[EMAIL PROTECTED] wrote: > > On Sun, Mar 16, 2008 at 12:35 AM, quekky <[EMAIL PROTECTED] > <mailto:[EMAIL PROTECTED]>> wrote: > > I do not have problem with UOB on F8 (other then 64bit IcedTea not > working and I have to use 32bit FF with sun java) > > However I'm unable to load any page on OCBC and all their sites > (FFP/Finatiq) on any ISP (I've tested singnet/pacnet/starhub and even > from the datacenter in Singtel building) I did try a simple wget, and > it only loads 1 packet and the rest is gone. From wireshark, I only > see 1 packet coming in, the remaining are somehow lost in cyberspace > > It seems to be affected either the newer kernel (2.6.xx) or networking > stack or something else, because even RHEL5 wget also doesn't work, > same for my friend on ubuntu is unable to load too. > > Strangely when I use a proxy (any proxy, be it singnet, pacnet, or my > own squid) it works fine > > > I previously had similar issues with hdb when accessing it using my > own proxy server. I eventually traced that to some tuning I performed > on the networking stack, where I'd increased the tcp buffers to > something very large (perhaps 1MB or 10MB per connection) to optimise > speeds over high speed, high latency connections. Perhaps their > firewalls disagreed with my buffer settings.
Another common cause of this type of problem is blocking ICMP fragmentation needed packets. Happens quite often when firewall admins filter ICMP without understanding the consequences. If the foreign network has a smaller MTU (quite common if they are using tunnelling of some sort) and the resulting ICMP is dropped, then this can result in the symptoms you describe. You could try setting a smaller MTU on your Ethernet device - say 1400 or try disabling path mtu discovery: echo 1 > /proc/sys/net/ipv4/ip_no_pmtu_disc _______________________________________________ Slugnet mailing list [email protected] http://www.lugs.org.sg/mailman/listinfo/slugnet
