Messages in this thread | | | From | Greg Lindahl <> | Subject | Re: Ken Thompson interview in IEEE Computer magazine (fwd) | Date | Fri, 7 May 1999 22:49:37 -0400 (EDT) |
| |
> What exactly do you mean by parallel? There's pretty much two clean ways to > do it. One system has access to the physical disks that hold the filesystem > or multiple systems have direct access to the physical disks. NFS/SMB/Coda > use the former. Sequent's CFS and the similar product for > Digital/Sun/HP/IBM use the latter. In the former you are limited by the > performace of the system that has access to the disks. In the latter you > are limited only by the speed of the disks and the interconnect (to the > disks). I can tell you without even thinking too hard which is faster. :)
How about the 3rd way: a cluster of systems, each having local access to their own disks. You are limited only by the speed of the network to the systems, and how fast each can talk to its local disk. This is how Legion does it, and some other parallel filesystems built on top of things like NFS (although CIFS would be better).
I assure you that my cluster with gigabit networking and 64 fast/wide SCSI busses is quite fast at doing parallel I/O. And if you have all the data mirrored to a different node, you can do rolling upgrades.
This has slid off topic. But there's more than one way to skin a cat.
-- g
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |