why won't write-behind stay disabled?

"Sandi Hardmeier - MVP" wrote in message
...
"cquirke (MVP Windows shell/user)" wrote in
message ...
On Fri, 20 Jan 2006 23:19:25 -0000, "Noel Paton"
"Jeff Richards" wrote

NOW you've kicked over the antheap! :-)

YAY!!! Here comes Chris! sitting back with fresh popcorn



I ain't playing this game! g

--
Noel Paton (MS-MVP 2002-2006, Windows)

Nil Carborundum Illegitemi
http://www.crashfixpc.com/millsrpch.htm

http://tinyurl.com/6oztj

Please read http://dts-l.org/goodpost.htm on how to post messages to NG's

On Sun, 22 Jan 2006 04:59:45 GMT, (Olive) put
finger to keyboard and composed:

All I know is using the principles in this article sped up internet
speed of my Win98 system dramatically. The principles, paraphrased
from the article, must be done in order--1, 2 then 3:

1) Start with your data link layer and make it solid (eliminate com
port overrunns).

COM port overruns are usually of insufficient frequency to be
noticeable, unless you have an old 486 with a 16450 buffer-less UART.

Enable FIFO.

Yes.

Reduce rate setting your com port advertises to modem.
For example, set modem speed to 56K but set comm port to
38K.

This is extremely bad advice. Doing this effectively restricts your
modem to a throughput of only 38400bps, ie 3.8k bytes/sec. You need to
select the maximum port rate possible for your hardware, usually
115200bps. In this way you will be able to take advantage of the
modem's data compression capability. A typical compression ratio for
text based data is 2:1 for V.90 and 3:1 for V.92. This means that your
port rate must be at least double your modem-to-modem data rate,
otherwise your throughput will be throttled by the COM port.

Make sure you have enabled error correction and data compression in
your DUN connectoid, and also enable software compression if your ISP
supports it. The latter will compress the data before it is received
by the modem.

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Sun, 22 Jan 2006 19:42:26 GMT, (Olive) put
finger to keyboard and composed:

In a nutshell the link between drive write-behind and internet surfing
is this quote "Anything that prevents your CPU from responding quickly
enough to interrupts from your UART can cause overruns."
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

True. However, certain types of modems, eg soft and controllerless
ones, are immune to overrun errors. Most internal hardware modems
(including my Rockwelloid but excluding some USR models) are also
immune.

My understanding is that (Com port) overrrun errors is a data "layer"
problem which slows data flow between "layers" and eventually slows
data transfer between your modem and internet servers.

Have you actually seen any overrun errors? I've only ever seen one,
and that was on my slow socket 7 system during a complete virus scan.

Your modemlog will show you the total number of serial overrun errors
during the last dialup session.

Modemlog:
http://www.modemsite.com/56k/modemlog.asp

Your PPPlog will show other error types including buffer overruns and
CRC errors:

PPP log:
http://www.modemsite.com/56k/ppp.asp

Above the data
"layer" is the PPP/SLIP "layer". And above that is the MTU "layer."
The top two layers depend on the bottom data layer to have as few if
any com port overrun errors.
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

The impact of overrun errors is usually insignificant. In the
following test I was able to achieve a throughput of 12-19kBps without
a single overrun error, even with the UART FIFO buffer disabled.
Others have achieved 40kB/s with an internal controllerless Lucent Win
Modem.

Internal modem / FIFO buffer / throughput test:
http://groups.google.com/group/comp....e=source&hl=en

Sandi, as for my definition of of internet speed, all things being
equal, a 56K modem should at best give you an average speed of 7.0
Kbytes per sec (K/sec).

It's a little more than 8 bits/byte if you allow for error correction
overhead.

Test results - compression and error correction:
http://groups.google.com/group/comp....e=source&hl=en

My average seldom went beyond 2.3K.sec.
Before all my tweaks I regularly saw 1.6 to 2.3K/sec for an internet
speed. After tweaks I saw 2.3 to 4.6K/sec.

The figure of 4.6K/s is impossible. Your COM port limits you to
3.8K/sec. You are probably seeing the result of rolling averages, or
some such artifact of your browser.

The regulars at comp.dcom.modems recommend System Monitor
(sysmon.exe). This ships with Win9x. It is accessed via Start -
Programs - Accessories - System Tools. It gives instantaneous and
cumulative receive and transmit byte counts and error counts. Any
other product that purports to give the same information is probably
snake oil.

