Will BEEP Simplify Network Programming? 195
hensley writes "There is a (not quite) new effort by the IETF to standardize
a framework for network applications, called BEEP, the Blocks Extensible Exchange Protocol. Standardized in RFC3080, it takes care of all lower level tasks an application level protocol has to like framing, authentication and capabilities negotiation in a modular and lightweight way. In the current issue of Internet Packet Journal (a quite nice and free-as-in-beer technical publication by Cisco) is a well written Introduction to this framework. Why isn't anyone adopting this protocol besides some Java libraries like beep4j and PermaBEEP and a C library called RoadRunner. I couldn't find any applications based on this protocol, regardless of it's promised capabilities. Is everybody still inventing his own application layer protocol?"
slashdot these days... (Score:1, Offtopic)
Is a simple question mark too much to ask for?
Re:slashdot these days... (Score:1, Offtopic)
Re:slashdot these days... (Score:1, Funny)
Re:slashdot these days... (Score:1)
I used it once... (Score:4, Funny)
Agreed...BEEP is great work (Score:5, Informative)
BEEP is Blocks Extensible Exchange Protocol (RFC 3080 [ietf.org]). More details can be found at here [beepcore.org].
When we were designing RTSP [rtsp.org], we looked for something like this, and at the time, the options weren't very appealing. We ended up using HTTP as a quasi-base protocol. I think it was the best solution at the time, but had BEEP been available, we'd have used it in a heartbeat.
Not a Good Sign (Score:1)
Location:
Internal Servlet Error:
java.lang.NullPointerException
at org.apache.jasper.compiler.JspCompiler.generateNe
at org.apache.jasper.compiler.JspCompiler.(JspCompil
at org.apache.jasper.JspEngineContext.createCompiler
at org.apache.jasper.servlet.JspServlet.doLoadJSP(Js
at org.apache.jasper.servlet.JasperLoader.loadJSP(Ja
at org.apache.jasper.servlet.JspServlet.loadJSP(JspS
at org.apache.jasper.servlet.JspServlet$JspServletWr
at org.apache.jasper.servlet.JspServlet$JspServletWr
at org.apache.jasper.servlet.JspServlet.serviceJspFi
at org.apache.jasper.servlet.JspServlet.service(JspS
at javax.servlet.http.HttpServlet.service(HttpServle
at org.apache.tomcat.core.ServletWrapper.doService(S
at org.apache.tomcat.core.Handler.service(Handler.ja
at org.apache.tomcat.core.ServletWrapper.service(Ser
at org.apache.tomcat.core.ContextManager.internalSer
at org.apache.tomcat.core.ContextManager.service(Con
at org.apache.tomcat.service.connector.Ajp12Connecti
at org.apache.tomcat.service.TcpWorkerThread.runIt(P
at org.apache.tomcat.util.ThreadPool$ControlRunnable
at java.lang.Thread.run(Thread.java)
Not so impressed (Score:5, Interesting)
BEEP's main proponent, Marshall Rose was one of the main wheels in the OSI project. So much of the initial buzz came from his name alone. People were talking about the protocol before they read the drafts (oh yes that is normal for the IETF).
I do have a bunch of quibbles technically. First using XML is a good idea, Using the obsolete SGML DTD mechanism to describe the protocol sucks. I think Marshall started to suplement the DTD with schema fragments but that makes things worse, not better, we now have two specs in one document, the schema version which is what people will implement and the DTD version which is normative.
The other problem is that SGML is a real baaad choice for an encoding at that level. The main complaint about http is that the encoding is too verbose leading protocol exchanges to require multiple round packets instead of one. BEEP does not address that problem.
Politically BEEP has bigger problems, first being that IETF does not have as much influence as it might appear when it comes to promoting new protocols. There haven't been very many IETF protocols that started in IETF process and took off like wildfire in the past ten years. HTTP took off and was brought into IETF process, same with TLS (SSL). Most successful IETF protocols had a userbase before the working group was formed.
The problem with BEEP is that Marshall did not start with a constituency who had a problem that BEEP was the solution for. Instead he wrote the protocol and then went off looking for consumers. So we start to see Marshall popping up at random in working groups like SACRED peddling the BEEP Kool-Aid. The problem being that if you are doing a researchy protocol like SACRED the last thing you should be doing is layering it on someone else's research.
After this happened a few times Marshall started to alienate folk like myself who might be interested in BEEP as an option but certainly were not going to allow him to insert himself onto our critical path.
The other problem is the nature of the IETF these days. The problem is that they talk a good talk about being open and such, but it is really an old-boys club. The old-fart faction is strong on the IESG and IAB, they have known each other for 20 years and they don't want anyone messing with their turf.
In theory the IETF process is open. In practice there are a bunch of shadowy cliques who make the real decisions in private. BEEP got to RFC status in record time because it was proposed by an IETF insider. Problem is that the IESG does not have much influence with the people in the Web Services world which is where all the interest in XML based protocols is at the moment.
Most of the people I see at W3C and OASIS Web Services meetings have no IETF experience at all. Of those who do, none are IETF insiders and so an endorsement by the IESG does not have much force.
For BEEP to take off it really needs an endorsement by one of the heavy hitters of the Web Services world which basically means IBM. Microsoft or Sun. I don't think that is very likely because everyone knows that there is a lot of work to be done to make Web Services work and there is simply nothing to be gained by putting BEEP on the critical path. People are more interested nailing down WS-Security, SAML, XKMS, geting WSDL to work and such.
