[comp.arch] Cache flush penalty at context switch.

dwc@cbnewsh.ATT.COM (Malaclypse the Elder) (02/01/90)

In article <1062@yarra.oz.au>, chris@yarra.oz.au (Chris Jankowski) writes:
> By the way affinity scheduler will help you only if:
> 	- there is enough processors to manage.
>         - there is typically more processes running than processors.
> 	- caches use direct mapping based on physical addresses.

i don't understand the need for direct mapping.  i always thought that
n-way set associative generally resulted in a higher hit rate (cycle
issues aside).  and can you get away with virtual addresses if you add
a process id or context id flag as part of the key?

danny chen
att!hocus!dwc

chris@yarra.oz.au (Chris Jankowski) (02/02/90)

In article <7846@cbnewsh.ATT.COM> dwc@cbnewsh.ATT.COM (Malaclypse the Elder) writes:
> In article <1062@yarra.oz.au>, chris@yarra.oz.au (Chris Jankowski) writes:
>> By the way affinity scheduler will help you only if:
>> 	- there is enough processors to manage.
>>         - there is typically more processes running than processors.
>> 	- caches use direct mapping based on physical addresses.
>
> i don't understand the need for direct mapping.  i always thought that
> n-way set associative generally resulted in a higher hit rate (cycle
> issues aside).  and can you get away with virtual addresses if you add
> a process id or context id flag as part of the key?

	I should have been more precise.
        I should say:
        There are cache designes that cause horrible penalty
        everytime you start a new process.
        I know of one (now obsoleted) design where you had
        to invalidate all entries in instruction cache in
        every processor everytime you started new process.
        The penalty is proportional to the number of processors
        and the rate you generate new processes.
        This penalty may be in some cases more significant
        then any gain you can get from affinity scheduler.
        But, yes it will help in this case as well.
	You are perfectly right.

      -m-------   Chris Jankowski - Senior Systems Engineer chris@yarra.oz{.au}
    ---mmm-----   Pyramid Technology Corporation Pty. Ltd.  fax  +61 3 820 0536
  -----mmmmm---   11th Floor, 14 Queens Road                tel. +61 3 820 0711
-------mmmmmmm-   Melbourne, Victoria, 3004       AUSTRALIA       (03) 820 0711

"Knowing how things work is the basis for appreciation,
and is thus a source of civilized delight."  -- William Safire