Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS |  [KS] experimental TCP window size
I made a experimental on TCP window size so I pushed upper beyond to 4MB TCP window size so I able getting steady holding at top speed 20Mbps so I'm glad that worked in testbed.
I am on premier service... |
|
Gamer
join:2006-12-09
Phoenix, AZ | Are you actually getting a true steady 20Mbps, or 20Mbps on a speedtest? If the latter, it's quite possible you're just causing excessive buffering of data resulting in a sudden burst. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
 ·Cox HSI
4 edits | Gamer, excessive buffer of data is good for broadband. I told you it is for broadband only. That's not for dial up. so however larger window size allowed be continuous without wait for acknowledge before sending or receiving a data packet... Also Microsoft tech support site said that windows 2000 pro TCP window size can be pushed up to 1 Gigabytes with window scaling enabled. Plus, There is no issue problem after pushed upper limited size so I set 4 megabytes of TCP window size and it's worked to steadying top speed at 20Mbps for few minutes and down to 15Mbps and go back up to 19Mbps to 20Mbps range that worked! I glad that my experimental really worked so good. Oh, I almost forget to tell you that win9x,NT,200,XP are TCP static, not dynamic. Gamer, Windows vista is right for you if you want dynamic TCP turning if you re so worrying about bandwidth delay product issue so you should look at windows vista that is newest operating system from Microsoft. And I am not worried about bandwidth delay product issue that I am sticking with windows 2000 as long as it works well and I am not interested in new windows vista operating system...
FYI, I pushed again from 4MB to 16MB window size now
That is steady in 20Mbps range that really works so well. |
|
tcp
@cox.net
| said by Ikyuao  :excessive buffer of data is good for satellite fixed it for you 
RWIN should be large enough to sustain continuous data transfer and no larger. Any larger will not increase speeds and will decrease responsiveness. |
|
Foosking5
join:2004-04-26
Glendale, AZ
| I reported this last month.
»{AZ} Speed Increase?
My tests still show up as this.
If you go and download a large file your thorough put wont reflect your 20meg connection. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
·Cox HSI
| reply to tcp
Ha ha ha, tcp I disagree with this "excessive buffer of data is good for satellite" after experimental new TCP window size so there is no responsiveness problems I am getting good responsiveness so no problems so again that is good for broadband only. Let me ask you a question if you can answer a question how come linux have autoturning feature in etc folder so linux is getting around 20Mbps at same location? If you re using windows vista, Try it and see it is same 20Mbps range at same location? Are you gonna to answer? But you wouldn't answer it because you do not understand so much  |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS | reply to Foosking5
Foosking, I would like to see if you can make a experimental TCP window size with window scaling enabled so try 16Megabytes window size in a experimental if you are using win 2000 or XP. |
|
tcp
@cox.net | reply to Ikyuao
i really don't need to say anything cuz this ur already demonstrating exactly how much you know  |
|
tcp
@cox.net
| reply to Ikyuao
u obviously missed my attempt at humor with the satellite reference...
btw, i don't have powerboost and am on preferred, so I guess I would have to post screens of da 8 - 9 Mbps speed test results i see from Vista, XP, Mac OS X, Linux, etc. don't know what da hell you're getting at, you're most likely seeing normal throughput variation that has nothing to do with RWIN. At that latency, an RWIN of 128000 bytes is sufficient to sustain 20 Mbps indefinitely, which if ur really getting powerboost and not permaboost you won't be able to do anyway, but feel free to continue ur experiments
changes in the responsiveness of the protocol is something that you most likely won't perceive in ur limited, uncontrolled testing. people's powers of perception are unfortunately rather poor, objective measurements in a controlled environment are required to come to any kind of worthwhile conclusion and the internet is anything but a controlled environment.
for giggles tried 16 MB RWIN on da more transparent OS's
didn't see any speed improvement on multiple downloading and speed tests
kernel memory usage unsurprisingly went up by almost 16 MB
so i can see that resource usage increased, but i can't see any benefit that came from it so i'll be going back to using a 2^16 bytes RWIN. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
·Cox HSI
1 edit | Tcp, windows 2000 TCP architecture is sufficient up to 1 Gigabytes of TCP window size with window scaling enabled according to Microsoft tech knowledge site or MS tech support site... Again, remember, that win9x,NT,2000,XP are TCP static.
