Re: [MirageOS-devel] Parallelizing writing to network devices

[Anil Madhavapeddy <anil@xxxxxxxxxx>]

> On 27 Nov 2014, at 16:56, Masoud Koleini <masoud.koleini@xxxxxxxxxxxxxxxx> 
> wrote:
> 
> Hi all,
> 
> Wondering about the semantics of Lwt_ring and Netif.write. I am writing a 
> code that sends an Lwt_stream of frames over a network device using 
> Netif.write with highest possible rate and preferably respecting frame 
> ordering. Another thread pushes frames into this stream.
> 
> To parallelize writing to the output device, a recursive iteration over the 
> stream extracts every element using Lwt_stream.next, and calls Netif.write 
> under Lwt.ignore_result. Threads stop functioning with no error printed on 
> the console at high packet rates. Any suggestion?

That should work, to my understanding. The next step is to figure out what's 
blocking:

- if an event is being lost, then try to prod the ring to wake up with external 
stimulus (e.g. pinging the interface).
- graph the ring utilisation to see if it's always full (Thomas Leonard's 
profiling patches should help here)
- try to reduce the parallelisation to see if some condition there alleviates 
the issue to track it down.

-anil


_______________________________________________
MirageOS-devel mailing list
MirageOS-devel@xxxxxxxxxxxxxxxxxxxx
http://lists.xenproject.org/cgi-bin/mailman/listinfo/mirageos-devel