netdev - Re: [PATCH NEXT 3/5] qlcnic: vlan lro support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Fri, 17 Sep 2010 10:28:26 +0200
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	"\"Oleg A. Arkhangelsky\"" <sysoleg@...dex.ru>
Cc:	Amit Salecha <amit.salecha@...gic.com>, netdev@...r.kernel.org
Subject: Re: [PATCH NEXT 3/5] qlcnic: vlan lro support

Le vendredi 17 septembre 2010 à 11:34 +0400, "Oleg A. Arkhangelsky" a
écrit :
> 
> 17.09.2010, 11:10, "Amit Salecha" <amit.salecha@...gic.com>:
> >>  -----Original Message-----
> >>  From: Eric Dumazet [mailto:eric.dumazet@...il.com]
> >>  Sent: Thursday, September 16, 2010 6:57 PM
> >>  To: Amit Salecha
> >>  Cc: davem@...emloft.net; netdev@...r.kernel.org; Ameen Rahman; Anirban
> >>  Chakraborty
> >>  Subject: Re: [PATCH NEXT 3/5] qlcnic: vlan lro support
> >>
> >>  Le jeudi 16 septembre 2010 à 06:14 -0700, Amit Kumar Salecha a écrit :
> >>>  LRO + GRO + vlan rx accleration support, performance increases
> >>>  around 20% and cpu utilization reduces around 70% on vlan interface.
> >>  Interesting. What is the workload that demonstrates such gains ?
> >
> > Just to demonstrate decrease in cpu utilization further:
> > I have taken numbers with various load. I used iperf application to run tcp traffic on vlan interface.
> 
> > There is around 80-90% of reduction in cpu utilization.
> 
> I'm wondering, how can this scale further. For example if I have 200K low speed simultaneous
> TCP connections (router appliance case with many users). Can GRO give us some gain here?
> 

I dont think it can help in a mixed trafic:

GRO in itself has an overhead and cannot handle more than 8 different
sessions per NAPI run.

UDP frames hit the overhead but not the GRO gain.

Because most tcp sessions are short lived and initcwnd is small, most
frames wont be coalesced.



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html