Question on speed measurements

[kanenas]

Hello.
I'm running Win2K SP3 with a DSL connection 3000/512.
I'm doing a test trying to set my system up so it's optimized for accesses to my proxy.
I'm using Cablenut values as shown by Lobo on his "DSL, W2K and XP settings" with only the RWIN changed.
Up to now, I get the best results with an RWIN of 292000 (I'll try even higher values later). The high results seem to work better since the proxy has an RWIN of 1M.
Here are the questions:
1) Using a 100MB file for download, IE6 averages 320KB. For the same test AnalogX Netstat Live averages at 710KB.
Does that mean that the TCP/IP overhead is their difference (390KB)? Isn't that too high?
2) On the same test, AnalogX was averaging an Upload of around 350KB.
Now, this has to be overhead since there was nothing else active and uploading.
These results were repeated at the same ratio with different RWINs.
Why these high figures? Both up/down overhead is almost the same as the real data. Is there a mistake somewhere in my configuration or calculations? I would think that the large RWIN would reduce the overhead.
3) This IRQxx tweak with its value set up in System.ini always refers to things like IRQ03 or 04. These values are for serial ports. Does the tweak also apply to Ethernet connections where the Ethernet card might be on a port like 72?
4) The BufferMultiplier registry entry. Some people recommend setting it to 6 and others to 400. What exactly does it do and what's the right value?
Thanks a bunch for any answers.

[mnosteele52]

First off if you are behind a proxy you won't get an accurate speed test result, depending on how the proxy is used (and allot are used this way), they keep data, web pages or downloads on a server and once you or someone else with the same proxy had visited a web page or downloaded a file it is cached on that server so the next time you or someone else visits the page or downloads that file it is much faster. If you are not required to use a proxy then don't. They also limit your TCP/IP settings so there isn't much point in tweaking them. Like I said though.... allot of proxies do this but not all, I don't know your setup. The system.ini tweak has always been debateable..... there is no hard proven evidence that it works other than people say they notice a difference. Some say you cannot assign memory to an Irq like you can a com port which is where the tweak originated, you can try it and see if it helps or not. The buffer multiplier entry.... here is the definition:

BufferMultiplier REG_DWORD Multiplier

DefaultReceiveWindow and DefaultSendWindow get divided by this value to determine how many massages can be sent/received before flow control is imposed.

Default: 512

This is an AFD parameter based on bytes of memory (1024) NOT a TCP/IP parameter based on bytes of data (1460). So you should set it to 1024 (400 hexidecimal) which is equal to a byte of memory, ALL AFD parameters are memory buffers and the DefaultReceiveWindow & DefaultSendWindow should be multiple of 1024 also since they are based on memory, it's not that they won't work if they are not, but you are assigning a certain amount of memory as a buffer for your data and when assigning memory it is based on how memory is measured.... in increments of 1024, you can also only buffer as much data as your ISPs caps this is why these values should be based on your caps in kilobytes X 1024... you can't buffer any more data than they can give you so much more is a waste. I hope this explained things.

:) ;) :D

[kanenas]

Thank you for the information.
I did some further testing and got more disappointed with the whole affair.
First off, the 100MB file I download has been set up on the same server where the proxy is. I can bypass the proxy and get to it through a different port. In case I get it through the proxy, it has been set up so as not to stay in the proxy's cache.
Besides the speed improvement we wanted to see what's the effect of the proxy. By the way, we have to go through the damn thing plus a firewall.
The results are not encouraging though.
I measured values at the mid-point of the 100MB download, by IE's speed indicator and by AnalogX's Netstat Live.
Ratios were always similar with the best results in my case of an RWIN around 256960 and 292000.

1) Bypassing the proxy:
IE = 315KB (about 84% of cap).
Netstat = 345KB (incoming)/8KB(outgoing).

2) Using the proxy:
IE = 320KB
Netstat = 710KB (incoming)/370KB(outgoing) !!!!

Unless Netstat has a loose screw somewhere, it looks like the proxy triples the data transferred (double on incoming) but somehow manages to keep the same throughput for the file's actual transferred data.
What's going on here?
If these values are correct, then I get data way past my cap, if I use the proxy.
It also implies that the capacity for transfers is way beyond what I get when I bypass the proxy so there might be something that cuts down the speed.
It's not RWIN since anything below or above my best value gives worse results.
It's not the BufferMultiplier either. I tried the 1024 value, 2048, 512, 256, 128. No measured differences.
Any comments on the above?
Isn't there anybody who's behind a proxy and managed to successfully tweak the settings?
Thanks.