ace-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Bram de Kruijff <bdekrui...@gmail.com>
Subject Re: Analysis for updating the MA
Date Mon, 05 Aug 2013 07:23:40 GMT
Hi Marcel,

On Thu, Aug 1, 2013 at 3:48 PM, Marcel Offermans
<marcel.offermans@luminis.nl> wrote:
> Hello devs,
>
> On the wiki I created a document that provides an analysis of how we could design an
update mechanism for the management agent, as described in ACE-342 [2]. I would love to hear
your feedback on this!
>

Just my 2 cents; Although I am sure solution A will work, my
preference would be to go for solution B for reasons of simplicity and
efficiency.

Simplicity because, why have two transport mechanisms if you can have
one? Obviously we can and, as you suggest, must make an effort to keep
the implementations aligned. Still I feel 1 is better then 2 :)

Efficiency because, as you state, a second mechanism requires a second
HTTP call. Leveraging keep-alive (that is what you mean with HTTP
pipelining?) reduces TCP/IP overhead so that is a good idea, but that
has several consequence. First, this requires the agent update to be
on the same schedule as the regular check for updates (which I think
that is good anyway as it limits waking up devices/connections).
Second, we can not guarantee that keep-alive works in all environment
(firewalls/proxies). Third, when using keep-alive we must ensure that
connections are being closed from the target side in a timely fashion
to prevent running out of connection slots on the server fast.

grz,
Bram

> Greetings, Marcel
>
> [1] https://cwiki.apache.org/confluence/display/ACE/Analysis+and+Design+for+updating+the+Management+Agent
> [2] https://issues.apache.org/jira/browse/ACE-342

Mime
View raw message