4 ms·
The point is not how much you can store and how fast you can access consecutive sectors. The problem is that spinning platters have been stuck at 100 random se
by lsb 18y ago
The point is not how much you can store and how fast you can access consecutive sectors. The problem is that spinning platters have been stuck at 100 random seeks / second for the past 20 years, and most database struggles have been based on that. The new Intel SSDs can do 100,000 "disk seeks" per second, which means that you suddenly have the speed of a thousand-disk raid on one SSD. That's why SSDs are interesting. Not the storage per disk, the random-access IO. People still use tape (not just .tar formats) because it's really cheap for dumping big backups.
- smanek 18y agoBut I don't care about random-access speed when I'm just using it for media (music, photos, video). I think within a few years most people will end up with a small (a few hundred GB) SSD for their OS and programs, and a very large traditional HD (a few TB) for archiving data that doesn't need random access. That's the system I'm using now. Two 73GB 15K drives in Raid-1 for my OS, with 3 1TB drives in Raid-z for media.
- lsb 18y agoYeah, the same separation between small assets available quickly and big assets available slowly is how Amazon recommends you separate data between SimpleDB and S3. I'm quite fascinated by all the ways we'll be able to persist data.
- smanek 18y agoYep, it isn't a new idea. The premise of Von Neumann machines is that you work around the cpu/data bottleneck with increasingly faster (and larger) caches. You have a tiny bit of ludicrously fast on chip memory (registers), a few megs of successively larger and slower L1/L2/L3 cache, a few gigs of kind of slow memory, and hundreds of gigs of unbelievably slow hard disk. This would just split the unbelievably slow disk into two separate layers - kind of slow SSD and really slow disk.