On 26 Aug 2004 08:08:51 -0400
jamal <hadi@xxxxxxxxxx> wrote:
> On Wed, 2004-08-25 at 13:53, Stephen Hemminger wrote:
> > Extend netem options to support new features.
> >
>
> > struct tc_netem_qopt
> > {
> > __u32 latency; /* added delay (us) */
> > @@ -409,6 +410,13 @@
> > __u32 loss; /* random packet loss (0=none ~0=100%) */
> > __u32 gap; /* re-ordering gap (0 for delay all) */
> > __u32 duplicate; /* random packet dup (0=none ~0=100%) */
> > - __u32 jitter; /* random jitter in latency (us) */
> > + __u32 jitter; /* delay sigma (us) */
> > +
> > + __u32 delay_corr; /* delay correllation (0=none ~0=100%) */
> > + __u32 loss_corr; /* packet loss correllation (0=none ~0=100%) */
> > + __u32 dup_corr; /* duplicate correlation (0=none ~0=100%) */
> > +
> > + __s16 delay_dist[0]; /* delay distribution table (optional) */
> > +#define TCA_NETEM_TABLEFACTOR 8192
> > };
>
> You should really send the delay_dist in its own TLV to avoid
> the kind of mistakes that showed up in u32 classifier.
> Also allows you to extend that structure in the future.
>
> cheers,
> jamal
It would be good but the whole qdisc api (init, change) is really structured
around a single TLV.
|