Re:Not so impressed (Score:5, Informative)
combined with tomcat. (if someone has some hints on making
netbsd/apache/tomcat run more robustly in the face of significant load
-- or if you can tell me how to get the thing to fail gracefully instead
of tossing its stack -- please drop me a private email -
mrose+mtr.slashdot@dbc.mtview.ca.us - thanks!).
second, i have to praise this post for the way it seemlessly blends a
little bit of truth, a little bit of wisdom, along with a healthy amount
of ignorance and untruth. (my favorite untruth: marshall as the OSI
guy. the actual truth: marshall killed OSI in the IETF on November 4th,
1993 with the "roadkill in motion" speech.)
it is easy to agree with the fact that beep suffers from both political
and technical hurdles. as usual though, folks can disgree on the actual
details.
on the technical front: the beep specs use DTD not schema -- there
aren't any schema fragments in either rfc3080 or rfc3081, you must be
thinking of something else. the choice of DTD over schema is simple:
DTDs are ugly, but schema sucks. that explains why everyone has their
own pet language for defining the acceptable syntax of an XML
document. if schema was a winner, we'd be seeing fewer alternatives
instead of more.
everyone has their own "main complaint" about http. i hadn't heard the
verbose encoding one before, but maybe we should ask Keith Moore to add
it to the list (cf., rfc3205).
certainly, there's a lot we can agree on with respect to the political
front, so i'll just focus on the part we don't agree on.
the actual consumer for beep is the ietf. there used to be this joke
that the apps area invented cloning, because all working groups formed
argued exactly the same issues, over and over and over again, regardless
of the problem to be solved.
with beep, no one gets to argue those things any more (e.g., how to
frame packets), instead they get to go off and presumably argue things
specific to their application domain. (if folks want to understand the
reasoning behind this, check out rfc3117.)
the sacred working group, that you're so pissed off about me cutting
into, is a perfect example. a bunch of guys focused on security issues,
trying to write an application protocol. sorry, wrong skill set.
contrary to popular belief, i don't need to go looking for trouble. in
this case, it was a couple of ADs leaving an early sacred meeting,
shaking their heads, and then asking me to beat some sense into some
folks.
if you're unhappy that i stuck my nose in your business, then all
i can suggest is you get more clueful in the application design space,
so "the management" doesn't feel they have to go out and get you
help. particularly help that you don't like, and especially help that
would rather be doing other things with other people.
beep isn't research. it's a "best hits" collection of stuff dating back
to 1981 that's known to actually work. the only new part of beep is that
it got integrated into one coherent spec. and that's the reason that the
iesg approved it after only a year. they were familar with what it did,
how it worked, and they had a problem.
of course, you are perfectly free to attribute this to an "old boys
network" (i'm sure allison would appreciate that). when i find the
actual "right wing conspiracy", i'll be sure to sign up. it will
certainly reduce my frustration in dealing with the 100's of procedural
hurdles that get thrown in my way at the ietf.
finally, as far as web services go, well, let's just say that those guys
could learn a lot from what xerox did back in the early 80's. in a few
years, they may actually have something that works half as well as what
xerox did...
Re:Not so impressed (Score:5, Interesting)
Ah so you are claiming that the Marshall Rose who wrote The Open Book : A Practical Perspective on Osi [amazon.com] is a different Marshall T. Rose. No sorry, Marshall you made a major technical contribution to OSI, or at least he claimed to have done so on the jacket cover of the copy I read. You did not 'kill it' at the IETF, OSI was killed in the marketplace long before 1993. The speech had the impact it did precisely because you knew the OSI stack.
on the technical front: the beep specs use DTD not schema -- there aren't any schema fragments in either rfc3080 or rfc3081
Well Duuuhhh read what I wrote. I checked the RFC just before posting and saw that. Schema may be 'ugly' as you put it, but none of the major XML programming platforms are based on DTDs, they are all based on schema. And BEEP is dead at Microsoft, Sun and IBM without schema, cold stone Deeeeeeaaaaadddd.
with beep, no one gets to argue those things any more (e.g., how to frame packets), instead they get to go off and presumably argue things specific to their application domain. (if folks want to understand the reasoning behind this, check out rfc3117.)
No, we have a five minute argument on whether to use BEEP or not, reject it and carry on.
the sacred working group, that you're so pissed off about me cutting into, is a perfect example. a bunch of guys focused on security issues, trying to write an application protocol. sorry, wrong skill set.
I wrote a fair bit of HTTP, I don't think that you have the right to go arround saying who is and who is not competent to write application protocols. If you want to get into a reputaion war you are going to loose this one. I think that BEEP is very naive when it comes to the problems that arose when it came to layering application protocols on it. I suspect that like LDAP and X500, by the time BEEP has been extended enough to be useful it will look like HTTP.
if you're unhappy that i stuck my nose in your business, then all i can suggest is you get more clueful in the application design space, so "the management" doesn't feel they have to go out and get you help. particularly help that you don't like, and especially help that would rather be doing other things with other people
I only attended the one SACRED meeting and your comment on 'the management' is quite illustrative of the George W. Bush style crony-standards process the IETF is becomming notorious for.
finally, as far as web services go, well, let's just say that those guys could learn a lot from what xerox did back in the early 80's. in a few years, they may actually have something that works half as well as what xerox did... /mtr
Standard old fart response 'we did it all twenty years ago sonny', 'and we did it better'. Yeah and you should have seen the anti-gravity machines we made twenty years ago.
I don't much care for the arrogance of the IETF 'management' as you call them. I certainly don't appreciate folk who think that they have the right to make the type of off-hand blanket pronouncements on other people's work that you and they make habittually without backing it up. Your Xerox comment is absolutely typical of IETF old fartism, you want to have the right to be dismissive, you don't have the technical arguments on your side. So instead of detailing a real technical issue you allude to an earlier system, the more obscure the better. The message: 'I am too important to have to justify my comments but I believe that you are not competent to work on this problem'.
