🚀 go-pugleaf

In <22...@chorus.chorus.fr>, m...@chorus.chorus.fr (Adam Mirowski) writes:
|Oliver Chung (khc...@pacific.mps.ohio-state.edu) wrote:
|
|| Your RAM estimates are quite wrong.  Not counting the diskcache,
|| HPFS only adds about 200K memory overhead.  Quite important in
|| DOS where every byte of conventional memory is precious, but not
|| so for 32bit OSes which have 8MB RAM to allocate.  You are going
|| to have one diskcache anyway, so whether it's a VFAT or HPFS
|| diskcache is irrelevent.
|
|What are you talking about? Disk cache in Windows 95?  This is an
|obsolete OS concept you still find in OS/2, as OS/2 is an old OS
|design. Windows 95 does not have a specific "disk cache". Neither do NT
|or modern versions on Unix.

I see, so in W95 or NT, you have to bear the cost of cache expansion
and contraction. This is an improvement?

|In Windows 95 all of your memory can serve to cache disk files. You do
|not have to reserve memory in advance. It is all dynamic and managed by
|the virtual memory system.

Great, so now the VM mgr gets to worry about disk cache too? Wow, what
an improvement! Disk cache in main memory is mostly a microcomputer
strategy, as small machines do not have the I/O bandwidth or sophisticated
hardware I/O architectures to make disk access fast enough. BTW, that
microcomputer assessment includes UNIX which has the unfortunate quality
of not being able to scale sufficiently for use on large scale systems.

|One of the most interesting consequences of that fact is that you can
|map files into memory, instead of accessing them with read/write
|calls. This way you obtain the best possible performance, as only one
|byte of RAM is used to cache one byte of a file at any given time and
|there is zero overhead in accessing that byte: it is in your address
|space. Of course, OS/2 does not allow you to map files into memory, as
|its disk cache is an opaque entity without relation with the address
|space of applications.

Memory mapped files and main memory disk cache are _not_ the same thing.
Show me _any_ OS that manages main memory in one byte increments....
BZZZZZT, sorry time's up; but, thank's for playing anyway! The day that
there is zero overhead to access something in a cache is the day we
won't need computers, 'cause pigs will be a flyin' and we'll all be in
Heaven (well, some of us...). As for the _system's_ cache being "in your
address space", well, that's called "the dangerous memory model" and they
call it that for a reason! One of the _major_ architectural draw backs
and security holes in W95 is exactly that design, giving R/W access to
much of the OS to _all_ processes.

|And HPFS is not a replacement for VFAT as it neither supports Unicode
|nor allows DOS applications to see long file names.

DOS? Are you still running DOS? Oh sorry, I almost forgot, W95 _is_ DOS
or least the graphical shell... Unicode? Oh yeah, just dying for that
one, don't know how I've done without for - let's see - close to 15yrs!
VFAT is no replacement for HPFS as VFAT is slow and subject to fragmentation
and _hugely_ space inefficient. I've got a 1300MB HPFS partition at work,
just how big would my _cluster_size_ be in VFAT? Get a clue, first you
try to talk about "modern" OS's, then you try to claim superiority for
VFAT! What's next, edlin as a modern editor?

|Adam Mirowski,  m...@chorus.fr (FRANCE),  tel. +33 (1) 30-64-82-00 or 74


Phil "Guido" Cava   TeamOS/2
Help, PC!
Let us help you achieve Warp Speed today!
email at: pcg...@helppc.com
> Computers are useless, they can only give you answers! - Pablo Picasso