Real World Computing
The promise of petabytes
To start creeping up on that petabyte target - whether you're on OS X or Windows - you'll need a slightly quicker kind of interconnect technology than IDE or SATA cable and, in the case of the VTrak (and hence the XServe, although this technology is eminently transferrable to PCs, too), the connection is made via Fibre Channel. This is a comms technology that's been associated for a long time with Enterprise-class server farms: you've probably seen the steady parade of Fibre Channel servers, drive-boxes and intermediate interlink hardware being rated and reviewed in PC Pro's Business section. It's a long-established technology, and the apparently low speed ratings of the interlinking connectors offer no guide to the effective performance you can expect from it, provided you have enough money.
Well forget all that now, because this modern kind of Fibre Channel shares only the name and the chips that load and unload at either end of a cable with its expensive older relative. But equally, don't commit the late-night Ebay scourer's classic error of thinking that sharing the Fibre Channel name guarantees interoperability between different products. Fibre Channel isn't like SCSI, where you can merrily plug a U320 drive into an 80MB/sec controller. The petabyte punch of the VTrak and XServe are achieved because "Fibre Channel" in this context refers to a dedicated connection medium, a copper conductor in the very simplest configuration, between one server and one storage enclosure - VTrak pushes the standard's envelope to the limit with a 4Gb/sec connection between server and enclosure.
Really, when you're looking at such systems you should ignore the interlink speeds, because what really matters is that at either end of that link is a lot of processing power devoted to shipping bytes of storage around. The Promise unit mixes and matches SATA and SAS drives inside the same enclosure, and then you can profile fast and slow storage, all within the same bit of rack space. This may sound like a crazy idea at first - what's the point of keeping any slow storage, or of mixing it with the fast? I think that should be pretty clear, because I've had a few terabytes hanging around in my basement for a couple of years now - and believe me when I say there's no greater temptation than that deadly phrase "What the hell, I've got room..." Even a few terabytes of storage can bring out your worst self, the one who swivels on his heel the second that progress bar has started to move and slopes off down the Chinese for Special Seafood Noodles. The management overhead for terabytes of typical business information rises very sharply indeed once you try to figure out which copy relates to which activity, and when you made that copy. This problem is an entirely human, not machine, one related to how coherently you name your data sets and how you structure them into ever-larger volumes. Even if Apple feels it's cracked the petabyte barrier, that's no automatic guarantee that us lesser mortals will be able to follow it over there.
Basic server; software RAID
Charity clients are difficult for the consultant because they typically present the worst-case scenario for a would-be server specifier: they need a machine that does everything, but costs nothing. This may sound do-able if you believe the hype that we live in some Nirvana of infinite performance and limitless disk space, but the fact remains that you can get yourself into a world of trouble by using consumer-grade computing hardware to do business-grade jobs.