Your comment on Web services only illustrates that you really don't understand what is going on, what people are trying to achieve or why previous efforts such as CORBA failled. I am not going to explain why or how Web Services are different because I am faaaar toooo important. I may not be old enough to have achieve old fart status but I can certainly play the part on the net.
Still you are right on one battle, if the IETF is to regain some relevance at the upper end of the protocol stack adopting XML as the way to author RFCs is the only way forward. However the IETF is going to have to do a lot more than produce its documents in a format that does not look like utter crap from a teletype before I am going to take any standards there. I want a genuinely open and geuinely transparent process. I want standards groups to complete in 18 months, not 10 years - and yes it is possible, the SAML group I was a member of developed the basic specs which have been adopted as the basis for liberty in 18 months.
Re:Not so impressed (Score:2, Interesting)
Sheesh, how on earth is knowing about OSI before publicly trashing it a crime? For people trying to make sense of the burgeoning pile of manure that was the OSI movement in 1990, MTR's books were a way to at least work out WTF OSI was on about, because it wasn't a picnic extracting this information from either: (1) the standards (unless you had a whole lotta time and masochism on your hands); or (2) the vendors (who just said "this is what you have to buy to be futureproof, shut up and buy it".)
MTR may have attempted to inject some sanity into the OSI standards process at some stage, but given that many governments were STILL treating OSI as manifest destiny in the early 90's ("it's definitely the future, we just don't know when it's turning up in a usable form"), this could be seen as an effort make the best of a bad lot on behalf of the people who were apparently going to get it inflicted on them, failure in the marketplace or not, rather than wholehearted support for everything OSI stood for. Jump in my time machine and take a job as an IT Officer in the Australian Public Service in the early 90's, then you might see this point a bit clearer. Back then, we honestly thought we were condemned to live with this crap forever, courtesy of the Government OSI Profile (remember that?). MTR's books were very very useful for arguing with our management in an informed manner about why OSI would not actually solve our problems. And it was pretty obvious if you actually read his books that MTR was not a big cheerleader for OSI.
Re:Not so impressed (Score:5, Insightful)
I'm sorry, but I can't buy much of this at all.
BEEP is a ridiculous waste of time which equates to XML-encoded-TCP over TCP/IP. The many unsubtle problems that causes should be obvious; if they aren't, here are some:
You point out that SGML is a bad thing for encoding "at that level", but in the same breath say that XML is a good thing. Since XML is SGML-compatible, and largely employs the same syntax (but doesn't allow SGML short-cuts like leaving out closing tags), I fail to follow your argument.
And in your musings about the IETF, has it ever struck you that maybe the IETF also think that adding an incomprehensibly slow transport layer on top of an existing and widely supported transport layer is a shit idea? BEEP is NOT an application level protocol, even if it wants to claim to be one.
Re:Not so impressed (Score:5, Informative)
Re:Not so impressed (Score:2)
If the IESG thought that it is strange that BEEP should be allowed to progress through the process at such a breakneck pace and then have ADs going round telling others they have to give reasons why they are not using it.
You are right about BEEP really being a transport layer protocol. Actually in the ISO model Marshall claims now to know nothing about it would be 'presentation'.
There are several reasons why such a layer is useful, the main one being that the old farts were not as clever as they think they were. TCP has a lot of problems and you often end up wanting a protocol layer to insulate your application from them. There are all sorts of hideous things that happen if you don't close out TCP/IP sessions correctly for instance which the Berkley sockets library won't fix for you.
The other reason people want a presentation layer is that TCP has pretty wierd performance characteristics. In the early days of the Internet you could litterally bring the system down by sending packets in faster than people could consume them. The solution to this was 'slow start' which essentially means that sessions start out at a slow speed and only ramp up the rate of packet submission as they get feedback indicating the recipient can handle it. All good stuff eh? Well yes but the parameters were set in the days of 300 baud modems and are completely inappropriate for broadband.
What this means is that once you set up a connection you really want to keep it going. That can be problematic however since simply serializing all your requests into one channel can get real ugly when you are interleaving 32 byte instant messages onto realtime video streams.
The way to fix a lot of these problems is to fix the TCP protocol. There is no technical reason why slow start could not be designed with memory so when you connect to CNN.com it remembers the speed it used last time. That would break the current RFC for the Internet standard but it is what a lot of vendors are actually doing on the quiet.
What the IESG and IAB do not get is that a lot of the Internet are regarding them as damage and routing arround them. The IETF is collectively working in interupt driven mode, people propose working groups, they do stuff in response. What I do not see is a group of people stepping back and examining the internet architecture as a whole and working out ways to improve it. We just have a bunch of incremental fixes.
In any IETF discussion the status quo is always accorded a ridiculously high status, even when it is clearly broken.
You miss the point about SGML encoding vs SGML data model. The SGML encoding of XML is simply an artifact. If you want to apply XML at that layer in the stack you should write a better encoding.
Re:MinXML (Score:2)
Thanks, this is very interesting! I have a page [eastcoast.co.za] where I've been messing around with some thoughts about XML and how it should have been - minXML gives me some more material to think about.
Personally I prefer having the structure represented by tags, and data only in attributes. It makes it a little easier to represent certain categories of information (IMHO).
Re:Not so impressed (Score:4, Interesting)
I am not one of the Editors, I am listed as a principal contributor before the group members.
I've met most of you, and like all of you, but quite frankly, at least one of you has a conflict of interest when it comes to opining on BEEP.
Not so in my case, I was not big into HTTP-NG, I was not involved in DIME or Multiplex. I am not HFN which should be bloody obvious if you know him. However I am a friend and my company does a lot of work with his (Microsoft), but I also do a lot of work with Sun.
