I wonder ... are the boxes inside purely WIN boxes? -- What sort of outside connection do you have? -- Do you have a rule forcing MTU size on the outside interface? -- I've found that some WIN installations for some reason dont do MTU discovery by default.... I've had to turn it on ... although I suspect that it was turned off by previous administrators... (just a similar, although odd situation I've run into in the past...) Alistair Tonner On 2002.10.13 09:08 Svein E. Seldal wrote: > Hi, > > I have read about a couple of previous discussions on the issues of > using iptables together with FTP and acheiving only really *slow* > throughputs. All Q's that I've read concludes that the remedy is to > either passthough auth (ident) packets or to reject it. > > I have a linux (2.4.18) running iptables. It is a router and FW, and > is NAT'ing the internal address range (192.168.0.x) to the external > range (1.2.3.x). ip_conntrack, ip_conntrack_ftp and ip_nat_ftp are > all loaded in the router. It is a 266 PII with 128Mb memory. > > For any machine that are placed on the inside, there are two methods > of getting to another machine on the inside: either using the direct > internal address 192.168.0.x or by using the external address > 1.2.3.x. The net result is the same, but the latter is routed through > the router (and will always be accepted). > > Now, the problem is this: I open two FTP connections from the client > machine 192.168.0.10 to a server, the first to 192.168.0.5 and the > other connection to 1.2.3.5. Again, this is the same machine, only > one of them is going via. the router while the other is not. > > The connection going via. the router is slow, very slow. The problem > is that is does not matter if I DROP, REJECT or ACCEPT the auth > (ident) port. It still is as extremely slow -- direct connection > takes 20 mins, while the routed version takes 2-3 hours!!. > > My hypothesis could be that the ip_nat_ftp module is causing the > considerable delay, since it must read all FTP data (I'm unsure about > FTP-data though). Another theory could be that the routing memory of > the iptables/kernel is too small. How can I increase it? > > Any one else got any ideas? (I've got a lot of angry customers on my > back...) > > > Thanks, > Svein Seldal > > >