Date: Sun, 31 Mar 2013 05:11:57 -0700 (PDT) From: Barney Cordoba <barney_cordoba@yahoo.com> To: Scott Long <scott4long@yahoo.com> Cc: Nick Rogers <ncrogers@gmail.com>, Adrian Chadd <adrian@freebsd.org>, Jeffrey EPieper <jeffrey.e.pieper@intel.com>, "freebsd-net@freebsd.org" <freebsd-net@freebsd.org>, "Clement Hermann \(nodens\)" <nodens2099@gmail.com>, Jack Vogel <jfvogel@gmail.com> Subject: Re: igb and ALTQ in 9.1-rc3 Message-ID: <1364731917.68949.YahooMailClassic@web121603.mail.ne1.yahoo.com> In-Reply-To: <F024F568-0749-4132-9AB1-5010ED531B04@yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Do you know anything about the subject, Scott? I'd be interested in seeing = your benchmarks with various queue counts, binding to cpus vs not binding, = and the numbers comparing the pre multiqueue driver to the current one. It'= s the minimum that any marginally competent network driver developer would = do.=0A=0AOr are you just hurling insults because you're devoid of actual id= eas?=0A=0ABC=0A=0A=0A--- On Fri, 3/29/13, Scott Long <scott4long@yahoo.com>= wrote:=0A=0A> From: Scott Long <scott4long@yahoo.com>=0A> Subject: Re: igb= and ALTQ in 9.1-rc3=0A> To: "Barney Cordoba" <barney_cordoba@yahoo.com>=0A= > Cc: "Nick Rogers" <ncrogers@gmail.com>, "Adrian Chadd" <adrian@freebsd.or= g>, "Jeffrey EPieper" <jeffrey.e.pieper@intel.com>, "freebsd-net@freebsd.or= g" <freebsd-net@freebsd.org>, "Clement Hermann (nodens)" <nodens2099@gmail.= com>, "Jack Vogel" <jfvogel@gmail.com>=0A> Date: Friday, March 29, 2013, 12= :42 PM=0A> Comedy gold.=A0 It's been a while=0A> since I've seen this much = idiocy from you, Barney.=A0=0A> Hopefully the rest of the mailing list will= blackhole you,=0A> as I'm about to, and we can all get back to real work.= =0A> =0A> Scott=0A> =0A> =0A> =0A> On Mar 29, 2013, at 10:38 AM, Barney Cor= doba <barney_cordoba@yahoo.com>=0A> wrote:=0A> =0A> > it needs a lot more t= han a patch. It needs to be=0A> completely re-thunk=0A> > =0A> > --- On Fri= , 3/29/13, Adrian Chadd <adrian@freebsd.org>=0A> wrote:=0A> > =0A> > From: = Adrian Chadd <adrian@freebsd.org>=0A> > Subject: Re: igb and ALTQ in 9.1-rc= 3=0A> > To: "Barney Cordoba" <barney_cordoba@yahoo.com>=0A> > Cc: "Jack Vog= el" <jfvogel@gmail.com>,=0A> "Nick Rogers" <ncrogers@gmail.com>,=0A> "Jeffr= ey EPieper" <jeffrey.e.pieper@intel.com>,=0A> "freebsd-net@freebsd.org"=0A>= <freebsd-net@freebsd.org>,=0A> "Clement Hermann (nodens)" <nodens2099@gmai= l.com>=0A> > Date: Friday, March 29, 2013, 12:07 PM=0A> > =0A> > Barney,=0A= > > Patches gratefully accepted.