XP come from old windows 2000 architecture that should work so fine and for windows vista there is autoturning feature is available as dynamic TCP turning. so tcp, what modem you re using? if using old early modems is no good. i recommended new modems like sb5100 or higher version series that will work so great. |
|
dvd536
as Mr. Pink as they come
Premium
join:2001-04-27
Phoenix, AZ | reply to Ikyuao
Get into packet loss situations then extra large receive windows suck because that whole 4MB will have to be re-transmitted.
--
You can never be too rich, too thin or have too much Bandwidth |
|
tcp
@cox.net
| reply to Ikyuao
Ikyuao, ur hilarious!
I'm well aware that you CAN set an RWIN of 1 GB in Win2K, but if you think that I or you SHOULD, well that's just sad.
I'm also well aware that Vista's TCP/IP stack automatically adjusts the RWIN to a value large enough to sustain continous data transfer and no larger.
I'm also well aware of what the words static and dynamic mean as applied to everything from RWINs to typed programming languages.
Don't know why ur asking about my modem. It's an SB5120 that's less than a year old. I'm on preferred in Tucson, which is advertised at 7 Mbps and I don't think we have powerboost here yet and the 8 - 9 Mbps I frequently get is well above what Cox says I have. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
·Cox HSI
1 edit | tcp, that's good you have sb5120 that should be work fine so i have sb5100... i have premier. u re on preferred... that's sucks. I don't buy preferred cuz it's slow than premier so i prefer to high bandwidth connection but I don't care about bandwidth delay product issue if i get some packet loss but inside of cox network is no problem for larger window size... |
|
jsouth
Jsouth
join:2000-12-12
Wichita, KS | Huh?? |
|
TheNerdGirl
join:2006-12-09
·Cox HSI
| reply to Ikyuao
Hi, guys, good evening  I'm on the preferred tier which is capped at 7Mbps down and I'm always getting this constantly:
8487/571(PcPitstop)
8247/559(Testmy)
I always get 8Mbps down and 559-560Kbps up night or day and that's using 64240 as the RWIN. I'm getting good response. If you guys meant by page loads, then pages does load very fast. I tried TCP Optimizer at 513920, all TCP Optimizer did was made my cable connection slow. I would have problems loading up pages. Removed TCP Optimizer and went back to DRTCP and 64240. Pages went back to loading faster. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
·Cox HSI
3 edits | TheNerdGirl, Did you turn window scaling on that allow you push up so upper limited of 64240 bytes? just like beyond limited of 64240 bytes.
Here mine:
And mine TCP window size is 4111360 bytes with window scaling enabled and timestamps enabled. My computer is directly to my cable modem so no firewall router here. any site pages are still loading so faster. and in future of case, when cox is offering 40Mbps or more then i can update my TCP window size from 4MB to 16MB size. |
|
Gamer
join:2006-12-09
Phoenix, AZ
2 edits | 3 second speed tests are NOT an accurate measure of CONSTANT data rates. Particularly when you're using things such as window sizes that exceed the size of the file the speedtest uses. The results may or may not actually be accurate. |
|
Ikyuao
Pro. debian Linux
join:2007-02-26
Wichita, KS
·Cox HSI
| Gamer, sorry to hear that you have doubtful about speedtest accurate what if i have windows vista with auto turning feature in TCP architecture then you have more doubtful again? i am sure you will be more doubt if i go with windows vista or linux and show you how much exceed? however auto turning feature in linux TCP stack is more stable than early windows9x,NT,2000,XP. so linux and windows vista doing are more accurate than old early windows (9x to XP)... |
|
Gamer
join:2006-12-09
Phoenix, AZ
2 edits | I use several Linux distributions (Primarily Fedora Core and Ubuntu), FreeBSD as well as Solaris and multiple versions of Windows (NT4, 2000, XP, 2003) on a daily basis. I've also played with every beta and RC of Vista, as well as a lot of the builds when it was still Longhorn, including the final RTM of Vista. The only thing I haven't really touched in recent years is Windows 9x; For the time being I'll continue pretending Mac's don't exist.
None of those speed test servers should be far enough from you to have a high enough delay that a larger window size will improve performance. Like I said before, it may be causing excessive local buffering or other innacuracies in the test itself. As stated by other users, packet loss now would heavily affect you.
I've achieved > 900Mbps on XP with completely default TCP parameters, per connection. Almost all machines acheive equal throughput given equal situations. |
|
TheNerdGirl
join:2006-12-09 | reply to Ikyuao
Tried enabling Window Scaling and it didn't do any good. |
|
