[nanomsg] Re: Persisted reliable messaging
- From: Jörg Singler <joerg@xxxxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Fri, 22 Nov 2013 17:33:43 +0100
Hi Paul,
my intention was exactly to work on this topic and making some tests for
various options and whether one benefits of doing compression or not. My
first iteration was more focused on the basics meaning keeping track of the
offset of the processed and the new messages. I used the following stacking
of patterns: Pub -> SUB (filtering on topics) -> Files -> req -> reply
(sending aknowledge only) -> PUSH --> PULL (distributing work to multiple
workers). Do you think the stacking should/can be optimized? I will keep
you updated...
Best, Joerg
2013/11/22 Paul Colomiets <paul@xxxxxxxxxxxxxx>
> Hi,
>
> On Fri, Nov 22, 2013 at 5:44 PM, Jörg Singler <joerg@xxxxxxxxxxxx> wrote:
> > Hi Paul,
> >
> > thanks. I have already a version which is running on top by using simple
> > files. That works fine so far by stacking the current patterns. It gives
> you
> > additionally possibilities to send messages in a bulk. Clearly my
> > implementation misses the proper handling of embedded pointers in the
> > message and some more sophisticated mechanisms. Btw. how do you handle
> > embedded pointers in a message in the core when using remote protocols?
> Does
> > that work via RDMA?
> >
>
> No embedded pointers are expected in the messages (except minority of
> cases with inproc://). The data is serialized before sending via the
> network.
>
> If you have a prototype, the questions that I'm curiuos about:
>
> 1. What is performance impact?
>
> 2. How do you compact the log, or whatever on-disk structure you use
> for persistence?
>
> --
> Paul
>
>
Other related posts:
