Higher RWIN's DO NOT Cause Packet Loss

mnosteele52 · Post by **mnosteele52** » Sat Apr 20, 2002 10:34 am

Not telling me anything I don't know

, just go ask where you did that test and see what kind of answer you get - hehehehehehe.

Lobo · Post by **Lobo** » Sat Apr 20, 2002 10:42 am

I did it for someone that said Higher RWIN's cause packet loss, the only reason I went to DSL site is they asked me, me sorry, forgive me, Jr. starts 2nd Row tomorrow at Talladega, Hmmm
<-----------------------------

illestdynasty · Post by **illestdynasty** » Sat Apr 20, 2002 11:16 am

uh oh :2cool: time to put the RWIN on max!!!!!!!

Lobo · Post by **Lobo** » Sat Apr 20, 2002 11:45 am

On Max, good luck

illestdynasty · Post by **illestdynasty** » Sat Apr 20, 2002 12:03 pm

Just tried RWIN Level of 730000 and was getting a reported 2+sec stall on DSLR tweak test, but still no packet loss

RWIN level of 642400 seems to be working good. Web pages load much much faster.

Lobo · Post by **Lobo** » Sat Apr 20, 2002 12:12 pm

Use 642400 then

TeenInternetAddict · Post by **TeenInternetAddict** » Sat Apr 20, 2002 9:23 pm

Here's my dslreports test. The rwin number is set at 642400 and 642400, webpages download fast and stuff is fast.

JawZ · Post by **JawZ** » Mon Apr 22, 2002 12:27 pm

Originally posted by Lobo
I see no indication of packet loss:

http://ny-monitor.dslreports.com/tweak/ ... via=normal =RWIN 730000

http://ny-monitor.dslreports.com/tweak/ ... via=normal =RWIN 584000

http://ny-monitor.dslreports.com/tweak/ ... via=normal =RWIN 578160

http://ny-monitor.dslreports.com/tweak/ ... via=normal =RWIN 554800

http://ny-monitor.dslreports.com/tweak/ ... via=normal =RWIN 548960

and I see no indication of test results LOL! Where da data go????

Lobo · Post by **Lobo** » Mon Apr 22, 2002 12:35 pm

I believe the powers that be deleted them, they backfired on me, I was trying to prove point, not have anyone change theirs, weep, weep

neo86 · Post by **neo86** » Mon Apr 22, 2002 3:56 pm

Doesn't matter how high you set your RcvWindow because windows just lowers it back down to whatever works best. It's just a matter of how quickly it does it that makes a difference.

More accurate the RcvWindow the better your speed will be.

If you have a RcvWindows of 1,000,000 and connection speed of 1.5mbps then you're receive window is not accurate.

Webpages load quickly initially with a large rwin because cable modems allow an unlimited burst when a connection to a server is made. Unfortunately, they slowly cap it back down which is why it won't be useful for large downloads.

Lobo · Post by **Lobo** » Mon Apr 22, 2002 4:00 pm

Were did you find this tibit:

Doesn't matter how high you set your RcvWindow because windows just lowers it back down

Kip Patterson · Post by **Kip Patterson** » Mon Apr 22, 2002 5:19 pm

Things have gotten a little out of hand.

When a TCP/IP connection is created, the machnies on either end agree upon RWIN's for both directions. The upstream RWIN can never exceed the RWIN of the distant machine, and the downstream RWIN can never exceed the RWIN of the local machine.

Once the connection is negotiated, data transfer starts. The remote machnie (which I will call the server) sends enough data to fill the RWIN of the local machine, which I will call the user.

If the RWIN is at least as large as the modem cap times the round trip time (latency), no increase in the RWIN will produce a faster speed. Period.

If the RWIN is larger than the modem cap times the latency, then data will accumulate at the router that is the choke point in the system, in this case the Cable Modem Termination System. If the CMTS is configured to retain the data, then there will be no packet loss. If the ISP elects to configure the CMTS to discard the data, then there will be continued packet loss. Moreover, if there is more than one TCP/IP connection as a result of more than one program being open or the use of a router, the latency will increase because the other data has to wait for the queue to be emptied. If you use an RWIN of 500k on a domestic cable connection, subsequent connections by the user or by other machines sharing the router will have latencies of 2 seconds!