As for the IETF being an old boys/girls club...well, yeah. Welcome to politics. I think the IETF has generally been a reasonably equitable old boys club.
It is reasonably equitable until you actualy try to get things done. Look at DNSSEC, they have been trying to deploy the thing for ten years. In the meantime the .com zone has grown so large that the scaling bug in the design of the NXT record means that the protocol is not going to be deployed by the registrar for those zones without modification. A fix has been proposed for two years and has been agreed on the list. But the fix is currently blocked by secret discussions between the 'DNS Directorate' which is a closed and entirely unaccountable group. The not so obvious strategy being to delay OPT-IN until the DS flag day has gone through and then reject OPT-IN as requiring a flag day. Just about the only reason that strategy makes sense is if the IESG wants to delay DNSSEC for at least a year and risk a lawsuit.
On IPSEC the group is currently facing two problems that have been known for five years. First IPSEC was deliberately designed to make it impossible to use through NAT devices. I remember the comments in the WG at the time, people took it as a badge of honor to sabotage NAT - pretty lame when one of the major reasons NAT is needed is a design blunder by the IETF itself when it choose the IP address space to be smaller than the human population of the world - and on reliable authority this was done to allow an address to fit into a single register.
The other problem with IPSEC is that they ignored the problem of PKI and so we now have a wildly successful deployment of IPSEC for VPNs, a function which it is particularly ill suited for, and almost no extra-net or genuine internet deployment.
At any rate, I encourage people to do some digging themselves on this. I can't say that my experience is overly deep in this area, and I certainly haven't tried to design a protocol on top of BEEP, but based on what I know, BEEP looks like a pretty good idea.
I haven't tried to implement a protocol on top of BEEP either, and I am even less likely to do so given Marshall's outburst earlier in the thread.
My point about old boys network is that I have no confidence that BEEP is ready for prime time. I will not be confident until someone actually does build a protocol on top and demonstrates the bugs have been ironed out. HTTP looked pretty straightforward until people tried to use it.
The IETF is in severe danger of becomming irrelevant, as attendence at recent meetings is deonstrating. The much bigger problem with the IETF is that they really don't feel any responsibility to anyone but themselves. They take their role as supreme Internet standards body for granted.
The average IETFer has pretty much a slashdot mentality. They have plenty of contempt for technology have nots while curiously tollerating computing environments from the stone age. MIME is an IETF standard but send an attachment to an IETF mailing list and the old-fart faction complain that their antiquated mailer can't handle it. Send HTML mail and they will have appoplexy. (And if you think HTML is a bad idea I presume you read Slashdot in the no-graphics, plaintext mode).
The average IETFer does not think that they have an obligation to the community of Internet users or the vendors that support them. So holding up a protocol for a year while an AD who does not have a clue about security 'gets comfortable' with a security modification advocated by the Security ADs, well that is not a problem. Designing a protocol that breaks through NAT, well that is not a problem, and so on.
I am certainly not antithetical to the aims of the IETF. However there is a reason why I helped take the Web standards out of the IETF and set up W3C, there is also a reason why I am currently helping take standards from W3C to OASIS. There is a free market in standards bodies and the IETF is in dire need of a severe kick up the pants.
Re:Not so impressed (Score:2)
Hah, the only time Henrik talks this way is after too many blue drinks with Mario.
Re:Agreed...BEEP is great work (Score:1)
I originally intended to dump my proprietary protocol code, but it turns out easier to maintain what I already have than to get BEEP to even function.
Must not be a programmer... (Score:2)
Re:Must not be a programmer... (Score:1)
Back to network design school with you!
just a quick example:
HTTP is a protocol, a web browser is a program that uses HTTP to transfer files.
Who wasn't a programmer? (Score:2)
Um... You remember what API stands for right? That'd be "Advanced Programming Interface" or some variation. If you'd ever had to code using one of these, particularly a "new" one, you'd realize that "Interface" is a very relative term.
So, is it easier to just use sockets as your application's API to the network? In many cases yes... Interestingly enough, TCP already provides multiple streams of cummincation, and port numbers are supposed to tell you how to behave when listening or connecting. Adding another layer on top of this to do essentailly the same thing seems redundant. The only benefit I can see to this might be effects on performance(for wireless) or the ability to "sneak through" firewalls. But at what cost complexity and efficiency?
That said, if anyone ever comes up with a useful presentation layer, I'd be happy to give it a whirl. (XML has been kind of fun that way...)
Re:Who wasn't a programmer? (Score:1, Informative)
Re:Who wasn't a programmer? (Score:2)
I believe that falls under *some variation*.
Either way, interface is still a very relative term...
Re:I think you are missing important BEEP feature (Score:2)
This automatically allowed for sessions
Which is what I thought TCP was all about... though there is overhead of course.
it supported asynchronious request-reply paradigm
I didn't manage to get through the whole specification, but this did seem to be one of the useful things in there... Of course, how useful is this really? It might be somewhat nice to have a tool I can use in my application to create an RPC mechanisms or the like, but what I'd much rather have is an RPC mechanim already built for me. (Like RMI in Java for example)
Don't get me wrong, BEEP will probably be useful in some applications, but I don't expect it to become a "canonical" implementation like XML or TCP.
Basically, I see a need for some level of tools to help in creating generalized TCP application protocols. In fact every TCP programmer I've met that does this for any length of time has created (or modified) their own toolset for this end. The problem with BEEP seems to be that it is a little too heavy and complex for many simpler applications. Also, the application protocol programmer, usually spends a lot more time writing code to communicate with legacy systems then coding new application protocols. I don't see a lot of functionality in BEEP for dealing with legacy issues.