That's double! And I
haven't even disabled drive write-behind yet. I hope disabling
write-behind will further increase my average internet speed.

Not a chance. You *may* be able to improve your throughput by reducing
the error rate of your modem. In certain cases you may be able to do
this by limiting the modem's top speed.

Limiting CONNECT speed:
http://www.modemsite.com/56k/x2-linklimit.asp

In any case you need to query your modem's last call diagnostic
report. This will tell you about the quality of the phone line (the
main limiting factor), Tx/Rx error rates, initial/final Rx/Tx speeds,
retrains, rate renegotiations, etc. You can view the report using
HyperTerminal (which ships with Windows).

Setting Up Hyperterminal:
http://www.modemsite.com/56k/x2-hyperterm.asp

Using Hyperterminal:
http://www.modemsite.com/56k/usehyper.asp

Modem Diagnostics:
http://www.modemsite.com/56k/diag.asp

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

Snake oil.

--
__________________________________________
Sandi - Microsoft MVP since 1999
http://www.ie-vista.com
http://inetexplorer.mvps.org

Inetexplorer has changed - for instructions on how to
find old URLs, go he
http://msmvps.com/spywaresucks/archi.../14/46971.aspx



"Olive" wrote in message
...

Sandi,
In a nutshell the link between drive write-behind and internet surfing
is this quote "Anything that prevents your CPU from responding quickly
enough to interrupts from your UART can cause overruns."
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

My understanding is that (Com port) overrrun errors is a data "layer"
problem which slows data flow between "layers" and eventually slows
data transfer between your modem and internet servers. Above the data
"layer" is the PPP/SLIP "layer". And above that is the MTU "layer."
The top two layers depend on the bottom data layer to have as few if
any com port overrun errors.
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

Sandi, as for my definition of of internet speed, all things being
equal, a 56K modem should at best give you an average speed of 7.0
Kbytes per sec (K/sec). My average seldom went beyond 2.3K.sec.
Before all my tweaks I regularly saw 1.6 to 2.3K/sec for an internet
speed. After tweaks I saw 2.3 to 4.6K/sec. That's double! And I
haven't even disabled drive write-behind yet. I hope disabling
write-behind will further increase my average internet speed.

But here is full quote that links write-behind to internet surfing.
I'm sure you can decipher because you're trained.
Me? It's mostly over my head.

http://www.cerberus-sys.com/~belleis...aq/overrun.htm
"Another cause is poorly written 32-bit disk drivers that aren't
WD1003-compatible (needed for Windows' caching
software to work properly), and which lock-out lower priority
interrupts (like com port interrupts) for an inordinately long time
while they dump-to-disk a large write-behind cache. While awaiting
longer term fixes by upgrading disk/drivers/BIOS, you can
get temporary relief by turning-off write-behind caching.

NOTE: Windows uses a Terminate-and-Stay-Resident (TSR) program
for disk-caching called smartdrv which is loaded by your
autoexec.bat file. Add the switch /X to turn-off write-behind
caching. Windows for Workgroups uses a VxD called VCACHE,
ignoring smartdrv except for floppy disk drives. Write-behind
caching for VCACHE is turned-off with a line in the [386enh]
section of system.ini that says ForceLazyOff=C (or =CD if you
have two hard drives) with no spaces and no : after drive
letters.

A fully compatible disk driver (like Western Digital's WDCTRL.DRV for
its Caviar drives, or Ontrack Software's Drive
Rocket) will enable Windows for Workgroups to use both 32-bit file
access (with a VxD called VFAT) and 32-bit disk
access which bypasses the DOS disk interrupt services through a
Digital Protected Mode Interface. This provides much faster
disk reads and writes to allow more time for handling com port
interrupts. "

Nice to meet you Frank. An excellent post.

--
__________________________________________
Sandi - Microsoft MVP since 1999
http://www.ie-vista.com
http://inetexplorer.mvps.org

Inetexplorer has changed - for instructions on how to
find old URLs, go he
http://msmvps.com/spywaresucks/archi.../14/46971.aspx


"Franc Zabkar" wrote in message
...
On Sun, 22 Jan 2006 19:42:26 GMT, (Olive) put
finger to keyboard and composed:

In a nutshell the link between drive write-behind and internet surfing
is this quote "Anything that prevents your CPU from responding quickly
enough to interrupts from your UART can cause overruns."
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

True. However, certain types of modems, eg soft and controllerless
ones, are immune to overrun errors. Most internal hardware modems
(including my Rockwelloid but excluding some USR models) are also
immune.

