Re: ICMP ping effecting network flow?

Moe Trin wrote:
On Fri, 16 Jun 2006, in the Usenet newsgroup comp.os.linux.networking, in
article <e6ve3l$mu3$1@xxxxxxxxxxxxxxxxxx>, Roy L Butler wrote:

I was told by someone at a systems conference about 10 years ago that if
you ping a server over the Internet while downloading a file from it,
your download will take place quicker.

I've come across documentation that some routers place ICMP traffic in
their high-priority queue.

That still doesn't sound right. Lessee, 10 years ago... On 31 Dec 1995,
the Internet consisted of 1.05 billion IP addresses assigned to 40865
networks world wide (for comparison, the figures earlier this week were
2.318 billion hosts and 73466 IPv4 networks) (Source: RIR Zone files
dated 06/15/2006). If you look at any contemporary documentation
(examples - the man page for the common LBL version of traceroute, or
W. Richard Stevens 'TCP/IP Illustrated Volume 1 - the protocols", Addison
Wesley, ISBN 0-201-63346-9, copyright 1994) you'd see that many
connections over the Internet require at least 6 hops (routers) to get
anywhere - a more common value is over a dozen. In the traceroute man
page, it took 18 hops to go from Lawrence Berkeley Labs (about 45 miles
Southeast of Oakland) to MIT in Boston - and no less than 13 to get from
LBL to a specific server in Berkeley (not more than 50 miles away). The
Stevens book shows similar hop counts as well. It's a bit of a stretch
to think that ten or fifteen intermediate routers would note a ping
going by from "A" to "B", and therefore remember to pass TCP datagrams
between those hosts with higher priorities. While not very likely at
that era, it is also possible that pings could have been routed over
different intermediate hops compared to TCP.

As far as prioritizing ICMP - can you give some cite? The IP header has
a TOS variable, and one of those is the "minimize delay" flag (see page
29 of RFC0791). These flags have in general been ignored, even back in the
early 1990s (see the Stevens book). RFC1340 (superseded by RFC1700, which
in turn was replaced by on-line databases) and RFC1349 (from July 1992)
specified how these flags were to be set, and the 'Minimize Delay' flag
was only set for 'Telnet/Rlogin', FTP control (but not data), TFTP, SMTP
(command phase, not data) and DNS queries using UDP. These were generally
interactive or setup conditions. There is also a 'Maximize throughput'
flag which was set for bulk data transfers (FTP or SMTP data, DNS zone
transfers).

Both these guys were heavily certified (which I know doesn't always mean
qualified),

A lot depends on the certifying organization. Certifications are generally
most useful in getting past the hiring filters set by HR and PHBs. The
actual experience/knowledge would be determined in interviews with the
technical people.

My guess is that they might have thought (wrongly?) that certain vendors'
routers placed all traffic to/from a source address issuing echo_requests
in such a queue, elevating lower-priority protocols.

How much memory do you think is in the "average" backbone router? How
much traffic do you think it might see in any given moment? (If you were
traveling by air in the 60s-80s, you probably knew that if you traveled
anywhere in the "South" of the United States (actually meaning Southeast),
you couldn't go ANYWHERE without changing planes in Atlanta. By the mid-90s,
the Internet was a good bit better than that, but there were still a lot of
routes where the path from "A" to "B" went through major hubs like SFO,
Denver, Chicago, and New York - even if you were going from Dallas to
Atlanta.

I can't remember their names and I agree the time to ask was back then.

You might try 'comp.protocols.tcp-ip' or even 'alt.folklore.computers'.
My news server also lists a 'alt.folklore.internet' which might be useful.

Old guy

Thanks for all of the good info. My specific source stating routers may be configured to treat ICMP echoes as high-priority packets is section 6.5 of _Wide-Area Data Network Performance Engineering_, by Cole & Ramaswamy (AT&T engineers), written in 2000. They do mention that the opposite (lower-priority) is common and I can understand how the whole thing seems even less plausible on a packet-switching architecture... I'll give up the position of devil's advocate. :)


Roy
.

Relevant Pages

  • Re: Why Ping does not Work
    ... nor can I ping my server here in Richmond from ... I can ping my home computer ... the rest of the internet. ... layer produces complete protection. ...
    (microsoft.public.windowsxp.network_web)
  • Re: AD, DHCP or maybe DNS problem?
    ... worked and my trace gets out to MSN, but internet still ... Ping statistics for 127.0.0.1: ... another host on the network. ... Even the server and the laptop that the internet works on. ...
    (microsoft.public.windows.server.active_directory)
  • Re: Network configuration problem
    ... I removed the Remote Access/VPN Server role from the server and am now able ... I can also ping the server's local address from the laptop ... Nor can I get to the internet from my laptop. ...
    (microsoft.public.windows.server.networking)
  • Re: ICMP ping effecting network flow?
    ... you ping a server over the Internet while downloading a file from it, ... LBL to a specific server in Berkeley. ... and one of those is the "minimize delay" flag (see page ...
    (comp.os.linux.networking)
  • Re: Urgent! New router and big disaster
    ... The SBS DNS server, running on ... its IP it means that your problem is now DNS. ... forward ports to it reliably in the router. ... I should have been more clear about internet connection.. ...
    (microsoft.public.windows.server.sbs)