Messages in this thread | | | Date | Wed, 03 Nov 1999 09:53:58 -0700 | From | "Jeff V. Merkey" <> | Subject | Re: Linux Buffer Cache Does Not Support Mirroring |
| |
The less time spent holding sleep locks the better. calling lots of redundant sync code to release a bndle of IO requests is dumb if you can release them in batch. NT uses something called MDL lists that allow multiple requests for a file to be transacted accross several processors at the same time, reducing latency.
Jeff
Rik van Riel wrote: > > On Tue, 2 Nov 1999, Jeff V. Merkey wrote: > > > The Statement relates to increased parallelism on FS code paths > > (they don't have to block), and allows multiple I/O requests to be > > handled as a "bundle" rather than one at a time. The benefit is > > immediately obvious since less time is spent calling semaphore and > > sync code in the kernel (shorter code paths, less rendudant > > You seem to make a somewhat outdated optimisation here, offering > up I/O bandwidth (disk time) to save on CPU time. With the way > the processor/memory/disk speed 'balance' is going, I don't know > if that would be a tenable situation in the future. > > The "I'm idle now, gimme work" idea is a very sane idea when > processors and memory are outrunning disk speed by many orders > of magnitude (and the gap is widening on a monthly basis). > > Of course, it doesn't make much sense to give a disk it's block > when we've only got one page ready to write out, but once we > have 64 or 128 kB, it all starts to make sense. > > regards, > > Rik -- There's no seek time like idle time > -- > The Internet is not a network of computers. It is a network > of people. That is its real strength.
- 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/
| |