPM will not suddenly be able to back up VMs on CSV LUNs.

Perhaps you are fine with them bloviating about what they intend to provide at some point in the future. I’m not, and I’ll write about it on VCritical. Please consider attacking me on my own blog and not Scott’s.

Eric