[
https://issues.apache.org/jira/browse/PROTON-591?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrew Stitcher updated PROTON-591:
-----------------------------------
Description:
The frame dispatch for the raw amqp and sasl layers use a table with 256
entries to lookup the handling routine for the incoming performative.
There are only 14 different performatives and there is a separate table for the
raw amqp and the sasl layer so there are in total 498 empty table entries and
only 14 actual entries.
Currently we don't need to add performatives on the fly so there is no real
need for a table that can be altered as in the current design. We can save a
lot of space and somewhat simplify the code by using a dumb
{code} switch(code) { ...case OPEN: ... break; }; {code} construct for the
dispatch.
On a 64 bit machine this saves about 4k per connection and on a 32 bit around
2k per connection (or only 1k if the SASL layer is disabled)
was:
The frame dispatch for the raw amqp and sasl layers use a table with 256
entries to lookup the handling routine for the incoming performative.
There are only 14 different performatives and there is a separate table for the
raw amqp and the sasl layer so there are in total 498 empty table entries and
only 14 actual entries.
Currently we don't need to add performatives on the fly so there is no real
need for a table that can be altered as in the current design. We can save a
lot of space and somewhat simplify the code by using a dumb switch...case
construct for the dispatch.
On a 64 bit machine this save about 4k per connection and on a 32 bit around 2k
per connection (or only 1k if the SASL layer is disabled)
> There are two large, mostly empty tables used for dispatching incoming frames
> -----------------------------------------------------------------------------
>
> Key: PROTON-591
> URL: https://issues.apache.org/jira/browse/PROTON-591
> Project: Qpid Proton
> Issue Type: Improvement
> Components: proton-c
> Affects Versions: 0.7
> Reporter: Andrew Stitcher
> Assignee: Andrew Stitcher
>
> The frame dispatch for the raw amqp and sasl layers use a table with 256
> entries to lookup the handling routine for the incoming performative.
> There are only 14 different performatives and there is a separate table for
> the raw amqp and the sasl layer so there are in total 498 empty table entries
> and only 14 actual entries.
> Currently we don't need to add performatives on the fly so there is no real
> need for a table that can be altered as in the current design. We can save a
> lot of space and somewhat simplify the code by using a dumb
> {code} switch(code) { ...case OPEN: ... break; }; {code} construct for the
> dispatch.
> On a 64 bit machine this saves about 4k per connection and on a 32 bit around
> 2k per connection (or only 1k if the SASL layer is disabled)
--
This message was sent by Atlassian JIRA
(v6.2#6252)
