…or more precisely already failed, is failing right now or definitively will fail. i’m getting this lesson one more time in a hard way. number of anomalies i’ve encountered during last 3 weeks exceeded critical threshold.
two most critical are:
- htb based traffic shaping system stopped to work correctly after hardware migration from sk98 gigabit card to e1000 card on some linux router. namely http downloads [that go over transparent squid proxy] ware not limited anymore. after numerous experiments i found it has something to do with tcp offloading that fortunately can be disabled with simple:
ethtool -K eth0 rx off tx offand then magically problem is gone… - dnat based access to w2003 file server does not work in deterministic way. or… maybe it does not work at all with some exceptions when it works. symptoms? you map one of network shares from w2003 server, you access it. after few seconds mapped share becomes ‘disconnected network drive’. file transfer from this share can succeed.. or can fail somewhere in the middle. i still do not know if it’s a general issue with windows networking and nating or maybe with particular kernel version. i ended up skipping linux router in a chain and multihoming w2003 box+adding some static routes on it. users are satisfied – after all it works, i’m completely puzzled – after hours of sniffing traffic at both ends of linux router doing dnat i cannot see any anomaly, to make things worse tcp traffic is not interrupted, no fin/rst flags are set, connection remains established but yet transfer is broken or drive gets disconnected. obviously after multihoming w2003 no anomalies are experienced anymore. and yes – it works perfectly fine when traffic is just routed.
i dont mind sharing painful experience of well described bugs, that authors of code are aware off. but it’s not the case. i haven’t found single entry on google groups describing similar anomalies with [very recent] versions of kernel i’m using.
deep breath…
after all it’s not that bed. i recovered from 3 weeks of being semi-sick. although i dont like winter, but weather out is fine – it’s sunny, -5 degrees. perfect conditions for some weekend walks and taking pictures
for the record. tests with windows networking were carried few more times on different pc hardware with 2.6.9 debian kernel+few self made 2.6.15 kernels. all of them failed. apparently windows networking does not like snat/dnats. it’s sad but seems to be true.