[CentOS] Yum/WGET/HTTP sourceforge etc.

Sun Apr 11 22:57:33 UTC 2010
tony.chamberlain at lemko.com <tony.chamberlain at lemko.com>

I am having a problem with 5.4 that I did not have with 4.5. The problem happens only sometimes but in specific
instances. Basically a summary of the problem is that certain network transactions timeout. The specfic instances
are with wget, rpm, http. The problem usually, but not always, occurs with pptp stuff. (NOT running pptp but getting pptp stuff).

For instance, the following command, which finishes in seconds on non-5.4 OS's:

wget http://pptpclient.sourceforge.net/yum/stable/rhel4/pptp-release-current.noarch.rpm

downloads about 20% then gets stuck. About 5 minutes later it downloads another 20% and
then gets stuck, etc. The same thing with rpm:

rpm -ivh http://pptpclient.sourceforge.net/yum/stable/rhel4/pptp-release-current.noarch.rpm

waits about 3 minutes and then gives an error. I think it does the same thing as the wget but
wget will keep trying, while rpm gives up. The error from rpm:

Retrieving http://pptpclient.sourceforge.net/yum/stable/rhel4/pptp-release-current.noarch.rpm
..five minutes later:
error: /var/tmp/rpm-xfer.DYY9e0: headerRead failed: hdr blob(7456): BAD, read returned 3568
error: /var/tmp/rpm-xfer.DYY9e0 cannot be installed

I can wget the above as I mentioned before and install it that way. Before I do it, yum works fine.
Afterwards, yum exhibits the same behavior of timing out (because it is using the pptp repository).
Also visiting the pptp web site from Firefox times out on certain pages.

I originally thought it was some problem with the pptp site, but I notice that log into hotmail.com
does the same thin (fine on other operating systems). A view with Wireshark on the wget (pptp)
shows the my machine receiving a reassembled TCPPDU from 216.34.181.96 (Sourceforge), sending an ack, receiving a
reassembled PDU, sending an ack, receiving, sending followed by the 5 minutes or whatever of nothing. Then sourceforge
sends an RST and a SYN and the process is repeated.

Here is what I tried. When I put the machine directly on an AT&T IP connection (12.147.X.Y) everything worked fine.
Same with Comcast on a direct link. The times I am having problems is when our router is hooked up to a Comcast
IP (70.88.X.Y) and assigns 192.168.5.X addresses to our machines. So when I was doing the above from 192.168.5.27
going through the router through Comcast is when I had the problem.

So it is probably something with the router, but it is hard to figure out since CentOS 4.5 and Fedora do not exhibit this
behavior, nor does 5.4 on most sites (mail.yahoo.com for instance). I did verify, at least from what I could, that ICMP
type 3 and 4 are not being blocked. If they were, the same problem would happen on other op systems. And I was able
to ping, albeit just locally, but we looked at the router settings and ping was not blocked.

Anyone else have this problem / know what might be wrong?