=0A> > =0A> > =0A> > =0A> > Adrian=0A> > = =0A> > =0A> > =0A> > On 29 March 2013 08:54, Barney Cordoba <barney_cordoba= @yahoo.com>=0A> wrote:=0A> > =0A> > =0A> > =0A> > =0A> > =0A> > --- On Fri,= 3/29/13, Pieper, Jeffrey E <jeffrey.e.pieper@intel.com>=0A> wrote:=0A> > = =0A> > =0A> > =0A> >> From: Pieper, Jeffrey E <jeffrey.e.pieper@intel.com>= =0A> > =0A> >> Subject: RE: igb and ALTQ in 9.1-rc3=0A> > =0A> >> To: "Barn= ey Cordoba" <barney_cordoba@yahoo.com>,=0A> "Jack Vogel" <jfvogel@gmail.com= >,=0A> "Nick Rogers" <ncrogers@gmail.com>=0A> > =0A> > =0A> >> Cc: "freebsd= -net@freebsd.org"=0A> <freebsd-net@freebsd.org>,=0A> "Clement Hermann (node= ns)" <nodens2099@gmail.com>=0A> > =0A> > =0A> >> Date: Friday, March 29, 20= 13, 11:45 AM=0A> > =0A> >> =0A> > =0A> >> =0A> > =0A> >> -----Original Mess= age-----=0A> > =0A> >> From: owner-freebsd-net@freebsd.org=0A> > =0A> >> [m= ailto:owner-freebsd-net@freebsd.org]=0A> > =0A> >> On Behalf Of Barney Cord= oba=0A> > =0A> >> Sent: Friday, March 29, 2013 5:51 AM=0A> > =0A> >> To: Ja= ck Vogel; Nick Rogers=0A> > =0A> >> Cc: freebsd-net@freebsd.org;=0A> > =0A>= >> Clement Hermann (nodens)=0A> > =0A> >> Subject: Re: igb and ALTQ in 9.1= -rc3=0A> > =0A> >> =0A> > =0A> >> =0A> > =0A> >> =0A> > =0A> >> --- On Thu,= 3/28/13, Nick Rogers <ncrogers@gmail.com>=0A> > =0A> >> wrote:=0A> > =0A> = >> =0A> > =0A> >>> From: Nick Rogers <ncrogers@gmail.com>=0A> > =0A> >>> Su= bject: Re: igb and ALTQ in 9.1-rc3=0A> > =0A> >>> To: "Jack Vogel" <jfvogel= @gmail.com>=0A> > =0A> >>> Cc: "Barney Cordoba" <barney_cordoba@yahoo.com>,= =0A> > =0A> >> "Clement Hermann (nodens)" <nodens2099@gmail.com>,=0A> > =0A= > >> "freebsd-net@freebsd.org"=0A> > =0A> >> <freebsd-net@freebsd.org>=0A> = > =0A> >>> Date: Thursday, March 28, 2013, 9:29 PM=0A> > =0A> >>> On Thu, M= ar 28, 2013 at 4:16 PM, Jack=0A> > =0A> >>> Vogel <jfvogel@gmail.com>=0A> >= =0A> >>> wrote:=0A> > =0A> >>>> Have been kept fairly busy with other=0A> = matters,=0A> > =0A> >> one=0A> > =0A> >>> thing I could do short=0A> > =0A>= >>>> term is=0A> > =0A> >>>> change the defines in igb the way I did in=0A= > the em=0A> > =0A> >>> driver so you could still=0A> > =0A> >>>> define=0A= > > =0A> >>>> the older if_start entry. Right now those=0A> are=0A> > =0A> = >> based on=0A> > =0A> >>> OS version and so you=0A> > =0A> >>>> will=0A> >= =0A> >>>> automatically get if_transmit, but I could=0A> change=0A> > =0A>= >> it to=0A> > =0A> >>> be IGB_LEGACY_TX or=0A> > =0A> >>>> so,=0A> > =0A>= >>>> and that could be defined in the Makefile.=0A> > =0A> >>>> =0A> > =0A= > >>>> Would this help?=0A> > =0A> >>> =0A> > =0A> >>> I'm currently using = ALTQ successfully with the=0A> em=0A> > =0A> >> driver, so=0A> > =0A> >>> i= f igb=0A> > =0A> >>> behaved the same with respect to using if_start=0A> in= stead=0A> > =0A> >> of=0A> > =0A> >>> if_transmit=0A> > =0A> >>> when ALTQ = is in play, that would be great. I do=0A> not=0A> > =0A> >>> completely=0A>= > =0A> >>> understand the change you propose as I am not=0A> very=0A> > = =0A> >> familiar=0A> > =0A> >>> with the=0A> > =0A> >>> driver internals. A= ny kind of patch or extra=0A> > =0A> >>> Makefile/make.conf=0A> > =0A> >>> = definition that would allow me to build a=0A> 9-STABLE=0A> > =0A> >> kernel= =0A> > =0A> >>> with an igb=0A> > =0A> >>> driver that works again with ALT= Q, ASAP, would=0A> be much=0A> > =0A> >>> appreciated.