[zeromq-dev] Fundamental perks?

[Pieter Hintjens]

Why even use an operating system? It's far safer to write the code straight
into hardware.

Seriously, anyone trying do build distributed systems who is not using some
kind of abstraction layer because of the extra dependency is... well,
they've not understood the different costs / benefits.

Without a messaging library of some kind you have to deal with dozens or
hundreds of issues in every single connection. The BSD socket API is very
low level. You have to handle all kinds of errors, some by retrying, some
by giving up, some by trying something different. To get any performance at
all you must build an event-driven framework so you can read and write as
sockets are ready, rather than blocking. To deal with more than a single
socket this becomes essential. Then you have to frame your data and deal
with edge cases and invalid data.

At the least you want to hide this from developers, because they will get
it wrong over and over, and this causes far worse complexity and
dependencies than using a mature and proven library.

it's easy to count. Hand-written TCP code measures thousands of lines for
any given case. The same code using a messaging API counts a few dozen.
Now, whether you use ZeroMQ or something else is secondary. ZeroMQ removes
the need for brokers, though they can help in some cases.

When you introduce technology to a business you will often get resistance
for reasons that have nothing to do with the technology itself.

-Pieter



On Thu, May 14, 2015 at 6:33 AM, Kenneth Adam Miller <
kennethadammiller at gmail.com> wrote:

> So, recently in the workforce the discussion came up that despite all of
> the nice work that has been accomplished with ZMQ, it should be removed
> because using it as a component for distributed analysis creates
> complication. Much to my dismay, they cannot be convinced otherwise, and
> see dependencies only as potential breakage and configuration nightmare.
>
> But our use case still needs to communicate between processes in the new
> model proposed. The idea is for regular raw sockets to be used to
> communicate between processes, because this would be the only
> communication, and the linux kernel would automatically replace any
> localhost communication with inter process communication (?).
>
> What fundamental benefits in such a use case does using ZMQ provide?
> Things like OS interrupts, OS abstraction, ect... I'm pretty sure that ZMQ
> does at least hide some complexity in buffering what it receives.
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20150514/b6c8a1fe/attachment.htm>