[rescue] ss2 under load

[Kurt Mosiejczuk]

On Thu, 14 Feb 2002, Dave McGuire wrote:

> On February 14, Joshua D Boyd wrote:

<snip>

> > Thanks.  That clears things up considerably.  So, which machine had the 64
> > contexts?

>   These are from memory, some may not be accurate:

Filling in Dave's table from my book... (and Dave, you WERE
accurate from memory =) )

pre-mbus machines
-----------------
4/100           16  <== The one Dave actually got wrong
4/200           16
4/300           16
4/400           64
SS1(+),SLC,IPC  8
ELC,IPX         8
SS2             16

sun4m machines
--------------
SM100	        4096
SuperSPARC      65536
SuperSPARC II   65536
microSPARC	64      (Classic and LX)
microSPARC II	256     (Voyager, SS4, SS5)

Now, the MMUs in the first part all seem to be "the same" (8 or 16 contexts).
However, there are lots of differences in number of PMEGs, pages/PMEG and
"total VM"...

Now... as additional info the pre-mbus machines use the "Sun-4 MMU" that I
described previously.  The mbus machines use the "SPARC Reference MMU".
It keeps it's TLBs in memory, instead of on-chip like the Sun-4 MMU.
Thus why they have so many more contexts.

Different tradeoffs.  The SPARC Reference MMU is more traditional...
And tended to work better for big workloads that would keep overrunning
the "cache" of hardware contexts in the earlier processors.

--Kurt