[rescue] Spark10 CPU question

Sandwich Maker adh at an.bradford.ma.us
Wed Dec 14 20:27:20 CST 2016

" From: Dave McGuire <mcguire at neurotica.com>
" On 12/14/2016 02:43 PM, Sandwich Maker wrote:
" > " From: john ferguson <jferg977 at aol.com>
" > " 
" > " 
" > " I'm running an old (but great)  CAD program on an SS10 using SunOS 
" > " 4.1.4.  I also have a SPARCPrinter 1, which also works and have compiled 
" > " a 2006+/- flavor of Ghostscript to accept more modern versions of 
" > " PostScript output and print them.  And, no, I didn't really know what I 
" > " 
" > " []
" > " 
" > " to the question:
" > " 
" > " My memory is that there is no benefit for multiple processors in 4.1.4, 
" > " and it seems unlike the application I use the most would benefit either.
" > 
" > there is, for heavy cpu-pig apps like cad, even if they're
" > single-threaded and under sunos4's primitive 'multiproc' - the app can
" > substantially hog only one cpu, but that leaves the other free to do
" > everything else a running system wants.
" > 
" > i saw this at a network-gear shop in the mid '90s; workstations with
" > only one proc were virtually unusable while doing sims, but ones with
" > dual procs were only a little slowed in the logic editor by the
" > 'background' task.  sims went faster too, since they were swapped less
" > with their 'own' cpu.
"   The big issue here is, as I understand it, is the lack of fine-grained
" locking in SunOS4.  There's one big lock going into kernel space, so
" only one processor can execute code in kernel space at a time.  Solaris
" has very VERY fine-grained locking, so lots of processes/threads can be
" running around in the kernel simultaneously.  In fact, if memory serves,
" each instance of a device driver also gets its own thread in Solaris,
" making things even smoother when there are lots (dozens) of processors
" involved.

that matches my understanding.

the giant spinlock seems like rather a crude hack for multiproc
support.  surely sun could've done better - but they didn't have to; i
was at another job that had a solbourne about the time they were going
under, and i thought it was a shame that their work on finer-grain
spinlocks in the sunos kernel was just landfilled [along with other
interesting things].  but sun was understandably not interested in
anything which would slow customers in moving from sunos to solaris,
especially when they were moving much slower than they wanted anyway.

"   How much of a performance benefit you get from multiple processors
" really depends on the workload, and how much of it involves heavy
" processing in user space vs. making frequent calls to the kernel for
" I/O, etc.

my understanding also.
Andrew Hay                                  the genius nature
internet rambler                            is to see what all have seen
adh at an.bradford.ma.us                       and think what none thought

More information about the rescue mailing list