Sounds Cool (Score:1)
That C Library doesn't actually use BEEP (Score:5, Funny)
This way RoadRunner goes really fast with
MEEPMEEP!(thp-thp-thp-thpppp) this was all just a funny lie
Re:That C Library doesn't actually use BEEP (Score:1)
Re:That C Library doesn't actually use BEEP (Score:4, Funny)
Given that we're discussing this within Slashdot and its context, I suggest that we rename it to Minimal Extensible Exchange Protocol Toolkit. Additionally, this protocol should perform every action within a single, long transaction encapsulating multiple exchanges, e.g.:
MEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEPT!
Coding (Score:1)
I love it
I'll be using BEEP for ... (Score:1)
As soon as I start working on it again, of course. If anyone wants to lend a hand, feel free.. ;)
iCalendar (Score:2)
BEEP: not appropriate for iCalendar (Score:2)
Since BEEP is intended as a peer-to-peer connection oriented protocol, it's not ever going to really be practical as an ICAP transport.
The main issue that a server deals with is a shared resource that belongs to the system, rather than participates as a peer. This becomes an issue when you want to schedule use of a confernece room or interview office, where these resources, unlike human beings, don't have a machine which participates in the process as a peer. It's not possible to totally distribute such a system, since the peers are not the only mutually contended resources (if "Bob" drops out, then you can have a request outstanding, but not accepted; no matter who else attends your meeting, though, the conference room *must* "attend").
As far as a peer-to-peer workgroup calendaring service, it *may* be possible to use MIME encapsulated calendar update notifications, similar to the way Microsoft Outlook can be a transport for Microsoft Schedule information (the message type involved lack standardization).
However, such an approach would have the drawback of "netsplits" (in the IRC sense) redulting in shared resources ending up with conflicting scheduling (just like "operator" status in an IRC network following recovery from a "netsplit").
In general, calendaring is a much more complicated problem than people tend to give it credit for being.
Right now, the best protocol bet for Calendaring is LDAPv3 with the "persistant search" or a similar notification mechanism (e.g. LTAP), except not as a proxy, and mandatorily integrated into the LDAP server itself (due to replication requirements).
One place BEEP *might* be useful is data replication between peered servers. The primary reason it could be useful here is that the entity relationship is *peer*, not *client/server*.
-- Terry
Re:BEEP: not appropriate for iCalendar (Score:2)
The newest incarnation of the iCalendar server protocol CAP is built on top of BEEP, and has an SQL-like query language that far exceeds the capabilities of ICAP.
Re:BEEP: not appropriate for iCalendar (Score:2)
The examples at:
http://www.imc.org/ietf-calendar/cap-examples/
Definitely contain the IMAP4 command tags, and uses the "." as a list output terminator. The draft specification which was located at:
http://www.imc.org/ids.html#calsch
8 months ago specifically references differences between IMAP4 and ICAP serversm and how a server can be both. I still have a copy of it in the directory with my Cyrus patches on my other machine.
I intentionally went out of my way to implement an ICAP server in the context the the Cyrus IMAP4 server, thinking that there would be clients, other than the PHP4 client I implemented. I was going to release the Cyrus patches at some point, once I had cleaned them up.
I guess it's a good idea to never implement anything until the standards have settled down, to keep them from changing out from under you. 8-(. Unfortunately, vendors love to change standards out from under you, so that their implementation complies and yours doesn't, and so that you can't beat them to market with a product (beats competing on quality of implementation).
IMO, the correct way to implement this is via an LTAP-type protocol. My personal bias is for adding the transaction and notification features to LDAP, since the Lucent patent is based on the use of a proxy.
I realize that the draft specifically references "BEEP" (not legal in RFC-land, that...;^)), but I still maintain that it requires a client/server relationship to correctly implement calendaring.
This basically means that there are two options: the first is that the BEEP protocol will be used in a client/server mode, rather than the peer-to-peer mode that the BEEP designers intended, OR people are simply going to be unable to implement usable calendaring applications with the newer iCAL, as specified by the new draft.
I still maintain that a peer-to-peer calendaring application is not viable, since it fails to have the minimal features present in, say, OnTrack's "Meeting Maker" (i.e. the ability to schedule a conference room, since peer-to-peer networks are designed -- and intended -- to support paroxial participation in the network).
Or in plain English: there is a lot of hype about peer-to-peer, and not a lot of delivery of the goods.
FWIW: I worked at Artisoft when the Artisoft LANtastic product was *the* preeminent peer-to-peer networking software out there, and was one of just 4 engineers involved in the UNIX (initially, Linux) port, so I'm not talking from ignorance here.
-- Terry
My own application layer protocol (Score:4, Informative)
By and large, yes... it's a symptom of the needs of applications being so varied.
(warning: blatant plug follows) For what it's worth, however, I've developed mine [lcscanada.com] over the course of three years and a dozen or so projects, to the point where I think it's pretty mature and useful; it's open source, and portable to most environments, although the IETF has of course never heard of it... ;^)
XATP, web services + pipelining (Score:1)
Is everybody still inventing his own application layer protocol?
Stateless, connectionless servers are a good idea (HTTP, NFS, SMTP); for this reason, most people are going with web services calls (XML-RPC [xmlrpc.org] or SOAP [w3c.org]) and using HTTP pipelining [mozilla.org] to erase the TCP connection negotiation overhead. This solves 95% of the problems that BEEP is designed for.
If you're still convinced that you really, really, really need a stateful connection, XATP [xatp.org] is much simpler and gets the job done just as well as BEEP.
Re:XATP, web services + pipelining (Score:2)
stateful connection (Score:1)
to clarify (Score:1, Interesting)
The idea is to introduce as little channel state (as few nested channels) as possible. BEEP is essentially "TCP over TCP".
