home *** CD-ROM | disk | FTP | other *** search
- Path: news.ultranet.com!usenet
- From: "Albert P. Belle Isle" <belleisl@cerberus-sys.com>
- Newsgroups: alt.winsock.trumpet,alt.winsock,comp.dcom.modems
- Subject: Re: My ISP's winsock settings?
- Date: Thu, 29 Feb 1996 11:59:19 -0500
- Organization: Cerberus Systems, Inc.
- Message-ID: <3135DB67.6FB6@cerberus-sys.com>
- References: <4g38o9$mk5@zoom2.telepath.com> <4ga313$1br@news.onramp.net> <4h41ni$c69@mserv.wizvax.net>
- NNTP-Posting-Host: apb-p5-90.cerberus-sys.com
- Mime-Version: 1.0
- Content-Type: text/plain; charset=us-ascii
- Content-Transfer-Encoding: 7bit
- X-Mailer: Mozilla 2.0 (Win16; I)
-
- Mark J. Restifo wrote:
- >
- > On Mon, 19 Feb 1996 14:56:31 GMT, in alt.winsock, nezsez@onramp.net
- > recently put forth:
- >
- > [snip]
- >
- > >| If you are sure your provider has enabled default PPP or Van
- > >| Jacobsens' header compression, than you can set the MSS to 3,5, or 8
- > >| less than MTU.
- >
- > Is the above a typo? I realize that VJ compression compresses the headers
- > only, but that seems like kindof a small gap. Could you elaborate a little?
- > Thanks.
- >
- > --
- > Mark R.
- > ===========================
- > mrestifo@wizvax.net
- > 102444,3171@compuserve.com
- > ===========================
-
- Mark:
-
- This sounds, at first glance, like an interesting trade-off.
-
- Single TCP/IP packets (like control packets requesting the opening of a port
- or socket for a subsequent FTP or HTTP file transfer) typically have 20
- bytes of IP header and 20 bytes of TCP header added to the data segment.
-
- Consequently, to avoid their being fragmented to fit its MTU limit by a
- router along the way, the TCP data segment (which is never larger than MSS)
- should be 40 bytes less than that router's MTU.
-
- However, Van Jacobson header compression does, indeed, drop the *average*
- header overhead to less than 10 bytes on long streams of packets to the same
- recipient, by only sending the header changes over your CSLIP or PPP link.
- With the slow speed of a modem (compared to the Internet) this is a
- noticable gain.
-
- Consequently, for local transmission over your CSLIP or PPP modem link,
- while single control packets would be fragmented and slowed, FTP or HTTP
- packet streams would not, giving you slightly higher efficiency where you
- most care about it.
-
- Unfortunately, out on the Internet (as opposed to over your local CSLIP or
- PPP link), I believe that "a packet is a packet," and each one in a stream
- can follow a separate route. (This, of course, is the source of out-of-order
- reception and occasional "unnacceptable segment" errors due to slow ACKs).
-
- Consequently, in order to avoid IP fragmentation on the routers out on the
- Internet (as opposed to avoiding fragmentation through your CSLIP or PPP
- link), I believe you have to limit TCP data segments to at least 40 bytes
- less than the *smallest* of the router MTUs on the *entire* download path.
-
- Unless I'm missing something (which is quite possible), it appears to me
- that the only advantage is that you could further *reduce* your MTU (the
- no-fragmentation limit your winsock announces to your ISP's dial-in server)
- by an additional 30 bytes or so - not increase your MSS by that amount.
-
- Why would this artificial MTU limitation be advantageous? Why not just match
- the (larger) IP MTU on your ISP's Local Area Network, and do your TCP tuning
- (MSS and RWIN) as needed for the downloads you do?
-
- Any knowledgeable people care to comment?
-
- Regards,
-
- Al
-
- --
- ==================================================================
- Albert P. Belle Isle
- Cerberus Systems, Inc.
-
- Al's Winsock Tuning FAQ -
- http://www.cerberus-sys.com/~belleisl/mtu_mss_rwin.html
- ==================================================================
-