[linux-elitists] ME roadshow

Eugene Leitl eugene.leitl@lrz.uni-muenchen.de
Wed Aug 30 12:38:46 PDT 2000


Paul J Collins writes:

 > Not a wrapper; it's a Linux super-server, much in the style of the BSD
 > super-server that NeXTstep uses on top of Mach.  Running Linux-L4

It works, without too high penalty on latency, doesn't it?

 > doesn't buy much from what I can see.  You're adding some overhead for
 > no discernible gain.
 
If I have a system with a million processors _and_ a small memory
grain size (say, about 1 MByte, or less) I certainly can't afford a
500 kByte kernel. All I need is something which lets me launch and
terminate code, and lets pieces of code in other nodes talk to each
other (most likely, this will be implemented by an opcode, with
interCPU switched matrix taking care of the matter).

 > I guess it's all perspectives.  But I doubt that your two kilobyte
 > Forth kernel provides everything that your fat Linux kernel does.
 
Wrong attitude. Why has everything to stay in the kernel? I should be
able either to swap code in, or refer to code residing in other nodes
by passing messages.
 
 >   $ ls -l /boot/vmlinuz-2.4.0-test7
 >   -rw-rw----    1 root     root       587871 Aug 28 19:49 vmlinuz-2.4.0-test7
 > 
 > What on earth do you have in your kernel?

It was a quick build in a hurry a while ago, and I almost never used
modules. I have 256 MBytes RAM on a legacy box, on that kind of system
+/- 500 kBytes is of little relevance. On an embedded, or a Beowulf
with small node grain size even a kByte extra means a lot.




More information about the linux-elitists mailing list