Virtualization has finally arrived at the urban yet bucolic campus where I spend my weekdays (and occasional weekends). I'm talking about the full deployment of a big and shiny new server with eight cores and 64 gigabytes of memory, which is further connected to a hardware RAID with over ten terabytes of storage. Nice, yes? Well, perhaps it is, but it is not without issues.
Just the thought of acquiring a big new server set our school’s excitable IT staff into planning mode. For the next year whiteboards in multiple locations were filled with hand-drawn maps of existing servers and their paths to the Really Big Server. After all, everyone else seemed to be virtualizing everything, and our folks were anxious remain on the cutting edge. Their belief was that virtualization would solve a number of problems, ranging from the need to replace old hardware that was well past its prime, to reducing the number of power supplies in a given co-located rack.
The latter is not a joke; the university’s on-campus IT department decided that the best way to encourage schools to reduce their colos’ demand for power was to establish a baseline of AC power consumption in their yearly contracts. They set this baseline to a value that was below the schools’ current usage, and then made it prohibitively expensive to move up to the next tier of power. Our school’s yearly bill for juice was going to skyrocket from $5000 a year to nearly $20,000; the only alternative was to eliminate some servers, along with those computers that acted as pseudo-servers, that had collected in our racks over the years. We faced a similar fate regarding the number of static IP cable drops allowed in each rack.
Fiscal plans were also made. Money was set aside, and our IT staff got to do some industrial-grade shopping for a solution. Vendors were consulted, bids were gathered, and decisions were made. The new server and storage solution arrived last August, but it took until late January of this year for the new server to be commissioned and the experiments begun. Along the way, our plans hit a snag — Apple left the server market altogether, leaving the XServe that services the computer lab high and dry. OSX Server 10.6 has now become a candidate for eventual virtualization.
I was aware that the servers used by my classes were being virtualized on the new hardware box, although I was not aware that I would become an unknowing guinea pig for an impromptu test today. Beginning about 9:30 a.m., I lost contact with the new virtual servers. In particular the Windows server which authenticates Mac users in the computer lab seemed to disappear, leaving a lab class full of students unable to reach their personal shares used in class. A call to the IT director for the school revealed that he had decided to install additional memory and a second NIC card in the server, taking it down in the process.
When I called him to see what he had to say, he explained that he knew the big new server would need a reboot, and he decided to see how long it took. The memory check for 64 gigabytes of memory added an extra seven minutes to the boot schedule, and the new shiny server was unavailable for nearly a twenty minutes. “Oh, sorry,” he calmly and dispassionately remarked, “I wondered if you were working on anything, but I wanted to see how long it would take to boot.” It took about fifteen minutes.
It would appear that we are putting a substantial number of eggs into the virtualization basket, and should the server ever lose power we’ll be able to get it back online in somewhere around fifteen minutes. That doesn’t count however long it takes to boot the individual virtual machines. Add to that the fact that while we have plenty of ports available inside the virtual machines, in the real world there’s only one Ethernet cable to share amongst all the servers, which is not helping our network throughput at the moment.
I’m sure at least some of these issues will be addressed, but given the lack of reliability in today’s hard drives and my unease about losing virtual servers because the power was reset and the server needed a re-boot, I’m wondering whether virtualization is the great idea the vendors said it was. For the moment, color me skeptical.