Re:to clarify (Score:2)
BEEP is NOT "TCP over TCP". Using BEEP instead of, say, sockets for implementing network application layer protocols is more like using Perl for your CGI scripts instead of C.
My point: Do you like mucking about with character arrays and pointers and making up your own regex implementations? Do you like having to worry about packet ordering algorithms when writing, for example, a client authentication protocol? Then use a standard TCP API. But if you like being able to specify your packet length with a funcion call then ship out your data without having to mung chop slice and dice all the while worrying about buffer overflows and packet framing, use BEEP.
It's pretty much a higher-level framework to create TCP-based application protocols. (higher-level than using the OS's standard socket API's)
If that sounds like "TCP over TCP" maybe you think Perl is "C on C"???
OK now that I've ranted I might add that I've never written a network application using C, nor have I designed an ALP using BEEP, so I could be full of shit. At least I read the article.
Re:to clarify (Score:2)
Hmm.. Perhaps that was the real question. None of this "how many roads" idiocy! :-)
Re:to clarify (Score:3, Informative)
UDP is great for DNS since queries are small and the overhead of using TCP is large compared to the data exchanged. UDP is also great for things like cache servers using ICP, since then you only need one socket descriptor that can serve however many sibling/child caches you have.
From RFC 768, entitled "User Datagram Protocol": The RFC is two pages long (as opposed to RFC 793 -- TCP -- which has 84 pages) and explains that it is simply a packet of data that does nothing but carry data. It does no handshaking in the protocol, and does not establish a connection, and is thus connectionless.
Re:stateful connection (Score:1)
Re:stateful connection (Score:3, Funny)
No state is maintained from one telephone call to the next.
No state is maintained from one quake session to the next.
By that definition, all protocols are stateless.
Re:stateful connection (Score:2)
Then why do all these long distance calls show up on my bill to places like Texas, Tennessee, Florida, etc?
--
Evan
Re:stateful connection (Score:1)
The difference with HTTP is that a user "session" (browsing your web site) encompasses multiple HTTP requests. Each request stands alone and involves a single query, single response - it's not interactive. (OK I'm simplifying, I know you can do more with HTTP 1.1.) Same with NFS - the NFS 'mount' call just sets up state in the client - the server does not maintain any state between calls. Every NFS call stands alone - the server does not even maintain open files on behalf of the client.
Various mechanisms are used to get around the lack of server state - HTTP cookies, magic session URLs, HTTP auth. In the NFS world, the lock manager uses a separate protocol from the NFS transaction protocol - because locking must maintain state on the server - that being the whole point. (That is why "lockd" is usually a separate process from "nfsd" - RPC mechanisms such as ONC make life much easier if you stick to one daemon, one protocol.)
By contrast, FTP, voice-over-PBX and Quake are not stateless. A single "connection" (the FTP port 21 control channel, the phone call, the Quake data channel) corresponds to the user experience of a single "session". Note that in the case of FTP at least, some clients implement "auto-reconnect" logic in case the server disconnects - this is to maintain the illusion of an unbroken session. Such is not needed with HTTP, since the server isn't maintaining a connection across user requests anyway.
See the difference?
Yes. (Score:2)
I suppose it all depends on your point of view. When I transfer a file via HTTP, it's just as stateless as, say, FTP.
Auto-reconnect logic, as you call it, would work just as well with HTTP as it does with FTP. It's a function of the client, not the protocol.
Re:XATP, web services + pipelining (Score:1)
This is the main reason why I implement my own protocol instead of just use HTTP. Supposedly BEEP can handle this case, but the code doesn't seem to compile, so I haven't tried it out.
Re:XATP, web services + pipelining (Score:1)
Yes, BEEP handles the case you talk about, but only if you have opened multiple channels within the connection. Each channel is processed synchronously, there is no guarante of synchronous or ordered processing between channels. This is not to say that the channel handlers can not impose this within the application, but the protocol itself does not.
Sweetums
Re:XATP, web services + pipelining (Score:3, Insightful)
Isn't this why UDP and other protocols exist on top of IP (or other routing layers). So you can forgo the overhead that a fancy session layer (like TCP) incurs.
It seems to me that most of these new protocols are just trying to get a free ride through existing firewalls... Why can't we all just use IP the way it was meant to be used? (and move to v6 to alleviate addressing pinches)
Re:XATP, web services + pipelining (Score:2)
According to the beepcore FAQ, it's because of NAT and server pooling - you're not guaranteed that multiple requests for connections between two machines actually will be between the same two machines.
As you mentioned, IPv6 would have prevented the problem. It's just a decade too late.
Re: (Score:1)
well (Score:3, Insightful)
Or GRE.
Or anything else we have yet to invent.
Because we don't yet know the best way to use the network.
Maybe we haven't adopted BEEP because you don't just 'create' a standard by declaring your stuff is better. Maybe it's because peopel ALREADY know how to do regular socket programming.
Who knows.
Re:well (Score:2)
Amen Brother!
I should point out for those that might've missed your point that TCP, UDP, and GRE are all "session" layer protocols, which are lower level than application, but still we don't have *just one* of them. So why should there be *just one* application layer?
What's missing from this conversation is the presentation layer. IMHO, this is what XML is all about. Of course, what I think is really missing in industry is a new layer between presentation and application that all of us can ignore as we begin to focus on the presentation layer...
Yes (Score:2)
The point was, we don't NEED a single protocol, we already have one, it's called IP.
Definitely not new, but... (Score:3, Informative)
Application Layer (Score:2)
Totally blank web page (Score:1)
XML is too much sometimes (Score:4, Insightful)
Hanging my head in shame, I'm one of those "still inventing his own application layer protocols". ASN.1 [elibel.tm.fr] and RPC [opengroup.org] were also supposed to save me from doing this. Lately, I've found I've been implementing my own protocols using the concept of netstrings [cr.yp.to] to suit my admittedly low-level needs better. Sadly, as XML and its derivatives mushroom in complexity, I find them less appealing.
