| sean finney on Fri, 2 May 2003 13:49:08 -0400 |
|
On Thu, May 01, 2003 at 03:26:10PM -0400, gabriel rosenkoetter wrote:
> If I'm going to want to read the whole of a piece of data from
> beginning to end every time I use it (do we use *any* data like this
> on computers any more?[1]), then it makes way more sense to have it
> on tape, because it's cheaper and I'll probably get it faster, since
> the disk will probably have to service someone else's random access
> request before I'm done reading.
>
> [1] Legitimate question. I don't think we do. I certainly don't.
i remember from my summer internship at the lawrence berkeley labs
nersc hpcf, they had this sweet storage system. it has 18 terabytes
of disk, which is used as *cache*(!) to a 2 petabyte tape storage system.
if your stuff isn't in the cache, the seek latency kind of sucks...
iirc it was used for storing massive amounts of data for/from parallel
experiments on the supercomputer or other cluster systems[1]. i got
to see the tape robots at one point. the entire unit was about the
size of my bedroom, and it had a radon tank sitting on top of it
for worst case scenarios.
sean
[1] and i think these experiments probably don't do more than a
sequential read/write of the data as far as storage i/o goes.
Attachment:
pgpp4PyFe9zMF.pgp
|
|