Date: Thu, 11 Mar 2010 22:46:44 -0800 From: Jack Vogel <jfvogel@gmail.com> To: beezarliu <beezarliu@yahoo.com.cn> Cc: rihad <rihad@mail.ru>, "freebsd-net@freebsd.org" <freebsd-net@freebsd.org> Subject: Re: Re: ixgbe input errors at high data rates Message-ID: <2a41acea1003112246qd1725f9ve55350c064a4583f@mail.gmail.com> In-Reply-To: <2a41acea1003112244n6a62083ag39696be4d37e134c@mail.gmail.com> References: <4B99114E.7060909@mail.ru> <2a41acea1003110942u717e2222hd984bd2859c3e477@mail.gmail.com> <4B9938F1.7090906@mail.ru> <2a41acea1003111056k42d9a188x3310bcedd2ae9e6a@mail.gmail.com> <201003121420185625042@yahoo.com.cn> <2a41acea1003112244n6a62083ag39696be4d37e134c@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
I should also point out that the FLOWID control only exists in 8.X. Oh, and if there is no flow id, then TX uses the queue that is bound on the CPU it is running on of course :) Jack 2010/3/11 Jack Vogel <jfvogel@gmail.com> > There is no user control, on the TX side there is a check of the FLOWID = of > the packet > which it gets on RX, if there is one it uses that to determine the right = TX > queue to use. > The TX side also has a flow control but this only exists on the 82599, no= t > 598. > > The RX/TX queue pairs are bound to a cpu via their MSIX vector, so I > believe you will > see the results you wish. > > As I implement more virtualization features more control may become > possible, we'll > see. > > Regards, > > Jack > > > 2010/3/11 beezarliu <beezarliu@yahoo.com.cn> > > >> Speaking of multiqueue support in intel 10G, such as 82598, >> if we use RSS feature, can we select certain ip/port pair to open a tcp >> connection, >> so that the returning packet will come back from a pre-known >> receive queue? >> >> >> 2010-03-12 >> ------------------------------ >> beezarliu >> ------------------------------ >> *=B7=A2=BC=FE=C8=CB=A3=BA* Jack Vogel >> *=B7=A2=CB=CD=CA=B1=BC=E4=A3=BA* 2010-03-12 02:57:15 >> *=CA=D5=BC=FE=C8=CB=A3=BA* rihad >> *=B3=AD=CB=CD=A3=BA* freebsd-net@freebsd.org >> *=D6=F7=CC=E2=A3=BA* Re: ixgbe input errors at high data rates >> I have typically not had the bandwidth to do performance work on the >> drivers, I am the >> only FreeBSD engineer in the wired network division here at Intel, howev= er >> for 10G I >> have changed the rules and have been working to make the driver perform >> well. >> >> >> 10G is quite different from 1G, in order for it to work well its not jus= t a >> matter of slapping >> in the hardware, the OS must be tuned, and the driver. Further, its an >> ongoing effort, >> so by recommending the CURRENT driver what I'm saying is that is my most >> recent >> code and thus, if all is well :), its also the best performing. >> >> >> The difference between things being tweaked vs not is quite dramatic, l= ike >> getting >> >> only 2 or 3 Gb versus getting 9.5 when properly set up. The multiqueue s= tack >> support >> in 8.0 is part of the equation, and in 7.X you wont have that. >> >> Hope this clarifies things some, >> >> Jack >> >> >> On Thu, Mar 11, 2010 at 10:39 AM, rihad <rihad@mail.ru > wrote: >> >> > Jack Vogel wrote: >> > >> > > Similarly, I have done lots of work in two years to the ixgbe driver= , >> > > I would even suggest that once you have 8 installed you get the driv= er >> > > from HEAD. >> > > >> > >> > You mean the driver in 8.0 isn't good enough and I'd need to somehow = mix >> > the driver from CURRENT with the rest of 8.0? >> > >> _______________________________________________ >> freebsd-net@freebsd.org mailing list >> http://lists.freebsd.org/mailman/listinfo/freebsd-net >> To unsubscribe, send any mail to "freebsd-net-unsubscribe@freebsd.org" >> > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2a41acea1003112246qd1725f9ve55350c064a4583f>