My understanding is that (Com port) overrrun errors is a data "layer"
problem which slows data flow between "layers" and eventually slows
data transfer between your modem and internet servers.

Have you actually seen any overrun errors? I've only ever seen one,
and that was on my slow socket 7 system during a complete virus scan.

Your modemlog will show you the total number of serial overrun errors
during the last dialup session.

Modemlog:
http://www.modemsite.com/56k/modemlog.asp

Your PPPlog will show other error types including buffer overruns and
CRC errors:

PPP log:
http://www.modemsite.com/56k/ppp.asp

Above the data
"layer" is the PPP/SLIP "layer". And above that is the MTU "layer."
The top two layers depend on the bottom data layer to have as few if
any com port overrun errors.
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

The impact of overrun errors is usually insignificant. In the
following test I was able to achieve a throughput of 12-19kBps without
a single overrun error, even with the UART FIFO buffer disabled.
Others have achieved 40kB/s with an internal controllerless Lucent Win
Modem.

Internal modem / FIFO buffer / throughput test:
http://groups.google.com/group/comp....e=source&hl=en

Sandi, as for my definition of of internet speed, all things being
equal, a 56K modem should at best give you an average speed of 7.0
Kbytes per sec (K/sec).

It's a little more than 8 bits/byte if you allow for error correction
overhead.

Test results - compression and error correction:
http://groups.google.com/group/comp....e=source&hl=en

My average seldom went beyond 2.3K.sec.
Before all my tweaks I regularly saw 1.6 to 2.3K/sec for an internet
speed. After tweaks I saw 2.3 to 4.6K/sec.

The figure of 4.6K/s is impossible. Your COM port limits you to
3.8K/sec. You are probably seeing the result of rolling averages, or
some such artifact of your browser.

The regulars at comp.dcom.modems recommend System Monitor
(sysmon.exe). This ships with Win9x. It is accessed via Start -
Programs - Accessories - System Tools. It gives instantaneous and
cumulative receive and transmit byte counts and error counts. Any
other product that purports to give the same information is probably
snake oil.

That's double! And I
haven't even disabled drive write-behind yet. I hope disabling
write-behind will further increase my average internet speed.

Not a chance. You *may* be able to improve your throughput by reducing
the error rate of your modem. In certain cases you may be able to do
this by limiting the modem's top speed.

Limiting CONNECT speed:
http://www.modemsite.com/56k/x2-linklimit.asp

In any case you need to query your modem's last call diagnostic
report. This will tell you about the quality of the phone line (the
main limiting factor), Tx/Rx error rates, initial/final Rx/Tx speeds,
retrains, rate renegotiations, etc. You can view the report using
HyperTerminal (which ships with Windows).

Setting Up Hyperterminal:
http://www.modemsite.com/56k/x2-hyperterm.asp

Using Hyperterminal:
http://www.modemsite.com/56k/usehyper.asp

Modem Diagnostics:
http://www.modemsite.com/56k/diag.asp

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Mon, 23 Jan 2006 15:37:57 +1100, Franc Zabkar
put finger to keyboard and composed:

On Sun, 22 Jan 2006 19:42:26 GMT, (Olive) put
finger to keyboard and composed:

My average seldom went beyond 2.3K.sec.
Before all my tweaks I regularly saw 1.6 to 2.3K/sec for an internet
speed. After tweaks I saw 2.3 to 4.6K/sec.

The figure of 4.6K/s is impossible.

Actually, it is impossible for an external serial modem but not for an
internal modem.

Your COM port limits you to 3.8K/sec.

This is true for those modems that have a real COM port, but not
necessarily so for those that have virtual one, ie soft or
controllerless types.

Sorry for the error.

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Sun, 22 Jan 2006 16:04:07 +0800, "Sandi Hardmeier - MVP"

1) Is it possible that Cacheman is messing around with your settings - it
runs in the background, yes?

Ahhhh yes, that is quite likely!!

3) I will ask why you can't update your BIOS.

I would be very, very reluctant to ever recommend a BIOS upgrade...
- risk of failed upgrade leaving you with no BIOS
- risk of breaking the software installation's assumption base
....as there's no ways I'd want to be responsible for fall-out I might
not be able to fix even if I had my hands on the machine.

