On 19/05/2007, at 4:05 PM, [EMAIL PROTECTED] wrote:
and another academic consideration is whether the snoopy cache
controller is write-invalidate , or write-update.
With write -invalidate, the writing processor invalidates the other
processors' copies of the item being written,
so they only know about it when they try to access it, whereas
write-update, the writing processor also sets
off an update of all the other copies. A lazy writer would prefer
not to play "nice" , so that valuable processing
time isn't spent being nice updating other processors copies for
their benefit, and just let them waste time
fixing it themselves when they need it, but since processors
shouldn't be anthropomorphised, its really
upto the designer to decide whether it makes any difference in
sales if the multiprocessor product uses
one scheme or the other.
An approachable article on reducing the number of TLB-miss and page
faults processing by the CPU /cache thrashing:
http://www.phptr.com/articles/article.asp?p=98836&seqNum=3&rl=1
On Sat May 19 5:47 , '[EMAIL PROTECTED]' sent:
yes, imagine Snoopy who is lying on the doghouse. Well ,each of
those processors need a snoopy dog
cache ; actually what it means is that the concept of snooping in
other's caches exists, so even though
the cache is yours, your neighbour is allowed to snoop in your
cache anytime.
On Sat May 19 0:29 , kuang oon sent:
On 18/05/2007, at 6:08 PM, David Guest wrote:
Which kind of leaves in a quandry.
I am trying to spec up a Xen box. It will run two debian guests
and three Windows guests. MSSQL will be the database server on
Windows and MySQL on linux. The MSSQL will be the machine working
the hardest, although I will need at least a gig of RAM for one of
the linux vboxen.
Horst, what would you recommend in terms of CPUs (2 or 4?),
hyperthreading (turned off?), RAM (8 gig?), 32 or 64 bit for this
configuration. Anything else I should consider?
If money is no object, 4 cpu, then you can try switching off 1 or 2
or 3 for speed. Since you mention so many processes, you probably
need all four. 8 because it is an inscrutable Chinese favourite
number. 64 bit if you plan to play heaps of computer chess as that
is a natural fit for computer representation of a chessboard.
Someone reported in my smalltalk list that In some multicore
scenarios a memory intensive process actually speeds up as you
throw another process at the multicore. The bottleneck with memory
intensive processes in multicore is the TLB or Translation
Lookaside Buffer which is the cache that contains fragments of
the page table, for speeding up virtual memory address translation
to physical addresses .
Kuangie
More speed, less haze, same meaning as English with Docle context
complete clinical codes.
_______________________________________________
Gpcg_talk mailing list
[email protected]
http://ozdocit.org/cgi-bin/mailman/listinfo/gpcg_talk
_______________________________________________
Gpcg_talk mailing list
[email protected]
http://ozdocit.org/cgi-bin/mailman/listinfo/gpcg_talk
