home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!crdgw1!rdsunx.crd.ge.com!rdsunx!barnett
- From: barnett@grymoire.crd.ge.com (Bruce Barnett)
- Newsgroups: comp.protocols.tcp-ip
- Subject: Re: Why would a Sun take 1.5 seconds to retransmit a missed packet?
- Message-ID: <BARNETT.93Jan8142235@grymoire.crd.ge.com>
- Date: 8 Jan 93 19:22:35 GMT
- References: <184841@pyramid.pyramid.com>
- Sender: usenet@crd.ge.com (Required for NNTP)
- Reply-To: barnett@crdgw1.ge.com
- Organization: GE Corp. R & D, Schenectady, NY
- Lines: 42
- In-Reply-To: lstowell@pyrnova.mis.pyramid.com's message of 7 Jan 93 20:34:04 GMT
- Nntp-Posting-Host: grymoire.crd.ge.com
-
- In article <184841@pyramid.pyramid.com> lstowell@pyrnova.mis.pyramid.com (Lon Stowell) writes:
- > Transmitting using what protocol? FTP, NFS copy or what?
-
- It's TCP based. Similar to FTP.
-
- >>Packets are missed by the archive system, so the Sun has to retransmit
- >>the missed packets. I have noticed large delays (0.5 - 1.5 seconds)
-
- > Those are pretty typical numbers for TCP layer time-outs...
-
-
- Thomas Narten mentioned that the granularity of the retransmit timer
- is a multiple of 0.5 seconds. Also for every ACK of one segment sent,
- you have to restart the timer for the next segment.
-
- I wonder if a finer granularity on the retransmit value would help if
- the peer is on the same network? I guess this is tough because some
- peers may be slow and some may be fast. It could keep track of the
- time for an ACK and try to pick a timeout near that time, I guess.....
-
-
- > If this is FTP, you'd probably be better off getting rid of the
- > fatal collisions on the net.....unless you can set the time-outs
- > to a very small number (very dangerous, but....)
-
- It's not collisions, because I saw the packets being sent. It seems
- the receiver _missed_ seeing the packets on Ethernet. People have told
- me the best solution is to upgrade the receiver. This could be a
- hardware upgrade (faster CPU, more memory, more ethernet ports,
- smarter ethernet ports, etc.) and possibly some software tuning. Does
- anyone have advice on the relative advantage of each option? How much
- would more memory help? A faster CPU? A second CPU? A second Ethernet
- card? A faster disk? What is the best choice? The most cost-effective
- choice? Who's gonna win the Superbowl? Well, forget the last question.
- I was on a roll..... :-)
-
-
- Thanks for all of the advice, esp. Stephen C. Trier, Aaron Leonard,
- Thomas Narten, Anay Panvalkar, Paul Hyder, and Roger-Hunen.
-
- --
- Bruce Barnett <barnett@crd.ge.com> uunet!crdgw1!barnett
-