=0A> > =0A> >>> =0A> = > =0A> >>>> =0A> > =0A> >>>> Jack=0A> > =0A> >>>> =0A> > =0A> >>>> =0A> > = =0A> >>>> =0A> > =0A> >>>> On Thu, Mar 28, 2013 at 2:31 PM, Nick=0A> Rogers= =0A> > =0A> >> <ncrogers@gmail.com>=0A> > =0A> >>> wrote:=0A> > =0A> >>>>> = =0A> > =0A> >>>>> On Tue, Dec 11, 2012 at 1:09 AM, Jack=0A> Vogel=0A> > =0A= > >> <jfvogel@gmail.com>=0A> > =0A> >>> wrote:=0A> > =0A> >>>>>> On Mon, De= c 10, 2012 at 11:58 PM,=0A> Gleb=0A> > =0A> >>> Smirnoff <glebius@freebsd.o= rg>=0A> > =0A> >>>>>> wrote:=0A> > =0A> >>>>>> =0A> > =0A> >>>>>>> On Mon, = Dec 10, 2012 at=0A> 03:31:19PM=0A> > =0A> >> -0800,=0A> > =0A> >>> Jack Vog= el wrote:=0A> > =0A> >>>>>>> J> UH, maybe asking the=0A> owner of=0A> > =0A= > >> the=0A> > =0A> >>> driver would help :)=0A> > =0A> >>>>>>> J>=0A> > = =0A> >>>>>>> J> ... and no, I've never=0A> been=0A> > =0A> >> aware of=0A> = > =0A> >>> doing anything to stop=0A> > =0A> >>>>>>> supporting=0A> > =0A> = >>>>>>> altq=0A> > =0A> >>>>>>> J> so you wouldn't see any=0A> > =0A> >> co= mmits. If=0A> > =0A> >>> there's something in the altq=0A> > =0A> >>>>>>> c= ode=0A> > =0A> >>>>>>> or=0A> > =0A> >>>>>>> J> support (which I have=0A> n= othing=0A> > =0A> >> to do=0A> > =0A> >>> with) that caused this no-one=0A>= > =0A> >>>>>>> informed=0A> > =0A> >>>>>>> J> me.=0A> > =0A> >>>>>>> =0A> = > =0A> >>>>>>> Switching from if_start to=0A> > =0A> >> if_transmit=0A> > = =0A> >>> effectively disables ALTQ=0A> > =0A> >>>>>>> support.=0A> > =0A> >= >>>>>> =0A> > =0A> >>>>>>> AFAIR, there is some magic=0A> > =0A> >> impleme= nted in=0A> > =0A> >>> other drivers that makes them=0A> > =0A> >>>>>>> mod= ern (that means using=0A> > =0A> >> if_transmit), but=0A> > =0A> >>> still = capable to switch to=0A> > =0A> >>>>>>> queueing=0A> > =0A> >>>>>>> mode if= SIOCADDALTQ was casted=0A> upon=0A> > =0A> >> them.=0A> > =0A> >>>>>>> =0A= > > =0A> >>>>>>> =0A> > =0A> >>>>>> Oh, hmmm, I'll look into the matter=0A>= after=0A> > =0A> >> my=0A> > =0A> >>> vacation.=0A> > =0A> >>>>>> =0A> > = =0A> >>>>>> Jack=0A> > =0A> >>>>> =0A> > =0A> >>>>> Has there been any prog= ress on=0A> resolving this=0A> > =0A> >>> issue? I recently ran=0A> > =0A> = >>>>> into this problem upgrading my servers=0A> from=0A> > =0A> >> 8.3 to= =0A> > =0A> >>> 9.1-RELEASE and am=0A> > =0A> >>>>> wondering what the late= st=0A> recommendation is.=0A> > =0A> >> I've=0A> > =0A> >>> used ALTQ and i= gb=0A> > =0A> >>>>> successfully for years and it is=0A> unfortunate=0A> > = =0A> >> it no=0A> > =0A> >>> longer works.=0A> > =0A> >>>>> Appreciate any = advice.