Forum: GNU Radio Misc questions, regarding Interrupt coalescing

371d411c739f4bcedc3ba4ee0816ab5b?d=identicon&s=25 Naceur El Ouni (naeo)
on 2013-10-21 22:20
(Received via mailing list)
Hello GR Forum,

I got some questions:

1/ Did anyone already tested the effect of Interrupt coalescing on
reducing
the latency when host and USRP N2X0 are communicating and how far did he
reduce this latency

2/ I want to check first if my NIC got this feature enabled ?
How do I have to proceed to check then how to test ?

3/ When running a stream of packets over the USRP I got the following
error:

thread[thread-per-block[24]: <gr_block message_source (8)>]: msg length
is
not a multiple of d_itemsize
This error is raised after a fixed number N of successfully sent packets
Could you please give me some hints on this issue.

Best regards,



--
View this message in context:
http://gnuradio.4.n7.nabble.com/Misc-questions-reg...
Sent from the GnuRadio mailing list archive at Nabble.com.
7d89a70df32c0ae27c1235016f9e5441?d=identicon&s=25 "Marcus Müller" <marcus@hostalia.de> (Guest)
on 2013-10-22 11:13
(Received via mailing list)
Hi Naceur,

while questions 1 and 2 definitely are more of material for usrp-users,
I'll try to answer them as briefly as possible, and puit usrp-users into
the CC:.
On 10/21/2013 10:19 PM, Naceur wrote:
> 1/ Did anyone already tested the effect of Interrupt coalescing on reducing
> the latency when host and USRP N2X0 are communicating and how far did he
> reduce this latency
I haven't specifically tested that.
However, Interrupt Coalescing should not decrease latency; on the
contrary, it should increase raw network latency but reduce CPU load.
This could even lead to decreased overall latency, however, as long as
NIC interrupts do not happen faster than they can be processed (leading
to packet loss in the long run), it probably won't.
> 2/ I want to check first if my NIC got this feature enabled ?
> How do I have to proceed to check then how to test ?
Good question. It's not only that your NIC needs to support input
moderation, your OS needs to do that, too!
ethtool is your friend under recent GNU/linuxes, try ethtool
--show-coalesce <ethernet device name>. However, on my devices, that
always tells me that it cannot get device coalesce settings.
> 3/ When running a stream of packets over the USRP I got the following error:
>
> thread[thread-per-block[24]: <gr_block message_source (8)>]: msg length is
> not a multiple of d_itemsize
> This error is raised after a fixed number N of successfully sent packets
> Could you please give me some hints on this issue.
Hm could you give some more hints on this issue? The error has something
to do with a message source, and you talk about packets, but you don't
say anything on the actual flowgraph you're using.

Hope I was able to help you a bit,
Marcus
5e623cc1b53ddefb15c9bad4245986a1?d=identicon&s=25 Johannes Demel (Guest)
on 2013-10-23 23:39
(Received via mailing list)
Unfortunately last nights mail didn't get to the mailing list. But here
it is.


Hi,

first of all you might be interested in [1]. Depending on the actual
setup of your NIC latency depends a lot on interrupt coalescing. There
is also some further reading on this page.

Have fun

[1] http://code.ettus.com/redmine/ettus/projects/publi...
Please log in before posting. Registration is free and takes only a minute.
Existing account

NEW: Do you have a Google/GoogleMail, Yahoo or Facebook account? No registration required!
Log in with Google account | Log in with Yahoo account | Log in with Facebook account
No account? Register here.