Sorry! This was, naturally, my April Fools prank for 2009 🙂
At the same time, with Linden Lab forging strong alliances with IBM, Intel, Sun, Novell, and even to a degree, with Google (which powers the in-world search through a Google Appliance Server), a notable absence has been felt: Apple. Rumours even claim that Steve Jobs has once commented that Apple would never be in Second Life, “because there is no money in it”.
Things definitely have changed. In a recent interview, Linden Lab’s Frank Ambrose, Senior VP Global Technology, and Joe Miller, VP Platform & Technology Development, jointly announced their plans for the new infrastructure that will be supporting the newest generation of Second Life simulators.
Data centres are usually ugly piles of cables all over the place, with a mess of different hardware solutions, all untidily arranged, forced to fit on racks or enclosures where they were never meant to fit. All data centres usually start neatly and pretty much organised; however, as time passes, new servers are placed on the old racks, which won’t fit in the same space — newer machines are usually slimmer and take less space, though, leaving some gaping holes where you cannot fit machines inside… so it meant opening each rack, taking all servers out, and place them back again, hoping to get things arranged. Doing that while the whole grid is running is always a daunting task, even taking into account that Linden Lab has plenty of spare servers to deal with these kinds of moves.
Linden Lab has gone through 5 different technology platforms — from Class 1 to Class 5 — and you can very well imagine that each time things get more and more messy, specially because LL buys all their servers and doesn’t lease them. Obsolete servers pile up on corners, but are still used for test grids or low-usage services. No wonder, though, that after a decade of operations a new solution needed to be implemented, to accommodate future changes, while at the same time increasing rack density — which will save LL precious money on co-location bills.
So the solution seemed to be obvious. You needed a class of servers that is fast enough to deal with the current demands of the simulator software. They had to be small — very small! — to allow more servers per rack. They have to be low power