MAC II ROM Upgrade

[Dr. Robin Lake]

In article <981 at aimt.UUCP> breck at aimt.UUCP (Robert Breckinridge Beatie) writes:
>In article <1558 at pt.cs.cmu.edu>, ralphw at IUS3.IUS.CS.CMU.EDU (Ralph Hyre) writes:
>> In article <8972 at apple.Apple.Com> phil at apple.UUCP (Phil Ronzone) writes:
>> >Yes, I've have seen the National Semi boards working. I use one myself.
>> >I have seen 5 boards in one system (88 megabytes of memory) running.
>>
>> Neat!  So, has anyone tried porting the BSD ramdisk driver to A/UX?
>> That's one way to sort of get more free space elsewhere, since you could put 
>> /tmp and multiple swap areas on multiple boards. ...
>
>Swap to ram?  I like the idea of /tmp (and /usr/tmp for that matter) on
>a ram disk.  And then the system might benefit from having the pipe device
>be a ram disk.  But swap space?  You're decreasing your available memory,
>which causes more swapping.  It seems that just keeping that memory available
>to programs would eliminate swapping almost entirely.
>
Experiments with UNIX and a solid-state disk [Sugit Kumar's Ph.D. dissertation,
Case Western Reserve University, late '70's] showed that the best use
of a zero-latency disk with UNIX was for /tmp, /usr/tmp, and commonly-
used system programs.  The reason that there is less gain with using
it for a swap device is that there is only ONE rotational latency, then
the whole swap area is written out to a contiguous disk file.  On the
usual file system files, you can incur a number of rotational latencies
per file.

Of course, I don't know how the AUX driver and hardware really work, nor
what the interleave factor is on the hard disk.  These would affect the
results Sugit obtained, as he was working with a word-addressable
disk.

More details upon demand, if any.
-- 
Rob Lake
BP America Research and Development
decvax!mandrill!nitrex!rbl
mandrill.CWRU.EDU!nitrex!rbl