Real World Computing
The promise of petabytes
One of the bigger thrills when we first became interested in computers was that sense of becoming a part of the future. Before 2000 at least, everyone in the business seemed to be looking forward - it was all about progress, progress, progress (okay, I'll admit I had one friend called Ishmael Skyes who carried a flintlock musket and marched through mud at the weekends, but all my technologist friends were forward-looking). This sense of excited futurism formed the basis of a lot of old-school science fiction and, no matter what your current opinion of the worth of that now-oversubscribed genre, there's no doubt those guys were fearlessly looking forward, too.
I was particularly reminded of the works of Rudy Rucker while sitting in a demonstration suite in Apple's new European HQ in Munich earlier this month. The Apple folk were doing a little hands-on session with XServes and XSAN2 - while explaining why they'd given up on manufacturing the XServe RAID drive enclosure - and in the course of the session they casually mentioned that the maximum volume size for XSAN2 is four petabytes. Now, the first time I came across this enormous unit was in Rucker's "Bopper Robot" trilogy, in which he thought it a number so huge that he could use it to stand for the quantity of information required to create an autonomous artificially intelligent life form. One petabyte is 1,024 terabytes (one terabyte is 1,024 gigabytes, one gigabyte is 1,024 megabytes and so on down).
By my reckoning, the complete works of Shakespeare occupies about 3MB in zipped plain-text format, so Apple's largest logical disk volume can now store 50 copies of the Bard's canon for every single human on the planet. I doubt that we'll all be rushing out to order one straight away, though, because this system requires a completely separate XServe, running just as its dedicated "storage processor", and adding up the number of hard drives you need to add to reach that magic number is pretty sobering. Simply assembling 1,024 one-terabyte drives won't actually be enough, because in a collection that large you can count on at least five of the drives giving up the ghost every week: let's be prudent and say it makes sense to employ 1,200 drives to give us sufficient fault-tolerance margin, all of them hot, online and spinning. Based on the drive mechanisms I'm buying at the moment, that little lot should draw two megawatts of power and output the same as heat (and that's assuming that the drive enclosures are 100% efficient, which they never are). Is your head hurting yet? Even allowing for the replacement of disks by solid-state storage, or for the now-traditional tenfold increase in storage sizes over the lifecycle of any server platform, four petabytes is a chunky number and represents a considerable vote of confidence by Apple in the future of its server platform.
This upbeat event was a complete turnaround from the start of the month, when Apple announced the death of the XServe RAID and prompted the usual prophets of doom to forecast complete market withdrawal, collapse of product range, and humiliation for all those proposing to disturb the peace and equilibrium of the server marketplace. Apple's salvation, so the session I attended informed us, lies with a vendor whose name I'd never have guessed before the event - to wit, Promise.
My exposure to Promise has been long-term but low-key, as the vendor of add-on IDE and SATA controller cards for the cheaper end of the market. My general rule concerning Promise's products, whenever I've found these cards anywhere near a server doing real work, has been to treat them as evidence of profound cheapskatery on the part of the supplier or network administrator. It's perfectly possible to build your own server using a bottom-end or last-generation PC plus a Promise card of any variety - I've done it myself, and the performance can be impressive with some careful picking and choosing - but the fact remains that users of Promise have always been, to put it as nicely as I can, extremely "price-sensitive", or exactly the opposite of the target audience for the VTrak SAN enclosure.
