On Mon, 30 Jan 2006, Trond Myklebust suggested tentatively:
> On Mon, 2006-01-30 at 16:55 +0000, Nix wrote:
>> On Sun, 29 Jan 2006, Trond Myklebust stipulated:
>> > As a general rule of thumb: if tcpdump/ethereal can see the reply on the
>> > client, then the engine socket should see it too. If tcpdump is indeed
>> > seeing those replies, you should check the RPC code by
>> > setting /proc/sys/sunrpc/rpc_debug to 1.
>>
>> tcpdump is seeing them.
>
> The complete packets, including all fragments?
Will check. Maybe one fragment or something is getting lost
(but if so, it's getting lost terribly *consistently*, as in,
every single time).
>> I *guess* that the `failed to lock transport' is the underlying error...
>> time to add some debugging and find out what task is locking the
>> transport. Back soon, must rebuild the kernel and reboot to clear this
>> lock ;)
>
> Nope. The congestion window is 1 request, and you do indeed appear to
> have one request on the "pending" queue. The problem in the above trace
> is a complete lack of data_ready messages, meaning that the socket is
> never seeing any complete packets come in.
O*kay*. I'll do some tcpdumps on both sides and compare them.
... you are quite right. I'm mounting with an rsize and wsize of 32768
(i.e., the default negotiated between a recent Linux NFS client and
kernel server), and, completely consistently, 31504 bytes are sent and
*only 27208 bytes are received*. This is so consistent that there's no
way that this could be due to network congestion (unless it's getting
jammed up inside the receiving NIC or something: it's an NE2K card and
they're rather crap so maybe the card is just too slow: my determination
to replace the card has just gone up a notch. But nonetheless if it was
getting lost by the card I'd see TCP retransmissions, which `netstat -s'
assures me I do not).
This explains why I don't see a problem with DNS: the number of DNS
packets >27208 bytes can be counted on the fingers of one foot.
Captures are available, but I gathered about half a minute's traffic
so they're quite large (1Mb apiece).
Tim? Any ideas? Is anyone else with this card seeing this problem?
--
`I won't make a secret of the fact that your statement/question
sent a wave of shock and horror through us.' --- David Anderson
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