Re:XML is too much sometimes (Score:3, Insightful)
Only Bernstein could think that an ASCII representation of Pascal strings is original.
Re:XML is too much sometimes (Score:2)
Anyway, I didn't see anything on that page stating that this would change the world. The "original" part is coming up with a simple way to encode them that's trivial to parse, can be recursively embedded, and has minimal overhead with some chance for error-checking.
It it earth-shattering? No, but it's clever.
Re:XML is too much sometimes (Score:2)
It seems much simpler to assume that a 4-byte binary length will be sufficient. I haven't had to deal with strings longer than 2^32 recently. If you DO have strings that large, your ASCII-encoded length is going to be 10 bytes long.
If you want to include error checking, put a 4-byte CRC/checksum/adler32 after the length.
.NET functionality (Score:1)
Re:.NET functionality (Score:1)
Sweetums
RoadRunner mailing list (Score:2)
List:
http://lists.codefactory.se/pipermail/roadrunner/
(codefactory is the company working on GtkHTML and Mr. Project, btw).
REST, Jabber, SOAP (Score:2, Interesting)
Re:REST, Jabber, SOAP (Score:1)
Beep? (Score:2, Funny)
Oh how it takes me back. The days spend on the old Beeb, learning to program (in basic!) and playing those wonderful old games. I feel quite nostalgic about the old Beeb.
Ah shit! Did I just admit that I actually learned to program on a BBC Micro? Dammit. There goes all my credibility. Nobody on Slashdot will take me seriously anymore! I can just hear the comments: Shutup Grandpa... go play with your 8 bit calculator Granpa... I have a question about Basic Granpa... what does "goto" mean Granpa...
Re:Beep? (Score:2)
It offered a programming environment which allowed for 6502 assember embedded in interpreted BASIC. How deliciously perverse!
GOTOs Represent! (Score:1, Insightful)
I once wrote a 2000+ line RPG without a single loop or set of code blocks in C++. (ie: I used GOTOs instead of brackets)
It ran beautifully, and still does. Not a single error in the whole thing.
I've seen others write a similar game using loops and code blocks, and watch as it hit infinite loops, memory errors, and whole hosts of other problems easily solved with a GOTO.
Say what you will about GOTOs being outdated, but when you compile a C/C++ program, all your if/else/loops all compile into GOTOs.
If they were useless, they why are they still included in all chips? (and prolly will be forever)
slashdotted, and then some (Score:2)
The server encountered an internal error or misconfiguration and was unable to complete your request.
Please contact the server administrator, webmaster@dbc.mtview.ca.us and inform them of the time the error occurred, and anything you might have done that may have caused the error.
More information about this error may be available in the server error log.
Apache/1.3.20 Server at beepcore.dbc.mtview.ca.us Port 80
Way to go, BEEP! I won't fuck with you anymore, because I'm no pedophile, OK?
new registries (Score:1)
the
Re:new registries (Score:1)
http://www.ietf.org/internet-drafts/draft-ietf-
As one of the folks involved in the EPP/BEEP discussions, I can confirm that most of the concerns had to do not only with a lack of experience with current BEEP implementations, but a wealth of experience with streaming data over TCP. Implementers just felt more comfortable working with tried and true technology. It will be nice, though, if we get a chance to implement this draft and see if any of the early performance and complexity concerns were justified.
/. effect -- note from beepcore.org (Score:1)
Re:/. effect -- note from beepcore.org (Score:1)
What about Net::Beep for Perl? (Score:1)
Unfortunately, for me there's still one obstacle: The lack of a perl Net::Beep module.
Yup, I'm that lazy! Until someone else(!) makes it easy in my favorite language, BEEP might as well not exist.
Anybody know of such a project in the works? My search of CPAN, Google, etc., about 6 months ago, turned up nothing.
Nullsoft Beep (Score:1)
to avoid confusion...?
The BEEP community is strong and gettng stronger (Score:3, Informative)
BEEP's an impressive protocol framework with even more impressive implementations.
I'm using it in a burgeoning open source project because of it's ability to multiplex bidirectional communication channels in a transparent fashion. Other features such as dynamic client/server roles, authentication, and channel encryption are just icing on the cake. The less I have to muck around with protocol state details the better!
When I first started looking at BEEP I was impressed by the spec but I was suspicious as to the quality and breadth of implementations. After looking through the high level Java abstractions and more specifically BeepCore-Java I was able to throw together a workable protocol that's proven to be extensible and quite robust in just a few days.
The BEEP community is alive and well!
There are a number of opensource BEEP implementations:There's an excellent IRC channel at OpenProjects [openprojects.org] with the kitschy name #beepnik.
There's the obligatory O'Reilly book, BEEP: The Definitive Guide [oreilly.com].
But the best source for information is the Beepcore.org [beepcore.org] site which has, among other things, an excellent whitepaper [beepcore.org] on the justification and design of the BEEP protocol.
Re:The BEEP community is strong and gettng stronge (Score:3, Interesting)
At the time, it seemed like our choices for letting partners have some say in network routing through our backbone were SOAP or BEEP, and we favored BEEP because of partner pressure from Sun against all things favored by Microsoft (go figure). Eventually, we used simple XML messages rather than an entire application layer.
Early on, BEEP wasn't a difficult protocol to implement; however as time passed, it grew more and more complex, until it maintaining BEEP in a closed-source environment outweighed any benefits. At that point we switched to simple XML messages.
BEEP isn't a bad protocol at all. It is a little over-designed: as a fan of eXtreme Programming, I'd have preferred that smaller versions of the protocol get wider use and more feedback before being expanded.
