Date: Fri, 01 Oct 93 16:54:15 +0900
To: vsta@cisco.com
From: Jonathon Tidswell <jont@hsa.com.au>
Subject: /@USER@/bin ( was Re: Latest stuff )
> From: Andrew Valencia <vandys@cisco.com>
[There is some logic behind separate partitions: /, /var, /usr, /home, ... ]
What is the logic behind separate disk partitions?
Would they be needed if unix had a better way to backup subsets of the
file system and disk quotas that worked?
Not attacking but trying to understand...
Chris
ps On fast IPC, the canonical paper is "Lightweight Remote Procedure
Call" by Bershad, Anderson, Lazowska, Levy in ACM TOCS (transactions
on computer systems) v8 n1 Feb 1990 p 37-55. The most significant
implementation of LRPC is the "Quick LPC" technology in Windows NT.
LRPC works by setting up a shared memory region at binding time before
any calls are made. An actual call consists of copying args to the
shared memory region and then calling a lightweight control transfer
function (such as a semaphore) to context switch to the destination.
The performance win is because of the shared memory region: the kernel
doesn't have to copyin, validate, and copyout this data.
As far as the IPC in Spring, I'm not sure why it's so fast. I have
two possible explanations. They use sparc register windows to avoid
saving and restoring cpu state. I imagine they save a lot of memory
traffic and get much better cache locality this way. But it's not
clear if you will get this kind of peformance in real life. Real
applications also do a lot more than sit in a tight loop bouncing IPC
messages off each other. (I'm looking at a copy of "The Spring
Nucleus: A Microkernel for Objects" by Hamilton and Kougiouris, Sun
Labs TR SMLI-TR-93-14. This might have appeared in the 1993 Summer
Usenix as well.).
This PS turned out to be longer than I thought...
Received on Fri Oct 1 01:10:36 1993
This archive was generated by hypermail 2.1.8 : Wed Sep 21 2005 - 19:37:12 PDT