Virtualization is all the rage. Youve heard the estimates: Between 70 percent to 90 percent of data center resources are underutilized.
Given the waste, the buzz around virtualization makes sense. It helps by pooling and sharing resources, allowing users to automatically grow and shrink resources based on application requirements.
Its no panacea, though. Attendees at Gartner Inc.s recent Symposium/ITXpo griped about virtualizations introduction of an untenable mountain of virtualized servers that make for management headaches. Plus, theres the issue of software licensing complexities, with the industry by and large still stuck in a per-CPU, per-user mindset, not yet adjusted to the more fluid dynamic presented by virtualized resources.
Hewlett-Packard Co. has long been a big backer of developing and bringing virtualization to its customers. More importantly, it also has approaches to working with the management and licensing headaches of virtualization.
Database Editor Lisa Vaas recently talked with HPs Nick van der Zweep, director of Virtualization and Integrity Server Software, to get the dope on how HP is trying to soothe the pain points of this up-and-coming technology. Van der Zweeps work includes developing HPs first-in-class HP Utility Data Center and HP Virtual Server Environment offerings. He also spearheads all virtualization efforts on behalf of HP Technology Solutions Group and is responsible for software marketing for HP Business Critical Servers, including HP-UX 11i, OpenVMS, High Availability software and the HP Virtual Server Environment.
Were hearing that a dangerous new problem is virtual machine sprawl—uncontrolled flexibility that is worse than useless.
Once you get into virtualization, the [technology] does things like take one server and run multiple applications on it at [the same] time. Its done through partitioning, on high-end systems, with virtual machines. With ProLiant [its done in] partnership with VMWare and Xen and Microsoft. But like [Gartner Symposium IT/Xpo keynoter Tom Bittman] said, you may have 10 servers but it looks like 100 because you have different operating systems running at the same time. If you say, No, I dont want to do it the old way, I want the new way, youll have 10 physical servers but 100 virtual servers.
The good thing is if server No. 10 or No. 95 is running at 100 percent, well, tough luck. You need less hardware because of virtualization, but now youve got to handle the complexity of 100 virtual machines sitting on top of 10 physical machines, and how do you handle that in management systems so they understand it? Or does it make you do something goofy, like create 200 virtual machines?
[HP has] software out there like System Insight Manager, which is really important to our whole strategy. [It allows users to] understand both the physical and virtual machines, and it covers a heterogeneous [mix].
Virtualization is great, but just watch out for the management complexity. Thats why weve been focusing on making the management simpler and preventing this great asset, to be able to reduce costs, add agility and prevent that from becoming a management nightmare.
Whats next on the agenda when it comes to the virtualization and utility computing portfolio?
Well always add more technology down at the low end to make it more powerful and simpler. We just added the capability to go to partitions that are 1/20th the size of a CPU. You can have many, many more virtual machines on a [server]. And were still adding virtualization to our storage.
But were really, really focused on the management area.
Virtualized storage is something that people are waiting for, before taking the leap to virtualization. Whats HP doing in that area?
Well focus on moving up, continuing to focus on more management software around virtualization and more automation software around virtual resources to automatically move CPUs around. We want to expand that [to allow users to] automatically allocate and de-allocate storage and storage cache. We already announced automation to automatically add [resources such as] a blade [server] on demand, so an application is running hot but you can get not just another CPU but another blade.
We do a lot of, when you start talking about vision for utility computing, youre going to a mode where more and more infrastructure is flexible and it automatically moves to the right application. Companies are doing that on an application-by-application basis, but eventually the entire data center will be virtualized.
We have sets of customers that have done that already. In most cases weve implemented these capabilities in multiple HP data centers around the world. We do a fairly large outsourcing business and extend our capabilities, not just technologies but our people who know how to run this and manage this.
Youre talking now about customers such as DreamWorks, right?
Companies like DreamWorks have taken advantage of this.
What they do is when they make a film like “Madagascar,” they have a lot of compute power to render all the frames. What they did was they said, Its not their core value proposition to be an IT company and sit there and run all those IT systems. They only need that power intermittently.
What they do is use a utility-rendering system HP has.
DreamWorks is able to use that service and submit their frames to us. We render them and ship them back fully rendered and digitized. They pay on a per-frame basis.
And we have SLAs as to what turnaround time we promise them and what we get paid.
How about the complexities of licensing when it comes to virtualization?
Licensings an interesting one. In September we announced more virtualization-friendly licensing from HP. Weve got multiple ISVs that are looking to move in the same direction as us. That is a big issue. As you automatically move resources around, you need a flexible licensing scheme as well. One minute you may need 100 CPUs of capacity running in an Oracle database, but in another minute you may just need 50. What kind of licensing do you pay when your usage has peaks and valleys?
Traditionally most software companies go on physical machine and size of box and go for the max utilization that could happen.
If you could use 100 CPUs of Oracle database, theyll charge you 100.
But were seeing things change.
At Oracle, they do more per-CPU licensing. So they handle partitions a little bit. At first they had the situation where if you took a machine and wanted to virtualize its eight CPUs, theyd want to charge you for eight CPUs.
But Oracle changed, and that was a good step. So licensing is changing out there. It needs to change to be a lot more flexible to handle this virtualization, where things ebb and flow.
What ISVs are you working with to change this?
Were working with multiple ISVs. We have this capability to meter usage. We already server-virtualized systems and storage. The more they use, the more they pay, and the less they use, the less they pay. We do that to our customers, so multiple software companies are looking at how we do that, our metering software, and asking should they do that. We have beta programs. Were not at liberty to say with whom, but we have them at multiple sites now.
Thats going to get interesting.
And SLAs, how is HP dealing with that in a virtualized environment?
What we have is within our virtual server environment, the ability to simply add in these SLAs, where you say, I need subsecond response time for this application, or 30-minute batch turnaround. And then you prioritize applications as well. Say you need OLTP subsecond response time but you have it at priority 1, because customers are using the OLTP database but internal folks are using [something else that gets a lower priority ranking].
Nobody else in the open systems space has anything like this.
Check out eWEEK.coms for the latest utility computing news, reviews and analysis.