Subject: Re: solaris disk I/O
To: Wojciech Puchar , <tech-kern@netbsd.org>
From: Lars Heidieker <lars@heidieker.de>
List: tech-kern
Date: 09/10/2001 08:47:52
Thats new in Solaris 8 the fscache is put on the standby memory queue so it 
will be reclaimed befor paging out anything else,
this would mean in terms of netbsd to put fscache(only) pages on the 
inactive queue...
This would stop the fscache to grow over 1/3 of physmem if memory is short 
and it would be easly shrunk to 1/3 if more memory is needed for other 
tasks. (It still can grow to full memory size if the memory is free)

Lars

At 01:29 AM 9/10/2001 -0700, Wojciech Puchar wrote:
>i've installed x86 solaris. while it's graphical tools (while i don't know
>now how to turn off while installing, and how to configure devices) it's
>terribly slow, java-bloated etc., OS itself looks fast.
>
>especially disk I/O. i found that while it buffers well small file I/O on
>it's dynamically resizable cache, it doesn't bloat memory with large file
>I/O. and doesn't "stall" when lots of writes are pending like netbsd.
>
>filesystem looks like ffs (it's called ufs there), but i found
>cylinders/group to be set to 1000 - value impossible for netbsd.
>
>could such behavior (large/small files) be somewhat implemented? i've
>tested -current integrated VM/disk cache and it's almost unusable. copying
>500MB files means wiping out everything from memory to swap.
>like in windoze.
>it's bad to see netbsd is worse in new version.