Visiting VMWare Virtualization 2010

"Up, up, up the stairs we go until we come to...the tunnel!"

Gollum's words from the Lord of the Rings films (Lord of the Rings, Wikipedia) echoed in my head as I walked the long hallways and sparse escalators of the Metro Toronto Convention Center. I followed the ambiguous, temporary signs that had been there for years but had never been corrected or replaced, promising the South Building was somewhere vaguely ahead. (Did nobody watch Rory Sutherland's discussion on the importance of signs in his talk on "Sweat the small stuff" (TED Talks)?) It was hard to believe that world leaders would be meeting here in just a few days for the G20 summit.

Luckily I knew the way. I had been to the South Building before and today I was on my way to the VMWare Virtualization Forum 2010 Toronto conference.

"You think there's any giant spiders here, Mr. Frodo?"

"Sam," Frodo would reply with an affectionate chuckle. "Don't be silly...Arrugh! It's a marketer! I'm done for, Sam. Take the ring and go on without me!"

By the time I reached the exhibition floor (descending an equal number of spare escalators), my legs were tired from the hike. In front of a small number of harried registration clerks a lineup of more than 100 people snaked randomly between the columns in front of Hall G as if to further test the endurance of anyone who made it that far. Computer professionals in T-shirts, jeans and backpacks checked emails on the iPhones out of boredom, the odd guy in a power tie trying to look like he had the skills to be in line. An even smaller number of women, perhaps 1 out of 20, waited but looked far more at home that the guys in suits.

Someone showed off his iPad, cradled awkwardly in his arm, which looked liked a iPhone designed for the visually impaired. It was a reminder that this was a conference with an agenda: VMware and its business partners were here to sell their products. They weren't going to talk about the risks, the costs or the alternatives. It was up to the attendee to use his (or, 1 in 20, her) good judgment and be on guard against exaggerated claims.

Once I got to the front of the line, I had to confirm my registration on one of the old laptops provided for that purposes. The clerks assembled badges and handed them over. They informed me that attendance was seriously underestimated: there was not enough seating available and I was advised to go early and get a chair where I could.

The secret to survive a room full of marketing people with slick hair and slide-show presentations is to come with healthy skepticism and a smile, tell a joke or two, and get whatever freebies you can. Head immediately for the vendor booths before the best stuff is gone. This year the best stuff was 4 GB flash drives. However, I wanted to hear the keynote addresses so I went there first.

The word "virtual" has been in computers for a long time. It refers to something that appears to be one thing but is actually something else. One example is virtual memory: using a computer's hard drive to simulate more memory than the machine contains. Software that virtually represents old computers and operating systems have been around since hardware began following Moore's Law (Moore's Law, Wikipedia). Two of the most ambitious and successful emulators are MAME, the Multi-Arcade Machine Emulator, which simulates the hardware used by more than 8,000 coin-operated video games and runs on virtually every operating system. The other one is Wine, the Windows emulator, for UNIX-based systems, able to run everything from µTorrent to Runes of Magic on Linux.

When desktop computers gained multi-tasking operating systems that could run more than one program at a time, these emulators could run in separate windows on the desktop. If you wanted to, you could run your old version of AppleWorks (Appleworks, Wikipedia) within Linux or Microsoft Windows. Eventually emulators, each with its own user interface, gave way to standard front-ends like Sun-now-Oracle's VirtualBox, the same way MAME provided one interface to many game hardware emulators.

"Virtualization" nowadays refers to simulating multiple, complete computers on a single machine. And it's not about running that old copy of WordStar (WordStar, Wikipedia) on a CPM emulator (CP/M, Wikipedia). It's about saving money through server consolidation: programs running on a simulated machine can be moved as easily as taking a screenshot and uploading it to a different computer. Setting up a new server no longer requires a complete O/S install and configurating software by hand. Now you copy and paste.

This risks and costs to this cost-saving strategy include competing applications fighting over resources, larger hard drive and backup requirements compared to the alternatives, having a hardware failure take out several applications at once and "virtual sprawl" (keeping track of the use and evaluating the usefulness of a particular virtual machine).

