Mixed Signals
Any sufficiently advanced information is indistinguishable from noise
Friday 11 July 2003, 3:06 PM
Friday
Friday 11/7/2003According to Jim Gray, head of Microsoft's Bay Area Research Centre, the best way of transferring lots of data is to put it on a hard disk, stuff it in bubble wrap and post it off. Storage has always led data communications in bang per buck, and this has always been true -- in fact, we may be about to take a step backwards in transmission speeds when it will no longer be possible to fill Concorde with DVDs.
The ever-improving cheap hard disk has caused changes elsewhere. It used to be true that the fastest, cheapest and least fussy way to back up a large system was to plug in a larger tape drive. Not that it was actually cheap, unfussy or fast, but it beat the alternatives. Nowadays, the smart cookie just plugs in a spare disk, copies everything across and leaves the duplicate on a shelf somewhere.
So why is this still so hard? Where are the well-designed front-panel mounted removable hard disk systems? They used to be everywhere, and if you look back at the preferred PC specifications issued by Intel, Microsoft et al in the late 90s there was even a specification for such things. Backups continue to be the most important, least observed ways to protect your system, and we've finally got a cost-effective, simple and efficient way of making them -- so where's the industry support? It's not as if there's any engineering to do, as serial ATA is hot-pluggable, and it can only improve the sales of hard disks, so its continued absence as a standard part of everyday computing is beyond mysterious.
As for costs: Gray goes further. For maximum convenience, he doesn't just send the hard disk, he sends a complete computer. Even that's cheaper than the alternatives, and at the far end the recipient just has to plug in gigabit ethernet and hoover off the data in short order. Gray also points out that with drives reaching towards the terabyte we will have to treat them as tape drives, not random access devices: taking a typical database model where you read stuff off chunks of a few kilobytes at a time, he points out, the slow speed of the disk seek means a terabyte disk will take a year to read completely.
That's nothing. I started Finnegan's Wake in 1994 and I'm still not halfway through.
Click here to see more of Rupert's diaries.
Previous
