Path: utzoo!utgpu!jarvis.csri.toronto.edu!clyde.concordia.ca!uunet!tut.cis.ohio-state.edu!ucbvax!VAX.FTP.COM!jbvb From: jbvb@VAX.FTP.COM (James Van Bokkelen) Newsgroups: comp.protocols.tcp-ip.ibmpc Subject: IBM LAN support programs Message-ID: <8912201719.AA28561@vax.ftp.com> Date: 20 Dec 89 17:19:02 GMT References: <1109@crlabs.CRLABS.COM> Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 31 Date: 19 Dec 89 23:57:47 GMT From: Chris Wiener .... Does anyone have performance figures for FTP Software's PC/TCP product with token ring support built-in versus token ring support obtained using the recently-posted packet driver.... If both were being used by the same TCP (e.g. our 2.04 ETHDRV kernel on the packet driver and our 2.04 IBMTR kernel going direct), I'd expect the direct to go at least 20% faster when talking to hosts on the local network. This is because the direct version can use a 1960-byte TCP max segment size (it knows it is on an 802.5 ring), while the other can only send 1460 bytes per packet (it thinks it is on an Ethernet, where the MTU is 1514 bytes instead of 2042). This wouldn't hold via a router, where the MTU is reduced, or when talkiing to 4bsd systems, which don't believe TCP segments can be bigger than 1024. The IBMTR kernel would also use more memory, because its default buffers are bigger. All bets are off if you aren't using the same TCP for each side of the test. By the way, we don't support the use of our ETHDRV kernel with the pseudo- Ethernet 802.5 packet driver. This is because the RIF-handling code in the packet driver will come into conflict with our Ethernet ARP module after having talked to some number of hosts; you'll get "It was working fine, but then I couldn't open a telnet session to foo.edu. I could still talk to bar.com..." James B. VanBokkelen 26 Princess St., Wakefield, MA 01880 FTP Software Inc. voice: (617) 246-0900 fax: (617) 246-0901