VMWare: "Hey, where'd all the servers go?"
One BS Bingo term I learned from Scott Huston was 'information sharing seminar', which IIRC involves one person who goes to one of those free-tech seminars to return and write up a simple report sharing what they learned. This is distributed some how to the rest of the department, and facilitates sharing information and skills rather than hoarding it.
I hate hoarding information, so naturally this was right up my alley.
I still have my notes from the Microsoft Public Developers Conference from a couple months ago to put together, but in the mean time here's my impression of the VMWare Virtualization seminar I attended in Seattle today.
The seminar was split into two, maybe 2.5 sections.
1: A virtualization overview of what VMWare is doing in the field right now;
1.5: A brief, clumsy overview of the features of those really expensive EMC storage solutions you can by for virtualized environments;
2.5: A fascinating Q&A with a Washington Mutual bigwig who had real knowledge and experience implementing large scale virtualization of servers and, the interesting part, virtual desktops.
Kevin Connor from VMWare began by talking about market cap, market share, number of employees, main product offerings, blah blah blah. He was smooth and fun to listen to, peppering his talk with words like potty, groovy, and my fav "a pillar to hang your hat on." He explained some of the leading edge stuff that VMWare is doing, like Vmotion. Neat stuff.
I couldn't stand the EMC guy and I really had to potty; when I got back, I'd missed it all.
By then the WaMu guy (sorry, didn't catch your name man) was talking. They'd started out with an impossible project, migrating 1500 call center workstations to two offshore locations on a tiny budget where no information could be housed outside US borders and do it in 45 days. They did it by virtualizing WinXP desktops and hosting them, and they did it under budget and under time.
I can believe that. One of the real benefits of virtualization that I can see is provisioning a new machine, and I'm not talking about IT work. I mean if you have to order a new machine, you need to spec it out, write up a cost justification, figure out who's budget it comes from, set it up as a capitalized expense (depreciated over three years), then order it and wait for it to arrive. Making a new VM? An hour to copy a template and run sysprep, assuming you've got the server available already.
Once nice thing about the WaMu guy was that he gave numbers and configuration examples. He's building VM farms out of 4-core, four socket machines. He sets up six of these things, and hosts 100 to 120 XP VM's. Nice advice I have no reason to doubt: No matter the scale of your farm, no matter the multi core or what, don't go more than 8 nodes. If you scale out that broad, you simply cannot recover from a node failure very flexibly. This makes sense to me on an intuitive level, i.e. eggs and baskets.
Ultimately, this is the first time I've heard real practical examples of utility computing in the enterprise. All the other times it's been research scientists doing the equivalent of mainframe timeshares back in the day. This guy was building out a farm, and when various departments called with a new project or server request (and budget to back it up), they'd carve out a VM from the pool of resources. Then they'd charge that department for it. While saving money from power and hardware. Faster. In fact, funny story, when they'd get a request for a new server or workstation, they'd intentionally delay the availability by a day. Internal customers thought they were getting gipped if the IT department gave them same day turnaround. By articifically delaying it 24 or more hours, by comparison, internal customers decided that the IT dept must have drop shipped the new hardware. For some reason, I find that really funny.
Lastly, I like that he did a full cost recovery model for their first VM farm. I'm interested in a similar exercise for the college. We've been using GSX, as it's free. ESX starts at just $500 for a nice hypervisor. I bet it makes a nice difference.
Labels: technology, virtualization, work
This is a post about how I solved an obscure iTunes/iPod problem; I'm only writing it so it's hopefully googleable and can help someone else. And cripes - I should've taken a screen shot. Anywhatski, this is the message I'd been getting, periodically, when trying to sync my new-ish iPod Shuffle (the 1 GB gen 2 one):