If BIOS *really* breaks, the only fix is to replace the chip.
Soldered in, as it usually is? Replace the mobo. Can't get that type
of mobo, e.g. old, proprietary or laptop? Replace the PC.

Software breakage can be PnP (similar to replacing the mobo, given
BIOS is a large part of how OS "sees" the mobo), bloody-minded vandor
payloads such as Product Activation, or more subtle tech issues such
as timing assumptions, etc.

What other autoloading software do you have running on this system that may
be taking up your limited RAM? A 48 meg system *is* going to freeze and
misbehave with virtually all modern software - all the tweaking in the world
isn't going to change that.

The trick with old PCs is to keep them in their time bubble - avoid
significant software written more than 2 years after the hardware was
made. If you do that, an old PC can run as well as when it was
shipped. The fly in the ointment is where this PC has to contact the
present world; data formats, exploit risks and so on.

So no, a 48M system doesn't have to misbehave, and won't if it is not
confronted with modern bloatware, especially modern bloatware that
runs underfoot, as so much trash does (take an ass-kicking,
QuickSlime, Adobe Reader et al).

Adding heroic RAM to an old PC won't always help either; sware that
assumes you have 256M RAM is also going to assume disk space, disk
performance (think PIO vs. UDMA) and CPU performance that the old PC
is not going to be able to offer.

Anyway, I've been referred to this thread to comment on your belief that
disabling write-behind somehow speeding up internet surfing.

That makes no sense at all...

I'll ask you define what you mean when you say "internet speeds".
Are we talking the time it takes for a page to appear? Modem
throughput? If the latter, there's no way write caching can make
a difference.

Agreed... if modem-bound, "Internet speed" generally = modem
connection speed, and any subtrafuges that avoid modem access, such as
web caching on HD and killing off bandwidth-stealing junk.

Logically, the only thing I can suggest is that a browsing slow-down, if not
related to your modem, phone line etc, is actually being caused by problems
with your IE cache, whether it be size, or corruption, or third party
software interference.

On the last, think av scanners and malware.

As a test, I'd ask you to turn write caching back on, and nuke your IE
cache, make sure you is not using the 'automatically' cache option
(IE settings) then run a scandisk and defrag. I'm betting this will improve
your browsing speed, and that once you complete these steps the
tweak will no longer make a difference. If you've got an overlarge or
corrupt cache, and the system is suffering from an extended period of time
without defragging, it will slow things down.

Yep. We are talking Win9x here, and thus FATxx file system, and thus
linear directory lookup. A stupidly-large cache (as is the duuuhfault
in IE, prolly the #1 reason why "Firefox is faster") will cause long,
slowly-growing (and thus fragmented) directory chains, onlt partially
mitigated by IE's laudible practice of using randomly-assigned use of
multiple subdirs within the cache.

Let's say you have a 256M web cache.
Let's say IE spreads cached material over 8 different subdir chains.
Let's say the average size of a cached file is 1k
Let's say all cached files have 1 extra dir entry for LFN
Let's say your HD volume uses 4k clusters

How many files will be in each subdirectory before the oldest ones
start to be automatically purged?

256M / 1k = 262144 files, call it 250 000
250 000 / 8 subdirs = 31 250 files per subdir, call it 30 000

OK; how many clusters will be needed for each subdir to hold all the
directory entries for these 30 000 files?

Let's say a dir entry takes 32 bytes (I can never remember!)
So 1 entry + 1 LFN segment = 64 bytes
So 30 000 such items = 30 000 x 64 = 1 875k = 469 clusters
That's 64 items per cluster; back-check 30 000 / 64 = 469

When one creates a new item in the web cache, the name has to be
unique, so as not to overwrite what is there. Let's say you want to
create a button.gif, you first look up that name, if found, try
button(1).gif, then button(2).gif, etc. Can take multiple searches
through that wretched fragmented 469-cluster horror-show, and then you
finally get to create button(37).gif - yep, that can be palpably slow,
when there may be 10 loose bits of gravel in a single HTML page.

This is one reason to prefer a 20M IE web cache. Another is to reduce
the actual size hogged by the cache - that nominal "256M" cache will
really cost more; 2M for each of the 8 subdirs themselves, plus 4 x
the size of the 1k files as each hogs a 4k cluster - over 1G.

