Re: Rep:Re: [f-cpu] little feed-back from the libre softawre meeting

[whygee@club-internet.fr]

hi !

>De: "Nicolas Boulay"
>De: whygee@club-internet.fr

>>>From discussion with a Hurd guy (Neal Wafield)
>hmmm... speaking about God and its saints .. ;-)
>we started speaking about IPC yesterday...
>
>>     16 kb for a page is a little bit to much
>why ? if you have huge datasets (megabytes, gigabytes...)
>you will spend your time in the trap handler...
>i count on the OS to "gather" contiguous small pages into
>larger ones, possibly based on usage statistics (that the
>CPU could gather).
>
>>>> For neal, the probleme came when a task ask for a very large amount
>of memory you gie him a large page but then he will not use the all
>memory so it could be good to catch all the unused memory. That's could
>only be done with smaller page size.   

i'm currently in the FreeBSD room ...
they gave me an indication :
http://docs.freebsd.org/cgi/getmsg.cgi?fetch=332902+0+archive/2002/freebsd-alpha/20020707.freebsd-alpha
which contains another link :
* http://shimizu-lab.dt.u-tokai.ac.jp/lsp.html
* http://www.cs.rice.edu/~jnavarro/mitosis

>>     8 kb could be enough
>>     all new processor have between 5 to 11 page size !
>cool.
>however too many sizes might make the LSU more complex :-(
>i guess that 4 is the best tradeoff.
>
>>>> i-64 give 4, 8,16,... kb page size (11 size !)
it will depend on the possibility to have several sizes...

>>Conclusion :
>>     implemented L4
>>     change specification of the tlb (no more the kind of caches use
>>     but an 8 bits fields)
>
>8-bit field if you want but will this solve virtual
>aliases problems ?
>
>>>> That's the sw problem ! you need bigger number than the number of
>line in the tlb (8 bits === maximum 255 line in the tlb)

hmmm ..... that's another problem...


>my current method is to have N parallel TLBs with 
>one size each. Say, 8 entries for 4KB, 8 entries for 32KB,
>8 entries for 256 etc...
>
>>>>That's few !!
let's start with something that works first, no ?

> It will be better to avoid so much different table but
>maybe it's the only solution.
>nicO (that go to see the RMS show ;p)

time to wake up :-(

*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/