[Info-vax] TCPIP performance for VMS

[jbriggs444 at gmail.com]

On Apr 22, 2:12 pm, Jim <mckinn... at saic.com> wrote:
> On Apr 22, 3:51 am, JF Mezei <jfmezei.spam... at vaxination.ca> wrote:
>
>
>
>
>
> > Found out that my old MAC can sustain outbound FTP transfers at roughly
> > 80KB/s (640kbps on an 800kbps ATM/PPPoE link (ADSL).
>
> > The Alpha VMS host, sitting next to the mac at the ethernet levl only
> > does 49 KB/s ( 392 kbps) sending data to the same FTP host on the internet.
>
> > going towww.speedguide.net:8080givesme some of the TCP parameters.
> > (you can curlhttp://www.speedguide.net:8080onVMS and then copy the
> > url in the results (a redirect done with evil javascript) and paste it
> > in a browser to see the results).
>
> > while pmtu_enable is set to 1 on the VMS host, speedguide reporrs MTU
> > discovery is set to off.
>
> > Setting ipmtu to 1492 (ADSL MTUs) with ifconfig make no difference to
> > the FTP performance on the VMS host.
>
> > I tried changing tcp_rexmit_interval_min to different values, and it
> > just seemed to decrease performance when the manual says it should
> > increase it for high latency links.
>
> > The TCPIP stack also doesn't seem to support timespamps in the TCTP
> > header. Anyone know if this can be enabled ? (this is used by modern
> > stacks to configure their congestion management).
>
> > Has anyone ever dealt with this ?
>
> > Is the answer: Migrate to Multinet ?
>
> I'm not a user of HP's TCP so I don't have answers to the questions
> that you ask - but I will offer this - the TCP send indow size can
> impact the speed of FTP transfers. Your MAC is faster than the VMS
> system when sending a file - does it use the same send window size as
> the VMS system? If not, try setting your VMS system to match it and
> see if you get similar performance.- Hide quoted text -
>
> - Show quoted text -

Irrelevant.  He's testing a transmit from VMS.  It is the receive
window at the receiver that would be relevant.  And even if we were to
suppose that an Internet speed testing site has an inadequate receive
window, surely that receive window would be the same size regardless
of whether a MAC or a VMS client were being tested.

I'd be trying to investigating the apparent PMTUD issue.  Ideal would
be to see a packet capture of the VMS data stream.

Given a packet capture it should be obvious at a glance whether we
have a bunch of 576 byte packets with DF clear or a bunch of 1500 byte
packets with DF set.

However there's the interesting possibility that the PPPoE
encapsulation is reducing the available MTU and the DF bit is either
clear or  ignored.  A MAC using a MTU less than 1500 could be getting
its packets through intact while the DSL modem is forced to fragment
full sized 1500 byte frames being emitted by the VMS box.  If the
speed testing site is detecting IP fragments, it could be falsely
concluding that PMTUD is not supported by the client when the truth
might be that it is not supported by the DSL modem.