Regarding this quote:

I was reading microsoft website about setting port speed
http://technet2.microsoft.com/Window...4f1a41033.mspx
when I read what this same article said about write-behind "You may
want to disable the write behind cache function, especially if you own
system critical applications, and ALWAYS shut down Windows AFTER
closing ALL running programs! This means all data will be immediately
written to disk, bypassing the cache."

I can't find the text you site at the URL you give.

I find the page, yes; now let's search for a text biopsy "cache"...
not in the base page, let's try one level deep into the link block...
no, I can't find "cache"; a sanity-check on method viability does find
"the", so the search process works.



---------- ----- ---- --- -- - - - -
Don't pay malware vendors - boycott Sony
---------- ----- ---- --- -- - - - -

On Sun, 22 Jan 2006 23:57:45 +0800, "Sandi Hardmeier - MVP"

Cacheman certainly does run in the background unless specifically set not to
do so - I made a point of checking into that point. It cannot show how much
memory is free "at any given time" unless it is loaded and actively
monitoring.

With only 48 megs of RAM, "some" autorunning programs will certainly be a
problem. With such a low level of RAM on Win98 you can't afford any
autoloading programs. Do you have an antivirus program? There goes what
little free memory you had available to you, right there? You don't have
antivirus? Sorry, but then I have to say that you are more likely than not
to have malware on your system of some type or other.

Yep. But I cringe when you say "free memory" - "free memory" is a bad
concept, because at all times, all RAM should be used for something,
and much notional "user (virtual) memory" is free of cost, i.e. it
doesn't exist in any has-to-hit-the-HD sense at all.

The only relevant memory statistics a
- swap file in use
- locked memory

I'll admit that I am, even today, maintaining some Win95 systems with 16 and
32 meg of RAM, and some Win98 systems with 64 Meg, but the only reason such
low stats are workable is because these systems are hooking into Terminal
Servers, with the server hosting the applications, meaning that the boxes
only need enough memory to render some bitmaps.

As mentioned before, even Win95 in 8M and Win98 in 16M are quite
useable as long as the app load is trim.

There are two general performance tactics:
- avoid HD accesses (e.g. add RAM)
- make HD accesses "cheaper" (e.g. reduce head travel)

I've always applied both tactics, with the latter being met via
partitioning that concentrates most disk access (and all swap file
paging) within the first few % of the HD. If you have "one big C:"
that is 90% full, YMMV drastically, because whatever disk accesses you
have, are REALLY gonna hurt.

You say that you freed up half a gig of drive space yesterday; you do
realise that that there is a maximum size hard drive that
Win98 can use, yes?

Within hardware shipped in the age of Win98, what is more relevant
would be the BIOS's capacity limitations, which would usually be 8G or
32G, making the 137G limit quite irrelevant.

I doubt whether any Win98-era PC will be limited to 2G (FAT16) or 508M
(an older BIOS HD capacity limit that can bite original Win95-era PCs)

Personally, I think you need to get back to basics. Stop tweaking on the
word of unknowns discussing operating systems that don't even apply to you,
and touting advice that is doubtful at best, and dangerous at worst. I
believe your system has been tweaked to death and tweaked in such a way to
do more harm that good. Untangling things will be well nigh impossible.
Given the choice I would wipe out your install and start afresh.

I would hardly ever recommend that, because the skill set that barfs a
system that badly is likely to snooker itself when attempting this
("why can't I read the CD drive?", "hey, where's my OS CD?", "why do
my ancient hardware devices not work and have a ! on them in Device
Manager?") and/or be under-patched ("why does it hurt when I pee?")

http://cquirke.mvps.org/reinst.htm refers.

You have to reduce its size. Simple defragging also isn't sufficient.

Amen! Scandisk, clear cache, shrink size, defrag - in that order.



---------- ----- ---- --- -- - - - -
Don't pay malware vendors - boycott Sony
---------- ----- ---- --- -- - - - -

On Sun, 22 Jan 2006 19:42:26 GMT, (Olive) wrote:

In a nutshell the link between drive write-behind and internet surfing
is this quote "Anything that prevents your CPU from responding quickly
enough to interrupts from your UART can cause overruns."
http://www.cerberus-sys.com/~belleis...aq/overrun.htm

This is true, though the significance varies.