What about the bandwidth cost? (Score:2, Insightful)
My poor eyesight (Score:2)
It's the debugging that demands sobriety....
Annoyed (Score:5, Interesting)
C: MSG 0 1 . 52 158
C: Content-Type: application/beep+xml
C:
C: <start number='1'>
C: <profile uri='http://iana.org/beep/TLS'>
C: <![CDATA[<ready
C: </profile>
C: </start>
C: END
S: RPY 0 1 . 110 121
S: Content-Type: application/beep+xml
S:
S: <profile uri='http://iana.org/beep/TLS'>
S: <![CDATA[<proceed
S: </profile>
S: END
Re:Annoyed (Score:4, Insightful)
XML maybe very cool and ideal to store data, for config files and all that. How cool would be a completly XML based
However I don't see any real benefits for protocols.
In example what does XML-RPC do any good? Is conventional RPC suddendly uncool? What can XML-RPC do, what normal RPC can't? Except that it uses 5 times more bandwith.
It's just sad, if you type RPC in google, the first five links you get is all about XML-RPC. So marketing words can boost a whole technology? Nobody seems to remember what RPC itself is, and a lot of developers mismatch it today if you talk with them about RPC, you say hey why don't we use RPC? Instead of stuffing sockets ourself, and they start to talk about XML, RPC does not need XML at all.
Re:Annoyed (Score:2)
Drive the market for massively high bandwidth, Gigabit switches, Pentium 4s and horribly expensive application level firewalls...?
Re:Annoyed (Score:2)
The main two things I've seen about XML are:
1. It makes medium-speed platforms like tomcat very, very slow performers.
2. It is hard to explain to users, who need to understand arcane concepts in order to exploit it.
3. The simple uses of it are indistinguishable from what can be done with config files.
4. When it really starts getting used you start seeing network competition effects like browser wars, differences in embedded languages, and rendering artificats.
5. It is relatively easy to bullshit vaporware, bloatware, and rapacious software houses if it has an XML in it (or an "e-commerce", a "secure key", or a "content management").
6. That said, it can be a great tool. But I don't want to eat, sleep, and breathe the stuff!
Re:Annoyed (Score:3, Insightful)
The "classic" example for an XML based protocol is a server which provides weather information to anonymous clients over the public net. Do you want to write your own protocol and try to explain it to everyone who wants your information? No, just package it in XML, provide a schema, and it's easy for clients to fetch.
Situations where you DO have control over both ends don't benefit from XML very much, especially when performance is important (and when isn't it?).
There is a place for technologies which consume extra CPU, memory, and bandwidth but provide something else: for example, reduced programmer development time (for some applications). Java is a good example. The benefits of XML are more limited, but they do exist.
Re:Annoyed (Score:2)
How about that you CAN telnet to the relevent port and type the right data. Text based protocols like SMTP, POP, HTTP, etc are very easy to write clients for, and to debug, because you can simply write & read text. Binary based protocols like DNS require special libraries or more complex programming.
So what's next (Score:2)
If anyone knows what I'm talking about, I'll be pleasantly surprised!!!
Re:It's Out Of Date (Score:2)
There's nothing stopping an app from having multiple BEEP connections open at once.
no authentication that the data being transferred is accurate aka Freenet
It's impossible to do that in a generic way, so BEEP doesn't do it.
Re:It's Out Of Date (Score:1)
TCP handles that just fine, it's called reliability. It's shitty application protocols that screw up your data.
Re:Do we need complex acronyms? by poopbot (Score:1)
Re:Acronyms... (Score:2)
"Do I have a scuzzy hard drive?! What kind of person do you think I am?"
Re:Acronyms... (Score:1)
ObScore+1Interesing: SCSI was intended to be pronounced "sexy". I don't know why "scuzzy" caught on instead. Probably for the same reason "BEEP" won't take off. It tends to defy any serious discussion.
Re:Acronyms... (Score:2)
If the coiners of the acronym really wanted it pronounced "sexy", they should have come up with some "E" word between "Small" and "Computer".
BEEP is pretty clearly pronounced how it sounds. Just so long as nobody confuses it with ping (or PNG, for that matter).
Re:Acronyms... (Score:2)
Of course, it used to be calles "SASI" for "Shugart Associates Computer Interface" (IIRC). And Atari had a variant called "ACSI", for ... Oh, you work it out. It had a funny D connector that you couldn't get when you wanted to make up cables, 19 pins or something.
Re:Acronyms... (Score:1)
Re:Acronyms... (Score:1)
Re:Slashdot story from 2000: (Score:2)
I believe TCP already solves this problem, perhaps herein lies the lack of excitement...
Re:Slashdot story from 2000: (Score:2, Informative)
The BEEP activity I noticed most recently was on a proposal for syslog-reliable, and I believe there are some intrusion detection things using it in the academic space.
There is also an excellent white paper (on beepcore.org I think) on the transport of SOAP using BEEP rather than HTTP, and it really is a pretty sweet fit, though early on they were in competition for mindshare at the IETF and neither wanted to be absorbed conceptually into the other.
O'Reilly has published a book on BEEP by Dr. Marshall Rose, the principal BEEP designer. It's good.
Sweetums
Re:Slashdot story from 2000: (Score:2)
I was working in the IDWG (intrustion detection working group) and was on team that started implementation of the BEEP version of the intrusion detection exchange protocol. It was certainly less work getting the BEEP version functional than the hacked-HTTP protocal that IAP (the first version of the protocol) was using. It also gave us some benefits vis a vis pluggable autentication/encryption. Most of the people involved got kinda busy, so I'm not sure how far the project has gone. I think we still have a webpage up under java-idxp on Sourceforge.