[ovs-dev] [PATCH v2 1/2] dpif-netdev: Expand the meter capacity using cmap

Tonghao Zhang xiangxia.m.yue at gmail.com
Fri Mar 27 09:52:02 UTC 2020


On Tue, Mar 24, 2020 at 8:40 PM Tonghao Zhang <xiangxia.m.yue at gmail.com> wrote:
>
> On Tue, Mar 24, 2020 at 1:11 PM Ben Pfaff <blp at ovn.org> wrote:
> >
> > On Tue, Mar 24, 2020 at 09:25:41AM +0800, Tonghao Zhang wrote:
> > > On Mon, Mar 16, 2020 at 12:06 PM Tonghao Zhang <xiangxia.m.yue at gmail.com> wrote:
> > > >
> > > > On Mon, Mar 16, 2020 at 11:07 AM Yanqin Wei <Yanqin.Wei at arm.com> wrote:
> > > > >
> > > > > Hi Xiangxia,
> > > > >
> > > > > The meter id is allocated by id_pool,  which can always return the lowest available id.
> > > > Yes, I added 70000+ meters, and it works fine.
> > > > > There is a light scalable data struct which supports direct address lookup. It can achieve several times performance improvement than cmap_find. And it has not copy memory when expanding.
> > > > > https://patchwork.ozlabs.org/patch/1253447/
> > > > >
> > > > > I suggest using this light data struct for >65535 meter instance. Would you like to take a look at it?
> > > > Yes, and I also wait Ilya to review, and offer a suggestion.
> > > Hi maintainers,
> > > Will you have a plan to help me to review those patches, thanks.
> >
> > I know that I skipped over this because I was waiting for v3 using
> > idpool.
> Hi Ben,
> The meter id is allocated by id_pool in ofproto-dpif layer(meter_set
> function use the id_pool_alloc_id to allocate it).
> I guess what Yanqin  mean is that I should use the patches[1] he sent,
> to compare the performance with cmap.
> The patches of Yanqin are not in upstream, but I use the meter cache
> to address the issue.
Hi Ben, Yanqin
Did I misunderstand ?
> [1] - https://patchwork.ozlabs.org/patch/1253447/
> --
> Best regards, Tonghao



-- 
Best regards, Tonghao


More information about the dev mailing list