Firstly, do this (sorry, no baby-steps; RTFM):
- find and rename away the modem log
- enable modem logging, append to log
- do your web surfing, etc.
- peruse log for serial port overruns

If no serial port overruns, forget this issue.
If some, then try the following...
- reduce modem COM port speed
- set UART buffer to trigger interrupt after fewer bytes

Again, no baby-steps; e.g. I'm not going to describe what the modem
log is called, how to find files, where to change these settings :-)

Also, check the UART your modem uses. An ancient modem may contain an
unbuffered UART, and an ancient serial port card may contain an
unbuffered UART to which an external modem is connected.

Use MSD from DOS mode to check what UARTs you have; I can't remember
the actual namesaccurately, but AFAICR 16550 is OK (16-byte buffer)
whereas 8450 is Bad (1-byte "buffer") and 14550 is also Bad.

Now let's look at the notion that "disabling write caching will help
avoid serial port overruns". Let's assume that disk activity has this
effect. The more disk activity, the more likely this will happen.
The more "urgent" the disk activity, the more likely it will happen at
a bad time for the UART.

By that logic, forcing every disk write to happen immediately, will
both increase the number of disk accesses, and be more likely to force
these to occur when the PC is "busy". On that basis, I'd expect
disabling write caching to make things worse.

My understanding is that (Com port) overrrun errors is a data "layer"
problem which slows data flow between "layers"

I don't find that a helpfullevel of abstraction for this topic.

What happens is:
- the buffer trigger point is reached
- the serial port generates an interrupt
- the OS services the interrupt, clearing the buffer

With an old UART, there is no buffer; each time a byte arrives, an
interrupt is triggered, and the OS has to clear the buffer before the
next byte arrives. Madness.

With a "new" UART, there is a 16-byte buffer and you can set the
trigger point to generate an interrupt after, say, 12 bytes arrive
(with 4 to go) or after 4 bytes arrive (with 12 to go) etc.

The more bytes you can swallow before generating an interrupt, the
fewer interrupts are generated, and the less often the CPU has to stop
what it's doing, push its context on the stack, branch to the
interrupt service routine, faff around wiping the UART's butt, pop the
context back, and carry on doing whatever it was doing before.

So you want to set the trigger point as late as you can get away with,
in the interests of general performance. But set it too late, and
there's a risk the buffer will fill and overrun before the cavalry
arrives to clear the buffer. So if you get buffer overruns in the log
(why guess, when you can know? Use the log, Luke) then back off the
gas on the buffer threshold and swallow the processing overhead.

Sandi, as for my definition of of internet speed, all things being
equal, a 56K modem should at best give you an average speed of 7.0
Kbytes per sec (K/sec). My average seldom went beyond 2.3K.sec.

Modem speed is in bits per second, and as communications adds extra
bits (framing, stop, start, whatever - generally 2 fluff bits per 8
data bits per byte) you divide this by 10 for the max bytes per sec.

So let's assume a 56k modem connects at 50k (which is really
best-case), you'd expect a maximum byte rate of 5k, all things being
equal. The modem can compress data up to 4 times, so the best-case
modem throughput could be as high as 20k, as long as the serial poert
speed can cope. In practice, your modem traffic is either
intermittent scraps of uncompressed HTML, or already-compressed .GIF,
..JPG, .ZIP and self-extracting installer .EXE, so 5k is best-case.

In practice, the best I expect to see is 2 - 2.5k on downloads, unless
I use a download accelerator [*1] to pull the same file down from 4
points in the file at the same time (like a 4-lane highway); then I
get 4k+ that would be close to max on typical 41k, 44k, 49k connects.

[*1] Use a clean download accelerator like Star Downloader, rather
than Download Accelerator Pus or something similar that's infected
with commercial malware

In contrast, even "slow" 192k ADSL gets 22k per sec and I don't have
to wait until it's finished so I can hang up the phone line to avoid
extra telcom charges. There's a large message in there.

All the above assumes the only modem traffic is that which you
generate. If you have a fistful of commercial and traditional malware
banging away, as well as background Windows updates, other 3rd-party
updaters, blah blah blah then YMMV.

Before all my tweaks I regularly saw 1.6 to 2.3K/sec for an internet
speed. After tweaks I saw 2.3 to 4.6K/sec. That's double! And I
haven't even disabled drive write-behind yet. I hope disabling
write-behind will further increase my average internet speed.

