Robert,

I am fine with your text.

<slightlyOffTopic>
The reliability issue is not rooted in disasters or other uncommon
scenarios. As of my testing, confirmed this month by Martin, the
probability of message loss even in a simple server restart case is very
high. For a busy network, I'd say it tends toward 100%, though I cannot
provide hard evidence - I didn't bother to do a statistically large
enough sample to back that statement. After all, it is irrelevant if
there is a 80% or 100% probability - the point is that it is very high
(probability of message loss increases with traffic volume).

In any case, I expect that an operator of an enterprise logging system
will routinely see message loss (if he cares to detect it) inside a
system using -transport-tls. Of course -transport-udp will be much
worse. But the point is that an auditor may very well question the
completeness of the log data under *normal* circumstances. Martin has
found a trick to get some better reliability for these normal
circumstances, but the trick only works in a fast, lossless network with
low traffic volume and without bursts. Plus, it is racy. But it is the
best you can do.
</slightlyOffTopic>

All of this off-topic should NOT go into the document, but it is the
reasoning behind my proposal.

Rainer

> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
> Sent: Thursday, June 05, 2008 3:23 PM
> To: Rainer Gerhards
> Cc: Joseph Salowey (jsalowey); syslog; [EMAIL PROTECTED]
> Subject: RE: [Syslog] Subject Name verification policy
> 
> "Rainer Gerhards" <[EMAIL PROTECTED]> wrote on 06/05/2008
> 08:01:37
> AM:
> 
> > I think I should have been more clear. I meant a note along these
> lines
> > (and only these lines, without any more specifics).
> >
> > ###
> > It should be noted that this transport does not use
application-level
> > acknowledgments. As such, there exists situations where loss of data
> > may occur. This protocol is not suitable if a 100% reliable solution
> > is desired.
> > ###
> >
> 
> How about just
> ###
> It should be noted that this transport does not use application-level
> acknowledgments. As such, there exists situations where loss of data
> may occur.
> ###
> 
> Maybe someone will find or write a good reference article explaining
> the
> application level issues with TCP/IPv4, TCP/IPv6, etc.  Then an in
> depth
> explanation could be given once for all the many applications that
> care.
> People who are serious about this will need the full details.  The
rest
> deserve just the short warning.  I don't want them blindly telling
> everyone that the syslog group has officially said that they shouldn't
> use
> syslog-tls because it isn't reliable.  That is what would happen if we
> include that final sentence.
> 
> There are unsophisticated users who routinely demand 100% reliability.
> I
> deal with them by asking about how they plan to handle power failures,
> hurricanes, earthquakes, floods, tornados, etc.  I know that when I
> cared
> about such stuff I could quote the local power grid reliability
> statistics
> off the top of my head, and could explain our decision tree on the
size
> of
> the fuel tank for the auxiliary diesel generator.  I was also on top
of
> disk drive reliability figures.  This was before RAIDs and hot swap
was
> very expensive.  It turned out that downtime for disk drive problems
> was a
> major issue as you went below 10min/year of downtime.  Their response
> to
> the basic disaster questions establishes what kind of response to give
> for
> applications level data loss.
> 
> R Horn
_______________________________________________
Syslog mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/syslog

Reply via email to