Citrix, VMware, Storage Vendors Invited To Talk
Without a doubt, desktop virtualization is the hottest topic in virtualization today. What keeps fueling this interest in desktop virtualization infrastructure technologies is the amount of use cases that you can now introduce to justify it.
Take the Motorola Atrix 4G example that I wrote about in this blog entry. There is a lot of excitement about DVI, but there are a lot of challenges with the implementation of DVI, a lot of technical challenges, such as boot-up storms, login storms, anti-virus storms ... let's just say the weather is frightful in DVI. Everyone is trying to come up with solutions to a lot of these challenges, you will see storage vendors running like chickens with their heads cut off trying to squeeze every inch of every IOPS to make DVI financially viable to deploy in enterprises.
Don't get me wrong. Storage vendors have been very innovative in some of the solutions that have come to market, such as dynamic tiering and the ability to detect and dynamically move workloads that get hot between the different tiers of disk to improve performance. Nonetheless, while the technology is cool and very promising, it cannot detect and react fast enough to avert the challenges presented by the different DVI storms we mentioned.
On the other hand, you will find that DVI vendors are also trying hysterically to come up with solutions to address these issues. Take Citrix, for instance. Its IntelliCache allows you to cache a copy of a centrally stored and managed virtual disk locally on the hypervisor host and then stream that image to all the resident VMs on that host. In theory, this sounds great -- you are now using cheap local disk, caching a copy of the centrally stored and managed VHD, you avoid boot-up storms and login storms, etc. ...
Sounds perfect right? Not quite. In a world moving more and more toward cloud computing, does Citrix really expect us to build these islands of siloed hosts that stream to locally resident VMs? What if a particular host is experiencing heavy utilization and I wanted to migrate some VMs? What if I want to dynamically live migrate VMs in order to recalculate load balance? Am I expected to ask the user to log out and log back in? This is supposed to be cloud computing. I want automation, and Citrix's Simon Crosby is one of the most enthusiastic people about cloud. So, how does this fit in?
IntelliCache is cool, but with all due respect, local disk is dead. Find me another solution where I don't have to give up any of the features or flexibilities gained with virtualization. When building DVI, a lot comes into play from an architecture and design perspective. What if I want to use Affinity rules to make sure that certain VMs are never present together on the same host? What if I want to separate VMs across different hosts in different blade chassis? Since in large deployments, it is inevitable to use blades.
So what is the answer? How about storage vendors and DVI vendors create a task force, a group of smart people that can sit at a roundtable and explain to one another the challenges? Instead of the storage guys trying to find metrics and trying to understand DVI, and the DVI guys trying to understand storage and building technologies around it, how about, hey, storage guys, we have this issue, we can give this data, how can we resolve it? Storage vendors have dynamic sub-LUN tiering, but it is not enough for DVI as it does not react fast enough. Great, can the DVI folks maybe provide more information so that the storage vendors can build technologies that can possibly detect a certain pattern and react quickly when they see this pattern? Maybe the storage array can detect a signature on the data that causes login storms or bootup storms and can move them immediately to a faster tier disk for processing.
Can we for example, enhance VAAI and StorageLink so that they can give the array more information, which would allow it to react faster? What if XenDesktop or VMware View had APIs similar to VAAI and StorageLink that tie directly into the array? In the event of a login storm, could XenDesktop or VMware View slow the amounts of VMs being powered on or logged into based on how much the array and the underlying storage can handle?
The bottom line is, virtualization created a marriage between storage vendors and software vendors that will not be dissolved anytime soon. So, while the technologies have married and have children already, the in-laws are still not getting the message that they need to get along.
I invite Citrix and VMware CTOs Harry Labana, Simon Crosby and Steve Herrod to form these task forces that can reach out to the EMCs, HDSs, HPs and IBMs of the world and talk to their storage folks, give them data they can use so that when they build these systems, they can support the innovations that you are developing.
Posted by Elias Khnaser on 01/18/2011 at 12:49 PM