On Wed, 2006-07-05 at 11:32 +0200, Balazs Scheidler wrote:
> On Tue, 2006-07-04 at 05:27 +1000, Darren Reed wrote:
> RFC3195bis is a nice idea, even though I previously disliked that
> protocol myself. I'll have to reread that RFC to form my opinion.

I have now refreshed my memories on RFC3195, and my technical opinion is
as follows:

beep in general:
- I like its extensibility and multiple channels, and its support for
optional authentication/encryption
- I don't like its complexity as we have to think with embedded devices
in mind
- depends on a whole lot of things, and as I see generating the XML
constructs should use a real DOM implementation, as constructing XML
with sprintf is asking for trouble (ie: escaping)

syslog/raw profile:
- this profile does not offer anything more than simple syslog over TCP
as currently implemented in syslog-ng and other products, it has
significant overhead though: 
  - bandwidth wise: about 30 octets per message, 
  - code wise: the beep framework (beep itself, xml, character sets,
entities etc), this can be simplified significantly by supporting a
single channel only, always using utf8 and hand-crafting XML with
sprintf/parsing it with sax/handmade parser
- this profile does not solve record termination, where the latest
consensus was to be completely transparent for any unicode character, in
syslog/raw CRLF line termination is used when multiple messages are
included in a single BEEP frame.
- the profile permits a maximum of 1024 bytes per message

syslog/cooked profile:
- I like the fact that every hop in the message path is properly
identified, and proper identification of senders can be performed,
- I like that application layer ACKs are present, however I think that
it is too verbose (about 30-35 bytes to acknowledge a single  message,
about 35-40 to nack it)
- I don't like that the new format still uses BSD timestamp, no year and
no timezone information

So as I see, there is absolutely no point in implementing syslog/raw,
using the existing protocols is enough for legacy support and
implementing syslog/raw is not any easier than using syslog/cooked. And
it has issues with record termination and maximum message length.

syslog/cooked could be updated with the latest header changes in
syslog-transport-udp and with some adjustments it could provide a nice
platform for new protocol features in the long run.

What this boils down, is that I could support RFC3195bis, especially if
we decide that we are bothered by the IPR claim of Huawei.

-- 
Bazsi


_______________________________________________
Syslog mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/syslog

Reply via email to