Hi: I am reading document about distributor library which is added in DPDK 1.7. The document mentioned the packets are dynamically load balanced between a set of worker cores. So I am wondering is this the only reason we need distributor library? What else could it give us? Do we have any performance number on this new library? Thanks.
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wang, Shawn
> Sent: Thursday, October 9, 2014 7:00 PM
> To: dev@dpdk.org
> Subject: [dpdk-dev] Newbie question about distributor library
>
> Hi:
>
> I am reading document about distributor library which is added in DPDK 1.7.
> The document mentioned the packets are dynamically load balanced between a
> set of worker cores.
> So I am wondering is this the only reason we need distributor library? What else
> could it give us?
> Do we have any performance number on this new library?
>
> Thanks.
Hi,
Distributor library only takes care of load distribution based on flow type i.e. based on rss ( which is calculated based on 5 tuple of the packet)
value of mbuf to workers. Packets with same rss will be given to same worker. Hence different flows goes to different workers.
15.5 mpps was the performance we achieved with 8 worker cores, 1 Rx core and 1 TX core using sample application.
Thanks,
Reshma
15.5 mpps is amazing.
Does the RX core run the distributor or there is another distributor core?
Thanks.
________________________________________
From: Pattan, Reshma [reshma.pattan@intel.com]
Sent: Friday, October 10, 2014 12:47 AM
To: Wang, Shawn; dev@dpdk.org
Subject: RE: Newbie question about distributor library
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wang, Shawn
> Sent: Thursday, October 9, 2014 7:00 PM
> To: dev@dpdk.org
> Subject: [dpdk-dev] Newbie question about distributor library
>
> Hi:
>
> I am reading document about distributor library which is added in DPDK 1.7.
> The document mentioned the packets are dynamically load balanced between a
> set of worker cores.
> So I am wondering is this the only reason we need distributor library? What else
> could it give us?
> Do we have any performance number on this new library?
>
> Thanks.
Hi,
Distributor library only takes care of load distribution based on flow type i.e. based on rss ( which is calculated based on 5 tuple of the packet)
value of mbuf to workers. Packets with same rss will be given to same worker. Hence different flows goes to different workers.
15.5 mpps was the performance we achieved with 8 worker cores, 1 Rx core and 1 TX core using sample application.
Thanks,
Reshma
Hi ,
Distributor was running as part of Rx core.
Thanks,
Reshma
> -----Original Message-----
> From: Wang, Shawn [mailto:xingbow@amazon.com]
> Sent: Friday, October 10, 2014 10:06 PM
> To: Pattan, Reshma; dev@dpdk.org
> Subject: RE: Newbie question about distributor library
>
> 15.5 mpps is amazing.
> Does the RX core run the distributor or there is another distributor core?
>
> Thanks.
> ________________________________________
> From: Pattan, Reshma [reshma.pattan@intel.com]
> Sent: Friday, October 10, 2014 12:47 AM
> To: Wang, Shawn; dev@dpdk.org
> Subject: RE: Newbie question about distributor library
>
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wang, Shawn
> > Sent: Thursday, October 9, 2014 7:00 PM
> > To: dev@dpdk.org
> > Subject: [dpdk-dev] Newbie question about distributor library
> >
> > Hi:
> >
> > I am reading document about distributor library which is added in DPDK 1.7.
> > The document mentioned the packets are dynamically load balanced
> > between a set of worker cores.
> > So I am wondering is this the only reason we need distributor library?
> > What else could it give us?
> > Do we have any performance number on this new library?
> >
> > Thanks.
>
> Hi,
>
> Distributor library only takes care of load distribution based on flow type i.e.
> based on rss ( which is calculated based on 5 tuple of the packet) value of mbuf
> to workers. Packets with same rss will be given to same worker. Hence different
> flows goes to different workers.
>
> 15.5 mpps was the performance we achieved with 8 worker cores, 1 Rx core
> and 1 TX core using sample application.
>
> Thanks,
> Reshma
>