Article View: comp.sys.ibm.pc.hardware.misc
Article #194Re: Cacheing Hard Disk Controllers
From: Keith Rohrer
Date: Sat, 12 Aug 1995 00:00
Date: Sat, 12 Aug 1995 00:00
38 lines
3101 bytes
3101 bytes
In article <5raMSlpJUjB@irata.Jena.Thur.De>,
Marc Mueller <MaMue@irata.Jena.Thur.De> wrote:
>Yes, and if the OS wants to write to disk, a write-back hardware cache is
>a great speedup, a software cache is nonsense.
If the OS actually multitasks, software cache is far from nonsense, esp.
when the hard drive controller can do DMA. If you're using anything better
than Windows, the machine goes on to do any other tasks that need doing
while it waits for a slow I/O device like the hard drive, the floppies,
the keyboard, etc. Plus, the money you don't spend on an expensive
controller can go to buy RAM for the mainboard (if you don't have 32
megs), which can be used for both cacheing file I/O and keeping more of
programs in memory, and more programs in memory (making better use of
CPU resources and eliminating even the overhead of an interrupt and
context switch for gobs and gobs of page faults); or you can spend it
on more CPU horses, if you're already keeping enough programs in memory
to keep your CPU busy. And if you're truly in need of disk performance,
you're either doing enough database work that no cache smaller than your
hard drive will save you ('cept for the indexes and directory info, which
an OS or app can know to cache, but a controller probably won't, short
of an LFU algorithm) from oodles and oodles of seek time, or you're doing
something more along the lines of multimedia, where playback speed and QOS
are important enough that you either want the RAM on the mb anyway to
keep the file in memory, or you save controller overhead on every single
request for the next chunk of that huge file...don't try to tell me the
cacheing controller can do read-ahead, a drive you'd get for that purpose
will be doing read-ahead anyway. Oh, one last thing...on an OS like Linux,
with all the RAM on the MB, any running program's code which is in memory
is stored at one place in physical RAM, counting space in the buffer cache;
with a cacheing controller, it's likely in two places at once.
Keith
--
Disclaimer: Despite the fact that I disclaim any connections between the
opinions expressed here and any other person, group, or entity which has not
licensed said opinions, my opinions are in fact not connected with any other
person, group, or entity which has not licensed them.
Technical Details
Message-ID:
<40hbjb$ruv@vixen.cso.uiuc.edu>Path:
arc1-ca.pugleaf.net!unknown!not-for-mailReferences:
<3vn231$7fp@cleese.apana.org.au> <5rPFb5roUjB@irata.Jena.Thur.De> <408nqk$hjf@aimnet1.aimnet.com> <5raMSlpJUjB@irata.Jena.Thur.De>Show Full Headers
Complete Article Headers
X-Google-Language: ENGLISH,ASCII-7-bit X-Google-Thread: 10d683,87705628d8f7e5d9 X-Google-Attributes: gid10d683,public X-Google-Thread: 109e38,87705628d8f7e5d9 X-Google-Attributes: gid109e38,public X-Google-Thread: f4d7c,87705628d8f7e5d9 X-Google-Attributes: gidf4d7c,public From: kwrohrer@ta7.cs.uiuc.edu (Keith Rohrer) Subject: Re: Cacheing Hard Disk Controllers Date: 1995/08/12 Message-ID: <40hbjb$ruv@vixen.cso.uiuc.edu> X-Deja-AN: 107976372 references: <3vn231$7fp@cleese.apana.org.au> <5rPFb5roUjB@irata.Jena.Thur.De> <408nqk$hjf@aimnet1.aimnet.com> <5raMSlpJUjB@irata.Jena.Thur.De> organization: University of Illinois at Urbana newsgroups: comp.sys.ibm.pc.hardware.misc,aus.computers.ibm-pc,comp.sys.ibm.pc.hardware.storage