[asterisk-bugs] [JIRA] (DAHLIN-290) dahdi_dynamic_eth have a huge jitter on transmit

[Shaun Ruffell (JIRA)]

    [ https://issues.asterisk.org/jira/browse/DAHLIN-290?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=201910#comment-201910 ] 

Shaun Ruffell commented on DAHLIN-290:
--------------------------------------

{quote}
some hints:
1. why don't you use the same (one) tasklet for either receive or transmit job ?
Too much #if #else #endif ...
{quote}

Ok, perhaps.  I thought it would make things clearer to indicate that only the flush of the transmit was happening in the flush_tlet.  But, perhaps not.

{quote}
2. ENABLE_TASKLET is not correct anymore.
dynamic will always have a tasklet (either for receive or transmit).
It's better to choose via module parameter.
modprobe dahdi_dynamic tasklet=0/1, receive/transmit.
{quote}

I'm not crazy about this, but then I'm not a huge dynamic span user.  What I think would be preferable is to make things work reliably without exposing a tuning knob like that. I believe that by just using the flush tasklet, things should be better now, but I could be mistaken.

{quote}
3. it's probably a good idea to start flush tasklet if (irqs_disabled()).
Otherwise, flush can be called directly.
{quote}

Isn't it only called in the context of dahdi_receive?  If not, then I agree with you.
                
> dahdi_dynamic_eth have a huge jitter on transmit
> ------------------------------------------------
>
>                 Key: DAHLIN-290
>                 URL: https://issues.asterisk.org/jira/browse/DAHLIN-290
>             Project: DAHDI-Linux
>          Issue Type: Bug
>      Security Level: None
>          Components: dahdi_dynamic_eth
>    Affects Versions: 2.6.1
>         Environment: tasklets enabled in dahdi_dynamic.c (#define ENABLE_TASKLETS)
>            Reporter: Pavel Selivanov
>            Assignee: Pavel Selivanov
>         Attachments: 0001-dahdi_dynamic-Use-a-tasklet-for-flushing-dynamic-dri.patch, 0002-Revert-dahdi_dynamic_eth-Move-tx-packet-flushing-to-.patch, dahdi_dynamic.c.patch, dahdi_dynamic_eth.c.patch
>
>
> At r10562 "Move tx packet flushing to process context".
> Seems, using a common "thread" can produce a huge jitter.
> Since this "common" thread can be used by any driver - it can't guarantee 1ms...

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira