netem/iproute2 solving correlated loss issues [0/5]
Subsystems (2)
PATCH 1/1 netem/iproute2 solving correlated loss issues [0/5]
Dear all,
we are going to send a patch that enhances netem module and the iproute2 utilities.
Our aim is to fix the problems with correlated loss generation in netem (as for example described in
https://lists.linux-foundation.org/pipermail/netem/2007-September/001156.html)
We have now included a more general loss model, as described in:
http://netgroup.uniroma2.it/twiki/bin/view.cgi/Main/NetEm2
The patch is backward compatible, we have not removed the previous correlated loss generated approach (though its use should be deprecated), nor we have changed anything else in the existing netem module.
Within our patch for iproute2 we also added a man page for netem (it was missing) in which we have pasted the existing documentation for netem and the usage of our model.
We had posted this work on the netem mailing list some months ago as a separate netem2 module... we received some feedback and made some bug fix since then. We believe the work is now ready to be inserted in the main netem/iproute2 development.
We will post in next messages the patches to netem module for 2.6.32 kernel and to 2.6.29 iproute2 (find below the list)
Comments are welcome.
Best regards,
Stefano and Fabio
netem kernel module patches:
patch 1/5 : linux-2.6.32/include/linux/pkt_sched.h patch 2/5 : net/sched/sch_netem.c
iproute2 utilities patches:
patch 3/5 : iproute2-2.6.29-1/include/linux/pkt_sched.h patch 4/5 : iproute2-2.6.29-1/man/man8/tc-netem.8 patch 5/5 : iproute2-2.6.29-1/tc/q_netem.c
Signature · 12 lines
******************************************************************* Stefano Salsano Dipartimento Ingegneria Elettronica Universita' di Roma "Tor Vergata" Via del Politecnico, 1 - 00133 Roma - ITALY http://netgroup.uniroma2.it/Stefano_Salsano/ E-mail : [email protected] Cell. : +39 320 4307310 Office : (Tel.) +39 06 72597770 (Fax.) +39 06 72597435 *******************************************************************
patch 1/5 : linux-2.6.32/include/linux/pkt_sched.h
include/linux/pkt_sched.h +15 -0
@@ -479,6 +479,21 @@__u32 gap;/* re-ordering gap (0 for none) */ __u32 duplicate;/* random packet dup (0=none ~0=100%) */ __u32 jitter;/* random jitter in latency (us) */ +__u32 p13;/* GI loss model-state 1 to 3 transition probability */ +__u32 p31;/* GI loss model-state 3 to 1 transition probability */ +__u32 p32;/* GI loss model-state 3 to 2 transition probability */ +__u32 p23;/* GI loss model-state 2 to 3 transition probability */ +__u32 p14;/* GI loss model-state 1 to 4 transition probability */ +__u32 gilb_p;/* Gilbert-Elliot model-state good-bad transition probability */ +__u32 gilb_r;/* Gilbert-Elliot model-state bad-good transition probability */ +__u32 gilb_h;/* Gilbert-Elliot-loss model loss probability in bad state */ +__u32 gilb_k;/* Gilbert-Elliot-loss model loss probability in good state */ +__u32 algorithm;/* loss model algorithm selection */ +__u32 logging;/* loss logging mode selector */ +__u32 query;/* loss query mode selector */ +__u16 pattern_length;/* deterministic loss pattern length */ +__u32 pattern_repetitions;/* deterministc loss pattern repetitions */ +__u16 *user_pattern;/* determinist loss pattern elements */ };struct tc_netem_corr
David Miller
maintainerFrom: Stefano Salsano <[email protected]>
Date: Fri, 18 Dec 2009 11:39:13 +0100
Quoted 4 lines
The patch is backward compatible, we have not removed the previous correlated loss generated approach (though its use should be deprecated), nor we have changed anything else in the existing netem module.
To heck it is!
The changed version of iproute2 will not work without the kernel changes installed. The kernel validates the size of the structures passed in, you can't change the user visible interface like this.
Add new netlink attributes to extend things, that's the "real" backwards compatible way to make changes to netlink interfaces.
Quoted 6 lines · David Miller
Quoted 1 line
The patch is backward compatible, we have not removed the previous...
To heck it is!
The changed version of iproute2 will not work without the kernel changes installed. The kernel validates the size of the structures passed in, you can't change the user visible interface like this.
Quoted 5 lines · David Miller
Furthermore it is fundamentally flawed in it's implementation in that the user exported data structures cannot be changed, you cannot change the layout and you absolutely cannot change the size of these things or else various tool and kernel combinations stop working.
thank you for raising these fundamental issues ! we will address them and resubmit the patch when done...
Quoted 2 lines
Add new netlink attributes to extend things, that's the "real" backwards compatible way to make changes to netlink interfaces.
point taken
Signature · 12 lines
******************************************************************* Stefano Salsano Dipartimento Ingegneria Elettronica Universita' di Roma "Tor Vergata" Via del Politecnico, 1 - 00133 Roma - ITALY http://netgroup.uniroma2.it/Stefano_Salsano/ E-mail : [email protected] Cell. : +39 320 4307310 Office : (Tel.) +39 06 72597770 (Fax.) +39 06 72597435 *******************************************************************