8d193171
XPost: comp.unix.programmer
From: grahn+nntp@snipabacken.se
["Followup-To:" header set to comp.protocols.tcp-ip.]
On Thu, 2010-12-30, David Schwartz wrote:
> On Dec 29, 9:26 am, "A. McKenney" wrote:
>
>> Based on other problems we have encountered,
>> I'm fairly certain that Solaris 10 returns from the
>> sendto() call before the message actually gets
>> sent (or not sent.)
>
> Of course.
>
>> OK, assuming that Solaris may fail to send a UDP
>> message after a successful return from sendto(),
>> where would one look for statistics on such lost
>> messages?
>
> If you need such support, code it. Have the other ends send you
> reception reports if you need them.
I didn't read the first postings as carefully as I should have, but I
assume/hope he *has* code to cope with packet loss, but sees losses
which are higher than he expected and wants to analyze the cause.
>> For messages lost on the receiving end, we run
>> "netstat -s" -- are there fields there that would
>> show messages lost on the sending host, too?
>
> Check your operating system documentation.
If Solaris is like Linux, it's hard to find such documentation, so I'm
not surprised he asked here.
> There is no requirement
> that it even have any way to establish this.
You'd expect a serious OS to count all drops before the link layer
*somewhere*, wouldn't you? Not per application, socket or port though
-- the stack loses track of such things pretty quickly.
My best bet would be netstat -s like he writes, and in particular the
UDP and IP (or IPv6) counters. Or some Solaris-specific alternative.
Also look in the packet filter stuff (whatever Solaris has instead of
Linux 'iptables -vL') because I don't think packets which get stuck in
the firewall show up in netstat -s.
/Jorgen
--
// Jorgen Grahn O o .
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
