[9fans] Ghost in the linker?

[9fans] Ghost in the linker?

[Eric]

forsyth wrote:
> 
> >>8l gets to 40Mb during the link!
> 
> that part is normal with ghostscript, as i recall.
> i had to relink the old version and i'm fairly sure i
> needed to do it on a 64mb cpu server.

I've compiled gs3.33 in 32mb on a pcdisk and it
pages pretty heavily.  Compiling parts of libtiff
also page on my old 32mb config.

I've been wondering what the 'typical' installation 
does (err, should do?) in cases where the cpuserver 
starts to page heavily.  It seems to me that w/o a 
paging disk on the cpuserver (or terminals paging 
heavily) there'll be an inordinate amount of traffic 
on the network.  Separate networks for cpu<>fs 
activity could help but are annoying to manage.

I think one could just put gobs of memory in the 
cpu and 'plenty' in the terminals and just take 
the hit in the rare occasion memory runs out... it's 
mostly pretty cheap these days, particularly compared 
to cheap disks that provide really too much storage 
for this task (plus hassle of having to maintain them).  
Paging doesn't happen often, either.

OTOH I don't really know what the paging statistics
look like for a cpuserver with a large user base; maybe 
a paging disk on the cpuserver is a win?  (then again,
what is 'large'?  50?  1000? ).

Eric Dorman
edorman@tanya.ucsd.edu

[9fans] Ghost in the linker?

[Eric]

> 
> edorman@Tanya.ucsd.edu (Eric Dorman) writes:
> | .. it's mostly pretty cheap these days, particularly compared 
> | to cheap disks that provide really too much storage 
> | for this task (plus hassle of having to maintain them).  
> 
> You could use them with cfs. 
>
Indeed;  What I'd really like to get at, however, is
having as few disks as possible (fewer moving parts
to go bad, configure, buy and generally worry about :) ), and 
those disks that do exist are in controlled environments.

Seems to me that would be useful in 'development' environments
where a broader spectrum of files are used more often, and
less useful in environments (e.g production, offices, etc )
where files (applications) are longer-lived and fewer in number.
I feel intuitively that the paging loads follow a similar
pattern, but I don't have any data to back it up :)

Looks like I need to write some activity simulations and
see what happens :)

Eric Dorman
edorman@tanya.ucsd.edu

[9fans] Ghost in the linker?

[Nigel]

Yes, 40Mb didn't surprise me at all. I don't have a swap partition
available
at the moment, but I do have 64Mb. It takes about 20 seconds to produce
the
duff executable.

>-----Original Message-----
>From:	forsyth@caldo.demon.co.uk [SMTP:forsyth@caldo.demon.co.uk]
>Sent:	Thursday, February 19, 1998 8:10 PM
>To:	9fans@cse.psu.edu
>Subject:	Re: [9fans] Ghost in the linker?
>
>>>8l gets to 40Mb during the link!
>
>that part is normal with ghostscript, as i recall.
>i had to relink the old version and i'm fairly sure i
>needed to do it on a 64mb cpu server.

[9fans] Ghost in the linker?

[schwartz+9fans]

edorman@Tanya.ucsd.edu (Eric Dorman) writes:
| .. it's mostly pretty cheap these days, particularly compared 
| to cheap disks that provide really too much storage 
| for this task (plus hassle of having to maintain them).  

You could use them with cfs. 

[9fans] Ghost in the linker?

[forsyth]

>>8l gets to 40Mb during the link!

that part is normal with ghostscript, as i recall.
i had to relink the old version and i'm fairly sure i
needed to do it on a 64mb cpu server.

[9fans] Ghost in the linker?

[Eric]

> 
> Before I dive where I've never been before, anyone know of any
> limits on segment size in the linker?
> 
> Specifically I linked Ghostscript 5.10 last night (the changes are
> minimal to make the standard release compile under APE), and
> got a very large executable (?4.6mb) which page faulted at address 0.
> 8l gets to 40Mb during the link!

Cool.  I'll have to try and give that one a whirl after I swap
boxes around tonight. 
  
Eric Dorman
edorman@tanya.ucsd.edu

[9fans] Ghost in the linker?

[Nigel]

Before I dive where I've never been before, anyone know of any
limits on segment size in the linker?

Specifically I linked Ghostscript 5.10 last night (the changes are
minimal to make the standard release compile under APE), and
got a very large executable (?4.6mb) which page faulted at address 0.
8l gets to 40Mb during the link!

"file gs" returns "386 plan 9 boot image".

In the words of Lord Gnome "shurely shome mishtake".

Any suggestions appreciated.