I'm running OS 10.1 on one of the new TiBooks. I'm connecting to the internet via an SMC wireless router (7004AWBR; latest firmware) attached to a 56K modem. With both wired and wireless connections, I'm getting terrible performance under OS 10.1; through-put tops out at under 1.5 kbytes/sec. Other machines (running various version of Windows) on the same network perform fine. If I switch to 9.2 it also works fine, so I don't think it's a hardware problem. If I set up one of the windows boxes to do routing/NAT it also works fine. So I think it's something to do with the OS X.1/SMC combo. Problem is, where to start? Any hints would be much appreciated.
slow tcp/ip & SMC router
- Thread starter iwanna
- Start date
scruffy
Notorious Olive Counter
Have you looked at http://www.practicallynetworked.com ? Lots of reviews and tips there, might be somewhere to start.
Also, is that 7004AWBR or 7004 ABR? Is there a difference? I'm thinking of getting a 7004ABR myself, so this is of interest to me...
Also, is that 7004AWBR or 7004 ABR? Is there a difference? I'm thinking of getting a 7004ABR myself, so this is of interest to me...
Are you connecting to the router via DHCP or statically? If you are using DHCP, make sure you have the correct DNS settings in your router. If you are using a static setup, try putting in your router's IP in the TCP/IP settings for the DNS server, and see if that works.
I'm using DHCP, and the DNS server (as specified by resolv.conf) looks correct. Even just FTP from the shell is slow. One of my test involves downloading a 1mb file via FTP from my website, and it was significantly slower than running the same test from one of my windows box. Looking at tcpdump output, I don't see anything that is hogging bandwith. I'm assuming that tcpdump shows *all* activitiy on an interface?
I'm pretty sure that tcpdump does show all activity. I seem to remember someone else having a problem like that but I can't recall the solution. I'll see if I can find it for you.
It appears that the SMC7004AWBR is not correctly handling RFC1323 "Large Windows" extensions to TCP. Looking at the netstat dump I saw a lot (something like 30% of received packets) of TCP bad checksum errors. Turning off RFC1323 did the trick:
sudo sysctl -w net.inet.tcp.rfc1323=0
Found some interesting links in my research: try googling "bsd tcp tuning"
sudo sysctl -w net.inet.tcp.rfc1323=0
Found some interesting links in my research: try googling "bsd tcp tuning"
It seems like you know more about that then I do cause I have no idea what any of that is, but glad you got it fixed.