Why does it seem as though large RWIN's help? If you start a download, the server sends an RWIN of data to the user, and that data stays in the TCP/IP RWIN buffer until the user tells the computer where to put it, in the case of a file download. The downloading program then reports the speed at whcih the data was transferred FROM THE BUFFER TO THE DRIVE, not the speed at which the data is transferred over the net.

Lastly, when you ping or traceroute a connection, the lost packets that appear have NOTHING to do with packet loss during a TCP/IP connection. Ping and traceroute do not use TCP/IP.

Kip

Norm · Post by **Norm** » Mon Apr 22, 2002 5:26 pm

Thanks Kip, you make it easy to understand.

Lobo · Post by **Lobo** » Mon Apr 22, 2002 5:37 pm

All those test scored 100%, they were meant to show no retransmitted packets, those RWIN's are not for DSL or PPPoE or anyone who has not been told to use them!!!!!!!!!!!!!

rmrucker · Post by **rmrucker** » Mon Apr 22, 2002 8:59 pm

Kip - Thank you, thank you, thank you.

One question - is the Receive Window always symmetrical? I was under the impression that the Send Window and the Rcv Window do not have to be the same size. [I am not really sure that is what you said, so I am trying to clarify this point.]

Thank you for stating that ICMP pinging and tracert's have nothing to do with RWIN. I got tired of saying that many months ago.

This is another point that is worth mentioning:

If the Tester only uses a download file that is only 146,000 bytes in size, then an RWIN that is much larger than this size CANNOT be adquately tested. The Rcv Window is never clearly "full" -- it cannot possibly be.

Retransmissions during TCP/IP are not necessarily due only to "packet-loss". A packet will be retransmitted if it is simply not ACKnowledged soon enough -- it does not have to be "lost".

Kip Patterson · Post by **Kip Patterson** » Mon Apr 22, 2002 10:12 pm

I hope that I did not suggest that the two windows were the same size. They are independant and are usually not the same size.

With respect to retransmissions, you are right on. I didn't deal with this aspect of TCP/IP, but would like to. I just got back from consuming a pizza I didn't really need, but pizza is a wonderful solvent for writer's block, and I got through several pages of writing dealing with these issues. I will remember to include retransmissions (and ask for you input as well.)

Kip

Lobo · Post by **Lobo** » Mon Apr 22, 2002 10:39 pm

This all falls on me, I was just trying to show people that retransmitted packets were not packet loss, the way I tell is if I download something and don't get all of it

Then I know I have packet loss

mnosteele52 · Post by **mnosteele52** » Tue Apr 23, 2002 12:37 am

Thanks Philip, see what you get for trying to be funny Lobo - hehehe.

Lobo · Post by **Lobo** » Tue Apr 23, 2002 8:26 am

I'm jumping, where's the couch, im afraid of heights, so unless your RWIN is smaller than 146,000 you have to take test with grain of salt at DSLR, we can use some of the info, not all. Thanks to rmrucker it's all becoming clearer now, Kip P. too, thanks guys,
and mnosteele52, im not going there, hehe

rmrucker · Post by **rmrucker** » Tue Apr 23, 2002 10:23 am

Kip, thanks for clarifying -- I was just misinterpreting your first paragraph.

Lobo, exactly. The Rcv Window is how much data (in bytes) that is 'buffered' during the transmission. If your window is very big (let's say 500,000 bytes) but the maximum size of the data being transferred is much smaller (say only 146,000 bytes), the Window is never full.

So, in essense, the largest RWIN that can be reliably tested at DSLR is 146,000. Anything larger than this exceeds the size of the data being transferred.

Of note, DSLR is working on an 'advanced' Tweak Tester that I believe will transmit around 1,000,000 bytes. This will likely never be available to the 'general public'. I suspect it will be limited to their "Premium Members".

SG Broadband Community

Higher RWIN's DO NOT Cause Packet Loss

Higher RWIN's DO NOT Cause Packet Loss

Re: Higher RWIN's DO NOT Cause Packet Loss