OTP vs. non-OTP (was: A Joeish Erlang distribution (long))

[Ulf Wiger]

On Fri, 31 Jan 2003, Martin Bjorklund wrote:

>It's very interesting to hear opinions from different
>people; people who have built shipping erlang products and
>people that have not. Interstingly, people who haven't
>shipped products seem to think that OTP is an unnecessary
>burden, and people who actually have built products tend to
>like the support OTP gives you.

I second that opinion. Most people here at AXD301 are not
that familiar with OTP (except for the parts that are
relevant to them). Most programmers use gen_server, mnesia
and the standard libraries; a few use things like gen_tcp;
many come in contact with SNMP somehow; and INETS, but only
indirectly, using our internal web page building support.

A handful of people really get OTP dirt under their
fingernails on a daily basis, working with the application
controller, release handler, etc. Supervision structures are
built using templates, and a few people get involved in
doing it once, and then forget about it. Very few actually
understand how it all works, but don't have to, either. (*)



>In your case, and klacke's (except for yaws, see below),
>you have other people around you that take care of the
>productification; building software packages, taking care
>of upgrades, start of the system, configuring the system
>etc.  You think this is better done w/o OTP? Fine, go ahead
>and prove me wrong.

I was expecting this retort. What kept you? (:

For those relatively new to Erlang, Martin and Magnus
Fröberg (also at Nortel) used to work on the OTP team and
support the application controller, release handler, global,
etc.(**)  Even then, most of the others (e.g. Klacke and
Jocke ;) in the group didn't really know all that stuff, and
didn't want to know. I have no objection to this, but
remember that both Klacke and Jocke have had the good
fortune to have some of the top experts in this area sitting
next to them. Then it's easy to just forget about all that
complexity and what it does for you. This is as it should
be.

I have stayed out of the discussion mainly because I'm one
of the OTP experts at AXD301, a Large Telecom Project (even
though it's not particularly large by Ericsson standards...)
Noone has questioned our need for OTP, and I felt that I
shouldn't necessarily speak for the rest of you.

I think what's needed is primarily better tutorials, and
better support for a gentle introduction to OTP. I fully
agree with Francesco and Martin.

/Uffe

(*) This has a downside, though. One of our finest
programmers once tried to demonstrate to some sceptics how
easy things were if you used Erlang, but he found that he
didn't know how to put together a distributed OTP-based
system from scratch. Some quick-start examples are badly
needed.

(**) Others (e.g. Lennart Öhman) have been involved as well,
but of the people I know to be at Nortel now, I think M&M
were most active in this area. Jocke designed INETS and
fathered Eddie, and Klacke, well... lots -- Distributed
Erlang and Mnesia (and, IIR global and rpc), for example. I
would not think of trying to belittle their achievements in
any way. Just suggesting that they could both afford to
forget about many of the problems that OTP solves.

-- 
Ulf Wiger, Senior Specialist,
   / / /   Architecture & Design of Carrier-Class Software
  / / /    Strategic Product & System Management
 / / /     Ericsson Telecom AB, ATM Multiservice Networks