Messages in this thread | | | Date | Mon, 1 Apr 1996 10:18:48 +0100 (BST) | From | Steve Davies <> | Subject | Re: Slow PPP (I had more than one problem) |
| |
On Fri, 29 Mar 1996, Linus Torvalds wrote:
> > Perhaps there is, but WHAT ?????? I've checked the link quality and > > everything looks perfect. It just seems to be going SLOW... Packets come > > in, packets go out, LCP stuff is exchanged every few seconds, there are NO > > errors reported by netstat, pppd-debug or kernel debug. > > What does /proc/net/dev say? Any errors?
I get a FEW errors (30 dropped packets in about 20,000) which I believe are caused by FCS disagreements, but such a small percentage isn't going to bother me, particularly as none of them appeared during my throughput testing.
I've solved part of my problem recently by turning bsd_comp back off. It got turned on by accident when I upgraded to modules-1.3.69 and re-did my conf.modules incorrectly. By also correcting my asyncmap I can now get data to stream at full speed. The FTP test I did last week now gives similar results for both Windows and Linux.
The problem I still get is with "interactive" TCP communications. NNTP being the best example. My 5.5Kb/s News transfers are down at about 2.4Kb/s, and have been since about the time this problem was reported. Using tcpdump on various communications seems to show that the begining of any exchange is constantly troubled with retransmits, but that things quieten down once data starts to flow one-way (apart from regular acks of course) - With NNTP, the constant request-send format of the exchange means that things never seem to settle down.
I forgot to bring the tcpdump output in to work, but if this problem persists, I'll try and find a good example of this.
> The silly thing is that the 72->73 transition (which people have told me is the > problematic one) doesn't seem to contain _anything_ that would explain it. It's > there somewhere, obviously, but certainly not easily seen.. > > Linus
I completely agree. As I said in a previous E-mail I've trawled through kernels 1.3.60+ without finding anything untoward...
Steve
| |