On Wed 14 Oct 1998, Drake Diedrich wrote: > d) Try using NFS. This slows down the i/o bound resource hog enough to > leave the machine usable for interactive tasks. Yes it's ugly, but > scheduling in 2.0 is suboptimal, and nice doesn't have much effect on i/o, > only CPU. An extra disk dedicated to the i/o hog would probably be better > than an SMP for this problem.
I've noticed that if something is doing heavy IO over a _large_ range of data, what happens is that RAM gets tied up in the buffers for the accessed data. Try something like "dd if=/dev/hda of=/dev/null bs=500k &" to see what I mean. It's unfortunate that there's no way to tell the kernel that the data being read sequentially, so there's no point in keeping that data around in the buffer cache after it's been passed to user-land... Paul Slootman -- home: [EMAIL PROTECTED] | work: [EMAIL PROTECTED] | debian: [EMAIL PROTECTED] http://www.wurtel.demon.nl | Murphy Software, Enschede, the Netherlands