Consolidation through virtualization is only a practical long-term strategy while Moore's Law holds true. There have been signs that Moore's Law has begun to struggle as chip makers have been unable to come up with low-cost ways to improve CPU speed (because faster speeds would melt the current chips, Tom's Hardware), relying on multiple cores (the poor man's multi-processor) which appear to support Moore's Law but having less value for the average grandmother's desktop than speed increases. More concerning, at the International Supercomputing Conference 2010, there were signs that Moore's Law had finally begun to falter for the newest designs, even with the cheat of multiple cores (University of Hamburg Lectures).

This year the advocates of virtualization had their head in the clouds: expounding how their products would make cloud computing possible. Cloud computing (Cloud Computing, Wikipedia), using other people's computers when they are not in use to use for your own programs, is still mostly science fiction. There are many political, legal and financial obstacles and, even if these were solved, not all programs are practical for running in such an environment. The VMWare spokesperson announced that if virtualization allowed programs to move between machines, then it could also allow programs to move to and from the (non-existent) cloud. VMWare would become the "Microsoft" of the cloud. However, with free virtualization like Xen, it seemed to me that VMWare was trying to fight Xen the way Microsoft tried to fight Linux in the server market. Even if companies truly believed that selling their computer time was easy money (which it isn't), why would they invest in expensive commercial products to make it happen.

When the speakers began using terms like "private clouds" (i.e. your business' data center), I think the crowd had a collective "oh, come on!" moment. "Private cloud"...like private Internet, easy money, or "simple yet powerful"...were oxymoron terms.

A speaker suggested virtualization would be the "framework" of the future of the Internet, comparing it to website frameworks. It seemed to me that Linux (and Xen) made a pretty good "framework" on the existing Internet and it would do just as well in the cloud, if it should come to pass.

The speech from a Telus manager had a more realistic vision. In their case, running thousands of Linux web site computers, virtualization allowed the web sites to be easily swapped from one real computer to another...or, in theory, even clone a hard-hit website across several actual machines. Even so, it's hard to imagine virtualization as a silver bullet. Although mathematics made it possible to maximize the average use of hardware, every ISP has peak periods and dry periods, and to provide good performance for peak periods, you still need a total hardware capacity equal to the challenge of the peak. The Telus theory sounded good but it didn't mesh with the reality that everyone was checking their favourite sites at 9 am and no amount of on-the-fly consolidation would solve that kind of problem.

The repeated call to "virtualize everything" seemed to me to be like Java's call that "everything is an object" (which is not true, "Objects Have Failed", Dreamsongs; "Everything is not an Object", National Technical University of Athens). When you have a hammer, everything looks like a nail.

A free box lunch was provided by IBM, which managed once again to deliver three kinds of food I hated. I wondered if this move, as I had seen at other conferences this year, reflected the number of out-of-work or under-employed IT professionals who couldn't afford a decent lunch.

During the break between sessions, I toured around the vendor booths. However, most of these products had little to do with virtualization, other than the assurances that they were VMware-compatible, which seemed a little silly since wasn't the goal of virtualization absolute compatibility? And none of the ones I saw had any tie-ins to cloud computing.

Near the end of a session, I leaned over to the journalist sitting next to me. "Too much marketing..."

"...and not enough meat," he finished.

It seemed to me that virtualization had a lot of practical uses: testing software against other operating systems, temporary environments, running software on obsolete systems. Using virtualization for consolidation was a mixed success, depending on one's business. Using virtualizating in cloud computing was much more dubious. It was clear that virtualization was a useful tool for certain types of businesses and applications and, as such, it was here to stay. Linux was also a preferred operating system for virtualizing many applications, including web sites.

Where open source seemed to be following behind was in virtualization tools. Xen was a free solution for virtualizing Linux but major players like Telus needed tools to manage not dozens but tens of thousands of virtual machines. VMware has vSphere. Monitoring and statistics tools like Nagios and Cacti exist for open source, but configuration was difficult, they didn't scale well to large data centers, nor do they provide a clean and consistent interface. Perhaps what was needed was a solution that combined these separate technologies into a complete and integrated product, like my old System Manager in a Box project was aiming for before the development team disbanded. The need that SMiaB was trying to fill was still there today and it was being driven by virtualization.

With the conference wrapping up, I needed to head back to Union Station. Maybe taking the Skywalk would be faster. After 20 minutes of long corridors and concrete stairs with ambiguous signs promising the end was near, I began to doubt the promises would be delivered before my legs gave out.

June 19, 2010