At 09:44 PM 10/31/01, John Tafasi wrote: >Here is an excerpt of what Mike Crane, author of the Cisco CID Exam >Certification >Guide from Cisco Press, is saying in his own words in this book, page 538: > >"Another adjustment is to increase the TCP window size to allow more >outstanding requests.
What he means is that one can increase the TCP window size to allow more outstanding packets that haven't been ACKed. ACKs are overhead. The fewer, the better. >This command can also minimize packet fragmentation >fragmentation, because larger packets means fewer smaller ones" That's where he goes astray..... >To me this make sense, because when you decrease the TCP window size you >will need more IP packets to carry the TCP Traffic. Here the packet >fragmentation occurs due to the TCP window size limit and not due to the IP >MTU size limit Typical values of the TCP window size are 8,192 bytes and 24,000 bytes, etc. They are much larger than a single packet size. It would be extremely bizarre for the window size to be smaller than a single typical IP datagram size (~1500 bytes). On the other hand, it's not a problem that the window size is much larger than an IP datagram and it doesn't affect fragmentation. The window size is used for flow control. Each side specifies in every packet how many bytes it is currently able to receive before the other side should stop and wait for an ACK. As I mentioned, this is usually a large number, which means the partner sends a few packets without waiting for an ACK for each one. Each side can have a different window size. PCs usually have a window size of about 8,192 bytes. I have seen big servers advertise a window size of 24,000 bytes. During the three-way handshake, each side also specifies the maximum segment (packet) size it is able to receive. Unlike the window size, which is advertised in every packet, the segment size is added as a TCP option to the TCP header in a SYN packet. The two sides do not need to agree, but they usually do. A typical size is 1460 bytes. Add the 20-byte TCP header and the 20-byte IP header, and you have a typical Ethernet packet. That's a good thing because otherwise fragmentation would be necessary. As an example, if a client says that its segment size is 1460 bytes and its window size is 8192 bytes, the server on the other end of the connection will send 5 1460-byte segments and then wait for an ACK. (Well, there's also the little matter of delayed ACKs and slow start, but ignore those for now.) I wish I had more time to explain it, but any good book on TCP (Comer, Stevens) explains it correctly. You should also take a look at a TCP session with a protocol analyzer and it will make sense. Unfortunately, the book you are reading does not seem to characterize TCP correctly. Priscilla >What you think guys > > >-- >watch your phone call records on the web at: >http://www.freedomstar.com/sh1885969 > > >| >""Circusnuts"" wrote in message >[EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > > John- I believe you've let this term confuse you. What you are seeing >with > > TCP windowing is how may packets you send before requiring an > > acknowledgement. This is better know as the sliding window principle, > > because consistent connections allow the window to open more freely (i.e. > > larger periods without an ack). When a network hiccup occurs, the window > > shuts and eases back open when the connection resumes. > > > > >http://www.cisco.com/univercd/cc/td/doc/cisintwk/ito_doc/ip.htm#xtocid223631 > > 9 > > > > All the best !!! > > Phil > > > > ----- Original Message ----- > > From: "John Tafasi" > > To: > > Sent: Wednesday, October 31, 2001 7:25 PM > > Subject: TCP Window size [7:24875] > > > > > > > Hi Group, > > > > > > I have a little but difficut question regarding TCP window size. How can > > the > > > tcp window size configured on a host affect ip packet fragmentation? >What > > > will happen if configure a host with a tcp window size that larger than > > the > > > ip mtu size? What is the command to configure tcp window size? > > > > > > Thanks > > > > > > John Tafasi > > > -- > > > watch your phone call records on the web at: > > > http://www.freedomstar.com/sh1885969 ________________________ Priscilla Oppenheimer http://www.priscilla.com Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=24899&t=24875 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]