Keeping Your Balance in a Virtual World

Arthur Cole

Last week, I touched on the subject of storage sprawl in virtual environments, noting a number of new management approaches hitting the channel designed to more closely match available resources with the rising and falling demands of virtual machines.


As a follow-up, I'd like to take a look at some of the other ways in which the storage farm can be more closely synchronized with virtual server and networking platforms. After all, virtualization may lay the groundwork for the dynamic cloud architecture of the future, but that doesn't mean there aren't real-world issues to grapple with in the here and now.


I was drawn to the subject after reading an article by Brian Peterson, storage architect for Forsythe Technology, in which he pointed out the myriad ways in which existing storage systems can be made more virtual friendly without major forklift upgrades. His list ranges from deduplication of OS images to snapshots and SAN consolidation, but the larger point is that you need to start rethinking about rebalancing the equation between servers and storage now that a single server can host 20 or more instances.


The other main disconnect in virtual environments is on the I/O infrastructure, says Henry Newman, CTO of Instrumental Inc. Increases in memory bandwidth and CPU performance are outpacing disk and bus performance, so much so that even the fastest SSDs will have trouble keeping up. What's needed, he said, is a concerted industry effort to upgrade the entire I/O stack, from the PCIe bus to the actual storage device, plus all related software. Without it, the entire stack could go the way of the dinosaurs in favor of something new, like phase-change memory (PCM).


The need to more closely integrate storage systems with virtual environments is starting to hit storage vendors of all stripes. RAID specialist Dot Hill Systems recently dropped a cool $12 million for Cloverleaf Communications, a developer of heterogeneous storage virtualization software. The plan is to integrate Cloverleaf's Intelligent Storage Network (iSN) system into Dot Hill's RAID portfolio as the company transforms itself into a full-scale storage and solutions company.


Elsewhere on the storage management stack, virtualization has made it clear that the small-time storage node of 1 or 2 TB is not much longer for this world. Witness DataCore Software's decision to bump its virtual disk technology from 2 TB all the way to 1 PB overnight. A 1 PB LUN will enable enterprises to manage the extremely large data sets that are now commonplace, even while employing techniques like thin provisioning to keep overall storage capacity under control.


Servers and storage are like the yin and yang of Chinese philosophy. Development that takes place on one side must be countered by development on the other or the system will be out of balance.


And before we can hope to engineer architectures of the future, we must make sure our foundation is secure.



Add Comment      Leave a comment on this blog post
Jan 5, 2010 10:07 AM Puzzled Puzzled  says:

I thought phase-change memory (PCM) writes at less than 1 megabyte per second (according to the datasheet of the only PCM chip that's commercially available).  On the other hand, Fusion-io's NAND-based SSD write at 1.4 gigabytes per second, or about 1400x faster.

Looks to me the I/O stack is doing just fine, while PCM (a 1970s technology) is going the way of the dinosaurs ...

Reply
Jan 6, 2010 10:58 AM Arthur Cole Arthur Cole  says:

I think the point is that PCM could move more storage out to the edge, so there wouldn't be as much need for high-speed I/O,

But you're right, it's all speculation at this point.

Reply

Post a comment

 

 

 

 


(Maximum characters: 1200). You have 1200 characters left.

 

null
null

 

Subscribe to our Newsletters

Sign up now and get the best business technology insights direct to your inbox.