I doubt it. The single biggest boost would prolly come from a
download accelerator, and that requires sites that support what the
download accelerator does. Your best-case max is modem x 10,
so 4.6k is nearly there; below that, other issues such as server
latency and other 'net traffic is limiting the speed.

http://www.cerberus-sys.com/~belleis...aq/overrun.htm
"Another cause is poorly written 32-bit disk drivers that aren't
WD1003-compatible (needed for Windows' caching
software to work properly), and which lock-out lower priority
interrupts (like com port interrupts) for an inordinately long time
while they dump-to-disk a large write-behind cache. While awaiting
longer term fixes by upgrading disk/drivers/BIOS, you can
get temporary relief by turning-off write-behind caching.

That also dates from the early EIDE years, thus more Win95-era than
Win98. This was when EIDE allowed data to be sent in larger blocks,
which is more efficient, but each block takes longer to pass and if
interrupts are disabled during this time, then you can indeed get
serial buffer overruns (esp. if unbuffered UART).

Newer systems are "used to this", plus UDMA trasfer modes allow the
CPU to work (and thus serve interrupts) while the block transfers take
place. So at this point, we need to know your hardware...
- motherboard chipset?
- HD capacity, brand and model?
- are you using PIO or UDMA?
- is BIOS set to enable large block IDE?
- what UART is the modem on?
- what are the UART's buffer trigger settings?
- are you getting serial overruns in the modem log?

NOTE: Windows uses a Terminate-and-Stay-Resident (TSR) program
for disk-caching called smartdrv which is loaded by your
autoexec.bat file. Add the switch /X to turn-off write-behind
caching. Windows for Workgroups uses a VxD called VCACHE,
ignoring smartdrv except for floppy disk drives. Write-behind
caching for VCACHE is turned-off with a line in the [386enh]
section of system.ini that says ForceLazyOff=C (or =CD if you
have two hard drives) with no spaces and no : after drive
letters.

Geez, this is from Win3.aaaaargh!!

Windows 3.yuk did indeed rely on DOS's SmartDrv.
Windows for Workgroups could use VCache instead, should be set so
Windows 95 original replaces all DOS disk services with VFAT

So unless you are in DOS Compatibility Mode, SmartDrv is utterly
irrelevant. In fact, I doubt if any Win9x can tolerate running on top
of DOS SmartDrv, even if in DOS Compatability Mode; I'm pretty sure it
kicks SmartDrv out of the loop once it boots, and will misbehave (you
don't wanna know) if some DOSware cacher is snuck in.

vomit ...rest of shuddering horrors from the evil days of Windows
3.yuk snipped. You've set my therapy back 5 years, you ^%#$ ! :-)



---------- ----- ---- --- -- - - - -
Don't pay malware vendors - boycott Sony
---------- ----- ---- --- -- - - - -

Kewl!!! The infamous Chris kinda agrees with me... almost... most of the
time... a red letter day ;o)

--
__________________________________________
Sandi - Microsoft MVP since 1999
http://www.ie-vista.com
http://inetexplorer.mvps.org

Inetexplorer has changed - for instructions on how to
find old URLs, go he
http://msmvps.com/spywaresucks/archi.../14/46971.aspx


"cquirke (MVP Windows shell/user)" wrote in
message ...
On Sun, 22 Jan 2006 16:04:07 +0800, "Sandi Hardmeier - MVP"

1) Is it possible that Cacheman is messing around with your settings - it
runs in the background, yes?

Ahhhh yes, that is quite likely!!

3) I will ask why you can't update your BIOS.

I would be very, very reluctant to ever recommend a BIOS upgrade...
- risk of failed upgrade leaving you with no BIOS
- risk of breaking the software installation's assumption base
...as there's no ways I'd want to be responsible for fall-out I might
not be able to fix even if I had my hands on the machine.

If BIOS *really* breaks, the only fix is to replace the chip.
Soldered in, as it usually is? Replace the mobo. Can't get that type
of mobo, e.g. old, proprietary or laptop? Replace the PC.

Software breakage can be PnP (similar to replacing the mobo, given
BIOS is a large part of how OS "sees" the mobo), bloody-minded vandor
payloads such as Product Activation, or more subtle tech issues such
as timing assumptions, etc.

What other autoloading software do you have running on this system that
may
be taking up your limited RAM? A 48 meg system *is* going to freeze and
misbehave with virtually all modern software - all the tweaking in the
world
isn't going to change that.

