[Info-vax] Slow FTP transfer Linux -> VMS

Richard Whalen WhalenR at process.com
Mon Aug 31 09:26:11 EDT 2009


The "burst limit" is the maximum number of packets the TCP_OUTPUT routine 
will transmit at one time (provided that there is enough data in the user's 
output buffer and TCP window available) to reduce the chance of causing 
congestion in the network and limit the amount of retransmits that need to 
be done in the case of a lost packet.  One argument that I saw for the value 
being 4 was that this is twice minimum ack requirements.  The TCP protocol 
specifies that an ack can be delayed for 200ms or until two unacked packets 
have been received.

"John Wallace" <johnwallace4 at yahoo.co.uk> wrote in message 
news:5b973764-047d-4067-ad8f-5be40963a931 at n11g2000yqb.googlegroups.com...
On Aug 28, 8:14 pm, JF Mezei <jfmezei.spam... at vaxination.ca> wrote:
> > does anyone know why the linux box doesn't speed up, and how to solve 
> > it?
>
> The only FTP performance parameter on VMS is the system-wide logical name:
>
> "TCPIP$FTP_WNDSIZ" = "4096"
>
> The doc says 4096 is good. I have no idea what the measure is (probably
> 4k of data ?). Not sure if it can be increased much beyond that. This
> affects VMS on the receiving end.
>
> I know that on the dsl line, the VMS box can't get anywhere near the
> maximum throughput (which the Mac next to it can).

But FTP sits on top of TCP, which has some configurable parameters
(e.g. window size stuff, though it seems both TCPIP Services and
Multinet have parameters to allow FTP to override the default), which
sits on top of IP, which has some configurable parameters (e.g. MTU
stuff), which sits on top of hardware which may have some configurable
parameters (eg high-end NIC+switch support for "jumbo frames" etc).

A sub-optimal configuration in any one of these could cause poor
performance.

Then there's "TCP slow start", where disappointing performance doesn't
even need a misconfiguration, just two differing allegedly-correct but
not-really-compatible implementations.

Not a clue what Richard Whalen's "burst limit" is, is it a widely
understood term?

Not a clue what VMS IP stack the original poster is using either,
might it make a difference? 





More information about the Info-vax mailing list