Virtual Insider

Blog archive

vSphere 5 Top 10: VMFS-5 at #3

If you've been reading this blog long enough, you know by now that I'm a big fan of VMFS. So, it should come as no surprise that VMFS-5 landed in the No. 3 spot on this countdown.

VMFS is a purpose-built file system optimized to run virtual machines. One of the big drawbacks of VMFS-3 was its size limitation of 2TB. With the introduction of the vSphere APIs for Array Integration, it was simply a matter of time before that limitation was lifted. VAAI has several useful features, but one in particular is more relevant to our current discussion: VAAI offloads SCSI reservation from the hypervisor onto the storage array, thereby significantly enhancing the performance of the VM in general.

VMFS-5 now supports datastores of up to 64TB in size, thereby stripping NFS from its only strategic advantage as far as I am concerned. The other feature enhancement that is now enabled within VMFS-5 is the ability to leverage VAAI with thin provisioning and deliver automatic free space reclamation. This is huge. Prior to this version of VMFS, many did not know that when you use thin provisioning and as you start deleting data, you need to go back and reclaim that free space. Well, the problem was that it had to be done manually. And so for those who did know about the free space that could be reclaimed probably had some manual methodologies or scripts that reclaimed the free space. Now, you will be happy to know that another annoying task has been removed from your to-do list and has been automated.

Now, I am sure the last two features in the countdown are probably going to be easy to guess. So, this time a new question: Between Storage DRS and the new High Availability, which one is your top feature and why?

Posted by Elias Khnaser on 08/18/2011 at 11:27 AM


Reader Comments:

Thu, Aug 25, 2011 Brandon

You still cannot have a single file over 2 TB - 512 bytes using VMFS. You can overcome it by creating a larger RDM in vSphere 5... but until the .vmdks can actually be bigger, NFS and I suppose, RDMs are the only ways to over come the limitation. It still exists. I agree you can now have a 64 TB datastore without extents and it makes RDMs (unfortunately) more attractive.

Sat, Aug 20, 2011 Elias Khnaser Chicagi, IL

yes yes, file locking, sorry. I did not say it was a performance limitation, i merely stated that it does exist and it does cause issues, i have most definitley seen that. As far as EMC, i don't represent them but after the acquisition of Isilon, i am not sure they would splash anything but... look i am not here to knock NFS, but you threw around dedup and that is a bit misleading, Dedup is a post process that requires resources, depending on when that post process is run, a negative performance hit could be observed. Furthermore, the version of NFS that is used with vSphere even vSphere 5 is NFS3, there are some security concerns with NFS 3, spoofing and other.... i have some concerns on native multipathing etc... it is not true that IPstorgae is cheaper, completely untrue, IP Storage DONE RIGHT! means you are going to isolate your traffic on dedicated switches, dedicated NIC (HW initiators) etc... that would cost just as much as FC if not more in some cases. IP storage is not as resilient as FC. we can talk forever and this is a religious war. i think both protocols will be around forever but to say NFS instead of FC in large ESX is a bit of a strecth. FC has a ton of advantages, loseless, guarantees delivery,congestion control etc... BUT i do think NFS is also a good protocol especially NFS4. the right protocol should be chosen based on the use case and the set of circumstances. in some cases FC is better in other NFS is better, there is no size fits all, all the time. Most of the deploymenst we do in the enterprise are mixed FC and NFS, even in VMware deploymenst for larger eneterprises it is mixed. High profile, sentive and performance "ANGRY" applications still go on FC while others go on NFS. it really depends. Eli

Fri, Aug 19, 2011 Nick

I believe you are referring to file locking as NFS is a file-level protocol. I haven't seen anyone bring this up as a performance limition on the NAS side. I'm sure if it was an issue EMC would have it plastered all over the web. There are other advantages to NFS protocol for VMware, especially on the NetApp platform, such as the space efficiency gained from thin-provisioning, snapshots (without LUN overhead) and block-based data deduplication on primary storage. These savings are compounded when you have redundant data centers with data replication for DR purposes. So, how many VMDKs can you put on a VMFS-5 data store? Can you grow and shrink the VMFS file system on the fly yet? Fiber channel may be the prevailing protocol in use today but I think that's changing rapidly as iSCSI and NFS environments seem to be easier to manage with a lower initial cost to implement. I don't work for the industry, just a customer who's trying to keep up.

Thu, Aug 18, 2011 Elias Khnaser Chicago, IL

Hi Nick, but NFS also has SCSI locking my friend :) however, with VAAI all SCSI locking is now offloaded to the storage array so that should not be an issue. NFS never had and as far as i can see never will have a strategic advantage in large ESX clusters over sayyyy FC. Most larger enterprise deployments are still on FC. BTW, thanks for starting a war in the commenst section :) if you don't know what i mean, wait a bit :) Eli

Thu, Aug 18, 2011 Nick

Unless VMFS-5 eliminates SCSI reservation locking NFS still has a strategic advantage in large ESX cluster deployments.

Add Your Comment:

Your Name:(optional)
Your Email:(optional)
Your Location:(optional)
Comment:
Please type the letters/numbers you see above

> More

> More

> More

Free White Papers

More Tech Library