Darryl, On Tuesday 19 December 2006 12:01, Darryl Gregorash wrote:
On 2006-12-19 12:47, Randall R Schulz wrote:
I wonder how much we can discern about the hosts on the other end when the failure occurs? We know that openSuSE 10.2 is one part of the equation, but since most peers do work, perhaps there's something instructive about the ones that do elicit the problem.
The OP has reported that reverting to the 2.6.16 kernel allows him to access the sites. Everyone reporting the problem so far has a 2.6.18 kernel, and the same problem has also been mentioned with other distros using that kernel version.
The TCP stack idea seems most on track, but it is curious that it doesn't affect everyone running 10.2. 32 vs. 64 bit doesn't appear to answer it, or does SMP vs. non-SMP.
I'll stay a bit on it.
Likewise. I just know that one of these days it'll crop up for a site I really need to access.
Felix suggested it is perhaps time for bugzilla.
Well, someone else can do _that_ particular bit of penance, this time. I tried to explore the possibility of a hardware ("offloaded") checksum generation bug. I found some reports of past bugs in some of the Ethernet drivers relating to this capability, and clearly it's being used on my system. So far I have not been able to find out how, if indeed it's possible, to disable the use of the hardware checksum offloading in the "skge" driver. Does anyone know how to disable hardware- (NIC-) based packet checksumming, either specifically for the skge driver or in general? However, just now I came up with another data point that pretty much refutes the erroneous hardware checksum generation hypothesis: I modified my system's configuration so as to use the other NIC, which is not the same kind of hardware and does not use the same device driver (before the system was using the skge driver for external traffic, now it's the sky2 driver). But the symptom remains unchanged. So whatever it is, its either something these two drivers have in common (they're not wholly unrelated, I don't think) or it's something that happens higher up in the protocol stack where the data paths are by all NICs.
I think I agree with him. The rabid sysclock issue is another area where reverting to an earlier kernel resolved the issue, and which is also one which affects only a few. I have been trying to keep an eye on things which could suggest a common denominator in all this, without success so far.
I don't have any real-time clock issues on this system. Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org