home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.nfs
- Path: sparky!uunet!noc.near.net!vaxeline.ftp.com!cream.ftp.com!jbvb
- From: jbvb@vax.ftp.com (James B. VanBokkelen)
- Subject: Re: NFS packet size
- Message-ID: <930105183618@cream.ftp.com>
- Nntp-Software: PC/TCP NNTP
- Keywords: TCP NFS
- Lines: 37
- Sender: root@vaxeline.ftp.com (vaxeline.ftp.com root account)
- Nntp-Posting-Host: whipped-cream.ftp.com
- Reply-To: jbvb@ftp.com
- Organization: FTP Software, Inc., Wakefield, MA
- References: <peterd.726170584@pjd.dev.cdx.mot.com> <51022@shamash.cdc.com>
- Date: Tue, 05 Jan 1993 18:36:18
-
- In article <51022@shamash.cdc.com> gsa@easyaspi.udev.cdc.com (gary s anderson) writes:
-
- I'm not sure about your version of TCP, but 1 sec tends to be a
- nominal value for determining when to retransmit. You must also
- accommodate "delayed acks" (200 msecs comes to mind for this value)
- when determining when to retransmit.
-
- In a good TCP, 1 sec. might be the initial value before any packets
- are exchanged, but the only way it would persist after the connection
- open would be if packet loss was such that you didn't get *any* clean
- send/ack cycles (without retransmits). A TCP might wait 200 ms.
- after receiving a single short segment, but it ought to issue the ack
- as soon as the RHS of the window should be moved (see RFC 813). This
- means that it normally won't dally when bulk data is being moved, and
- even if the last packet is small enough that a dally occurs, the ack will
- not wait any longer than the time to complete & reply to the RPC.
-
- You will also discover that NFS has an interesting delay pattern when
- using TCP because of "delayed acks" (i.e. the last packet
- is not a full MSS). Disabling Nagle's algorithm does eliminate this delay
- (some implementations allow you to do this).
-
- Many versions of 4bsd TCP don't really have a correct implementation of Nagle.
- Nagle says that if there is no unacknowleged data in the retransmit queue,
- send the segment regardless of size. 4bsd chose to wait rather than send a
- short segment when nothing else is outstanding. This produces the behaviour
- you describe, and may well vanish if you "disable Nagle", but Nagle himself
- isn't to blame.
-
- Is there some major point that needs rehashing?
-
- The fact that TCP largely de-couples the network RTT from the server's
- disk is not anything I'd heard before, but that doesn't guarantee it's
- new.
-
- James B. VanBokkelen 2 High St., North Andover, MA 01845
- FTP Software Inc. voice: (508) 685-4000 fax: (508) 794-4488
-
-