Hi How can i disable Selective-ACK in windows 7 SackOpts in registry is set to 0x0 (0) But when i look with TCP analyzer at http://www.speedguide.net/analyzer.php?DATA_OFFSET=40&TCP_Options_string=020401f001010402&IP_MTU_DISCOVER=1&WIN=65472&RWIN=65472&MSS=496&SCALE=0&TTL=48&TSOPT=0&SACK_PERM=1&IP_TOS=0&IP=79.223.15.71×tamp=1438603387 it says SACKs: ON « SpeedGuide.net TCP Analyzer Results » Tested on: 2015.08.03 08:03 IP address: 79.223.xx.xx Client OS/browser: Windows 7 (Firefox 39.0) TCP options string: 020401f001010402 MSS: 496 MTU: 536 TCP Window: 65472 (multiple of MSS) RWIN Scaling: 0 bits asked 03 Aug '15, 05:05  NT Lucky showing 5 of 11 show 6 more comments |
This is a static archive of our old Q&A Site.
Please post any new questions and answers at ask.wireshark.org.
Why on earth would you want to disable SACK? Is this another of those fancy snake oil network speed up tricks? :-)
1: Don't do it. 2: Read https://support.microsoft.com/en-us/kb/2525390, it may be the reason.
I'd also be really interested to know why you would want to? With a link to Info, if you have one please.
1 :D just do it ^^ 2. I already tried the hot fix but when i try to install it it says "This update is not intended for your computer" :/ F... Windows 7 , xp was so nice ^^
3.Yes it is " fancy snake oil network speed up tricks" dont know what it means but i cann imagine :D
I have a latency between 15-23 without changing something. After changes 7-15 latency on 2 pc with windows XP Sp3 and i want to get windows 7 working with same settings .I have no Packet loss or something else .And for me bandwidth is not even importent :D
Try these settings an check your latency ^^
MTU 250-500 default tcp receive window RWIN Scaling: 0 bits SACKs: off TTL: 64 TcpMaxDupAcks 0 TcpAckFrequency 1 TcpNodelay =1 TcpDel/AckTicks 0
If you just change MTU and TLL you will note better latency :D
Then have you ever tried disabling flow control and IPv6? Or perhaps you can provide a trace so with a little bit of luck the root cause can be seen there. Or you really have an halfduplex network or at least an duplex mismatch as I mentionend parlier?
I highly doubt that. Any kind of measurement you're doing will be heavily biased except you're
1) using professional fiber optical, completely passive network taps
2) a dedicated professional network capture appliance with confirmed ns timestamp resolution.
3) using consistent tests with same size packets
If you measure latency without both 1) and 2) any measurements are not trustworthy. This may sound tough, but forget measurements taken on a normal PC, especially if it's the same PC you're running a client/game software on. You'll never get accurate results, so it's just a waste of time to do that ;-)
@Jasper: Of course I agree with you. The times in those traces are not much worth. But some other could perhaps be seen. (retransmissions, mtu things). It was more meant as a check and maybe there is something uncommon in that trace.
Flow control and IPv6 are disabled and Networkadapter is 100Mbit fullduplex
Witch trace i can give u or what u need to know ? Christian_R
Jasper i have a lot of time ^^ to waste . I used WinMTR for for about 24h on differnt ips with mtu 1492 and 500 and averrage latency was about 8-12 ms better with low mtu
@NT Lucky: First of all I have to repeat that I fully agree with Jasper. So I think we should start at the beginning if you like. What kind of latency you want to improve oneway or the twoway latency? And for what protocols icmp (ping) udp, tcp (time of an acknowledgement sent) or do you have a speciL application protocol that you use?
Flow control and IPv6 are disabled already :/ Its 100Mbit Fullduplex
What protocol do you want to improve?
I think you want to improve some kind of TCP Protocol, to gain a little bit more realtime behaviour. If I am right, then you normally can forget the most of your tunings from my point of view. To reach this goal(TCP Realtime Tuning) the follwings parameters can be interesting for you:
And if you take a trace and you see less then 1% Retransmissions then it is ok. But if you see a higher value then you should figure out why it is so.
And all this can figrued out in a proffesional way only with a reproducable test results/concept and professional trustworthy capture equipment.
And a pingtime has not much to say, if it is not the application you want to tune.
Hope this Infos will help you or at least some other person who read this thread.