[comp.sys.m68k] Adding memory to MVME147 / Unix kernel tweaking

peter@stca77.stc.oz (Peter Jeremy) (08/23/90)

We have a 8 MB Delta 1147 which is showing signs of running out of memory
and I am currently looking at ways of alleviating this situation.  It uses
an old (pre-S) MVME147 with mezzanine DRAM.

The problems are:
a) Motorola never developed high-capacity mezzanine boards for the 147
   (although the addressing logic can support up to 32MB), and now that the
   board has been superceded by the MVME147S it is unlikely they ever will.
   Although a 16MB 147S board has been announced, it is not supported as
   a system product and no upgrade path from 147 to 147S boards has been
   announced.
b) The 147 does not support VSB.  Accessing a normal A32/D32 VMEbus RAM
   board takes ~3 times as long as on-board DRAM accesses, and is
   unlikely to provide much benefit.  In addition, the kernel uses low
   memory, thus placing user processes in slower memory.

The possible solutions I can see are (in descending order of preference):
A) Upgrade to a 141 or 161 (this would be nicest, but justifying the cost
   would be virtually impossible).
B) Find a 3rd party mezzanine board.
C) Move some of the large kernel data structures (disk buffers, streams
   buffers and the like) into VMEbus RAM.  This would result in a system
   with a higher disk buffers to user memory ratio than optimum, but
   should provide an overall improvement in performance.
D) Add VMEbus RAM and page into it.  The only problem is that there is no
   support for multiple-level paging and paging activity is shared evenly
   across all paging devices.  (Also, R3V5 didn't allow paging onto RAM disk).
E) Put up with the system as it is.

My questions are therefore:
1) Has anyone else had this problem?  If so, how did you solve it?
2) Does anyone know of a third-party manufacturer of high-capacity 147
   mezzanine DRAM boards?
3) 
4) Does R3V6 support paging into RAM disks?  (Under R3V5 the kernel would
   panic if the paging activity became non-negligible).

Also, whilst looking through kernel data structures, I noticed a relatively
large (~128K) initialised data block named `c37x_memory'.  Presumably this
is the download code for the MVME374.  Since I don't have a 374, I would
prefer to have the 128K as user process space, just on principle.-- 
Peter Jeremy (VK2PJ)         peter@stca77.stc.oz.AU
Alcatel STC Australia        ...!uunet!stca77.stc.oz!peter
240 Wyndham St               peter%stca77.stc.oz@uunet.UU.NET
ALEXANDRIA  NSW  2015