Hi all,

Based on discussions I had outside of this WG, I would like to ask for
the possibility to specify a kind of "simple reliable syslog protocol".
RFC3195 is definitely well done work. However, I see some serious issues
with it:

- (Un)availibility of BEEP
As it looks, BEEP has limited availability to say the least. I have
posted a question on the BEEPwg mailing list last week asking for
library implementations. I got no response at all (except for Marshal
Rose who requested that new libraries should be registered at
beepcore.org). So as it looks, BEEP is limited to BEEPcore (which has
some serious issues right now) and Roadrunner (which is hard to use
without redisigning your whole application [and also hard to use for
closed-source projects]).

- Complexity of BEEP
I wonder if we will ever see an implementation of BEEP (and thus
RFC3195) in routers and especially low-end devices like printers and
small vendor's routers.

- Need to redesign your application
If I got it correctly, there is strong need to redisign an application
in order to use it with BEEP. Honestly, we do not want to redisign our
full product line just to add support for one additional protocol.

I have followed the relevant lists and so far have the impression that
most implementors stay away from RFC3195 for one or more of the above
reasons.

So in (current) reality, we are back to the usage of "traditional"
(maybe not even RFC3164) syslog if we would like to receive messages
from a heterogeneous set of sources. The good point is that we have a
widely accepted protocol and it performs relatively good most of the
time. But there are weaknesses. Most stem from UDP usage.

I hope that it is possible to specify a protocol that fills the gap
between RFC3164 and RFC3195. A *simple* syslog reliable protocol, that
provides:

- reliable transport via TCP
- a much larger allowed message size than 1024 characters
- support for non-western characters
- some more meta data (like the full-blown timestamp, but this might go
into the content area and as such might be outside the charter of this
WG. I am still mentioning it so at least some header changes can be
considered)
- maybe hooks for optional encryption and authenticy - but this might go
too far
- most importantly is easy to implement and does rely on right now
readily available libraries, which unfortunately means sockets, only...

I wonder if this WG would be willing to step forward into that
direction. I am sure such a protocol would not the least be as elegant
and advanced as RFC3195, but I think this is it's key advantage: being
simple, we would be able to get more implementors to actually implement
it. So we wouldn't achieve the 100% that RFC3195 brings, but maybe the
important 50% of it. Well, and that looks much more than we currently
have...

Keep in mind: my key point is that improovement is needed *now* (which
means quickly), so I am looking for a solution for the next few years.
Maybe this would even be an interim-solution until BEEP becomes actually
available...

Any feedback is highly appreciated,
Rainer Gerhards
Adiscon


Reply via email to