=0A> > =0A> >>>>> =0A> > =0A> >>> =0A> > =0A> >>> Do yourself a favo= r and either get a cheap dual=0A> port=0A> > =0A> >> 82571 card or=0A> > = =0A> >>> 2 cards and disable the IGB ports. The igb=0A> driver is=0A> > =0A= > >> defective, and until=0A> > =0A> >>> they back out the new, untested mu= lti-queue=0A> stuff you're=0A> > =0A> >> just neutering=0A> > =0A> >>> your= system trying to use it.=0A> > =0A> >>> =0A> > =0A> >>> Frankly this proje= ct made a huge mistake by=0A> moving=0A> > =0A> >> forward with multi=0A> >= =0A> >>> queue just for the sake of saying that you=0A> support it;=0A> > = =0A> >> without having=0A> > =0A> >>> any credible plan for implementing it= . That=0A> nonsense=0A> > =0A> >> that Bill Macy did=0A> > =0A> >>> should = have been tarballed up and deposited in=0A> the trash=0A> > =0A> >> folder.= The=0A> > =0A> >>> biggest mess in programming history.=0A> > =0A> >>> =0A= > > =0A> >>> That being said, the solution is not to hack=0A> the igb=0A> >= =0A> >> driver; its to make=0A> > =0A> >>> ALTQ if_transmit compatible, wh= ich shouldn't be=0A> all that=0A> > =0A> >> difficult.=0A> > =0A> >>> =0A> = > =0A> >>> BC=0A> > =0A> >> =0A> > =0A> >> I may be misunderstanding what y= ou are saying, but=0A> if the=0A> > =0A> >> solution is, as you say "not to= hack the igb=0A> driver", then=0A> > =0A> >> how is it defective in this c= ase? Or are you just=0A> directing=0A> > =0A> >> vitriol toward Intel? Mult= i-queue is working fine=0A> in igb.=0A> > =0A> >> =0A> > =0A> >> Jeff=0A> >= =0A> > =0A> > =0A> > It's defective because it's been poorly implemented a= nd=0A> has more bugs=0A> > =0A> > than a Manhattan hotel bed. Adding queues= without a=0A> proper plan just add=0A> > =0A> > more lock contention. It's= not a production-ready=0A> driver.=0A> > =0A> > =0A> > =0A> > As Jack once= said, Intel doesn't care about=0A> performance, they're just=0A> > =0A> > = example drivers. igb is an example of how not to do=0A> things.=0A> > =0A> = > =0A> > =0A> > BC=0A> > =0A> > ___________________________________________= ____=0A> > =0A> > freebsd-net@freebsd.org=0A> mailing list=0A> > =0A> > htt= p://lists.freebsd.org/mailman/listinfo/freebsd-net=0A> > =0A> > To unsubscr= ibe, send any mail to "freebsd-net-unsubscribe@freebsd.org"=0A> > =0A> > = =0A> > =0A> > _______________________________________________=0A> > freebsd= -net@freebsd.org=0A> mailing list=0A> > http://lists.freebsd.org/mailman/li= stinfo/freebsd-net=0A> > To unsubscribe, send any mail to "freebsd-net-unsu= bscribe@freebsd.org"=0A> =0A> _____________________________________________= __=0A> freebsd-net@freebsd.org=0A> mailing list=0A> http://lists.freebsd.or= g/mailman/listinfo/freebsd-net=0A> To unsubscribe, send any mail to "freebs= d-net-unsubscribe@freebsd.org"=0A>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1364731917.68949.YahooMailClassic>