The trick with old PCs is to keep them in their time bubble - avoid
significant software written more than 2 years after the hardware was
made. If you do that, an old PC can run as well as when it was
shipped. The fly in the ointment is where this PC has to contact the
present world; data formats, exploit risks and so on.

So no, a 48M system doesn't have to misbehave, and won't if it is not
confronted with modern bloatware, especially modern bloatware that
runs underfoot, as so much trash does (take an ass-kicking,
QuickSlime, Adobe Reader et al).

Adding heroic RAM to an old PC won't always help either; sware that
assumes you have 256M RAM is also going to assume disk space, disk
performance (think PIO vs. UDMA) and CPU performance that the old PC
is not going to be able to offer.

Anyway, I've been referred to this thread to comment on your belief that
disabling write-behind somehow speeding up internet surfing.

That makes no sense at all...

I'll ask you define what you mean when you say "internet speeds".
Are we talking the time it takes for a page to appear? Modem
throughput? If the latter, there's no way write caching can make
a difference.

Agreed... if modem-bound, "Internet speed" generally = modem
connection speed, and any subtrafuges that avoid modem access, such as
web caching on HD and killing off bandwidth-stealing junk.

Logically, the only thing I can suggest is that a browsing slow-down, if
not
related to your modem, phone line etc, is actually being caused by
problems
with your IE cache, whether it be size, or corruption, or third party
software interference.

On the last, think av scanners and malware.

As a test, I'd ask you to turn write caching back on, and nuke your IE
cache, make sure you is not using the 'automatically' cache option
(IE settings) then run a scandisk and defrag. I'm betting this will
improve
your browsing speed, and that once you complete these steps the
tweak will no longer make a difference. If you've got an overlarge or
corrupt cache, and the system is suffering from an extended period of time
without defragging, it will slow things down.

Yep. We are talking Win9x here, and thus FATxx file system, and thus
linear directory lookup. A stupidly-large cache (as is the duuuhfault
in IE, prolly the #1 reason why "Firefox is faster") will cause long,
slowly-growing (and thus fragmented) directory chains, onlt partially
mitigated by IE's laudible practice of using randomly-assigned use of
multiple subdirs within the cache.

Let's say you have a 256M web cache.
Let's say IE spreads cached material over 8 different subdir chains.
Let's say the average size of a cached file is 1k
Let's say all cached files have 1 extra dir entry for LFN
Let's say your HD volume uses 4k clusters

How many files will be in each subdirectory before the oldest ones
start to be automatically purged?

256M / 1k = 262144 files, call it 250 000
250 000 / 8 subdirs = 31 250 files per subdir, call it 30 000

OK; how many clusters will be needed for each subdir to hold all the
directory entries for these 30 000 files?

Let's say a dir entry takes 32 bytes (I can never remember!)
So 1 entry + 1 LFN segment = 64 bytes
So 30 000 such items = 30 000 x 64 = 1 875k = 469 clusters
That's 64 items per cluster; back-check 30 000 / 64 = 469

When one creates a new item in the web cache, the name has to be
unique, so as not to overwrite what is there. Let's say you want to
create a button.gif, you first look up that name, if found, try
button(1).gif, then button(2).gif, etc. Can take multiple searches
through that wretched fragmented 469-cluster horror-show, and then you
finally get to create button(37).gif - yep, that can be palpably slow,
when there may be 10 loose bits of gravel in a single HTML page.

This is one reason to prefer a 20M IE web cache. Another is to reduce
the actual size hogged by the cache - that nominal "256M" cache will
really cost more; 2M for each of the 8 subdirs themselves, plus 4 x
the size of the 1k files as each hogs a 4k cluster - over 1G.

Regarding this quote:

I was reading microsoft website about setting port speed
http://technet2.microsoft.com/Window...4f1a41033.mspx
when I read what this same article said about write-behind "You may
want to disable the write behind cache function, especially if you own
system critical applications, and ALWAYS shut down Windows AFTER
closing ALL running programs! This means all data will be immediately
written to disk, bypassing the cache."

I can't find the text you site at the URL you give.

I find the page, yes; now let's search for a text biopsy "cache"...
not in the base page, let's try one level deep into the link block...
no, I can't find "cache"; a sanity-check on method viability does find
"the", so the search process works.



---------- ----- ---- --- -- - - - -
Don't pay malware vendors - boycott Sony
---------- ----- ---- --- -- - - - -