From: Jamal Hadi <hadi@shell.cyberus.ca> To: Andi Kleen <ak@suse.de> Cc: Robert Olsson <Robert.Olsson@data.slu.se>, Bogdan Costescu <bogdan.costescu@iwr.uni-heidelberg.de>, "David S. Miller" <davem@redhat.com>, sim@netnation.com, ralph+d@istop.com, xerox@foonet.net, fw@deneb.enyo.de, netdev@oss.sgi.com, linux-net@vger.kernel.org Subject: Re: 3c59x (was Route cache performance under stress) Date: Wed, 11 Jun 2003 08:08:00 -0400 (EDT) [thread overview] Message-ID: <20030611075703.R39786@shell.cyberus.ca> (raw) In-Reply-To: <20030611100520.GB27119@oldwotan.suse.de> On Wed, 11 Jun 2003, Andi Kleen wrote: > eth_type_trans checks the ethernet protocol ID and sets the broadcast/multicast/ > unicast L2 type. > > Some NICs have bits in the RX descriptor for most of them. They have a > "packet is TCP or UDP or IP" bit and also a bit for unicast or sometimes > even multicast/broadcast. So when you have the RX descriptor you > can just derive these values from there and put them into the skb > without calling eth_type_trans or looking at the cache cold header. > > Then you do a prefetch on the header. When the packet reaches the > network stack later the header has already reached cache and it can be > processed without a memory round trip latency. > I have done prefetching experiments with a NAPIezed sb1250.c driver on MIPS. I never got rid of eth_type_trans() just prefetched skb->data a few lines before calling it. I did see eth_type_trans() almost disappear from the profile (it was way low to be important). Andis idea is even more interesting. I did see i think about 10Kpps more in throughput. Robert, this means our biggest bottleneck right now is cache misses. The MIPS processor i am playing with is SMP and has a large shared L2 cache. What i am observing is that this is quiet useful for SMP. I am limited by how much traffic i can generate right now to test it more. I can do 295Kpps L3 easy. This board is an excuse for you to come down to Ottawa in July ;-> > Caveats: > On some cards it doesn't work for all packets or can be only done > if you don't have any multicast addresses hashed (that's the case > for the e1000 if I read the header bits correctly). The lxt1001 > (old EOLed card) can do it for all packet types. > So can the sb1250. I'll try this out. > Often prefetch size is limited so you should not prefetch more > than what you can store until the packet reaches the stack. > Good point. So is there a systematic way to find out the effects of the prefecth size or you just have to keep trying until you get it right? cheers, jamal
next prev parent reply other threads:[~2003-06-11 12:08 UTC|newest]
Thread overview: 216+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <87d6iit4g7.fsf@deneb.enyo.de>
[not found] ` <20030517.150933.74723581.davem@redhat.com>
[not found] ` <87iss87gqd.fsf@deneb.enyo.de>
2003-05-18 9:31 ` Route cache performance under stress David S. Miller
2003-05-19 17:36 ` Jamal Hadi
2003-05-19 19:18 ` Ralph Doncaster
2003-05-19 22:37 ` Jamal Hadi
2003-05-20 1:10 ` Simon Kirby
2003-05-20 1:14 ` David S. Miller
2003-05-20 1:23 ` Jamal Hadi
2003-05-20 1:24 ` David S. Miller
2003-05-20 2:13 ` Jamal Hadi
2003-05-20 5:01 ` Pekka Savola
2003-05-20 11:47 ` Jamal Hadi
2003-05-20 11:55 ` Pekka Savola
2003-05-20 6:46 ` David S. Miller
2003-05-20 12:04 ` Jamal Hadi
2003-05-21 0:36 ` David S. Miller
2003-05-21 13:03 ` Jamal Hadi
2003-05-23 5:42 ` David S. Miller
2003-05-22 8:40 ` Simon Kirby
2003-05-22 8:58 ` David S. Miller
2003-05-22 10:40 ` David S. Miller
2003-05-22 11:15 ` Martin Josefsson
2003-05-23 1:00 ` David S. Miller
2003-05-23 1:01 ` David S. Miller
2003-05-23 8:21 ` Andi Kleen
2003-05-23 8:22 ` David S. Miller
2003-05-23 9:03 ` Andi Kleen
2003-05-23 9:59 ` David S. Miller
2003-05-24 0:41 ` Andrew Morton
2003-05-26 2:29 ` David S. Miller
2003-05-22 11:44 ` Simon Kirby
2003-05-22 13:03 ` Martin Josefsson
2003-05-23 0:55 ` David S. Miller
2003-05-22 22:33 ` David S. Miller
2003-05-29 20:51 ` Simon Kirby
2003-06-02 10:58 ` Robert Olsson
2003-06-02 15:18 ` Simon Kirby
2003-06-02 16:36 ` Robert Olsson
2003-06-02 18:05 ` Simon Kirby
2003-06-09 17:21 ` David S. Miller
2003-06-09 17:19 ` David S. Miller
2003-05-23 0:59 ` David S. Miller
2003-05-26 7:18 ` Florian Weimer
2003-05-26 7:29 ` David S. Miller
2003-05-26 9:34 ` Florian Weimer
2003-05-27 6:32 ` David S. Miller
2003-06-08 11:39 ` Florian Weimer
2003-06-08 12:05 ` David S. Miller
2003-06-08 13:10 ` Florian Weimer
2003-06-08 23:49 ` Simon Kirby
2003-06-08 23:55 ` CIT/Paul
2003-06-09 3:15 ` Jamal Hadi
2003-06-09 5:27 ` CIT/Paul
2003-06-09 5:58 ` David S. Miller
2003-06-09 6:28 ` CIT/Paul
2003-06-09 6:28 ` David S. Miller
2003-06-09 16:23 ` Stephen Hemminger
2003-06-09 16:37 ` David S. Miller
2003-06-09 7:13 ` Simon Kirby
2003-06-09 8:10 ` CIT/Paul
2003-06-09 8:27 ` Simon Kirby
2003-06-09 19:38 ` CIT/Paul
2003-06-09 21:30 ` David S. Miller
2003-06-09 22:19 ` Simon Kirby
2003-06-09 22:54 ` Robert Olsson
2003-06-13 6:21 ` David S. Miller
2003-06-13 10:40 ` Robert Olsson
2003-06-15 6:36 ` David S. Miller
2003-06-17 17:03 ` Robert Olsson
2003-06-09 22:56 ` CIT/Paul
2003-06-09 23:05 ` David S. Miller
2003-06-10 13:41 ` Robert Olsson
2003-06-10 0:03 ` Jamal Hadi
2003-06-10 0:32 ` Ralph Doncaster
2003-06-10 1:15 ` Jamal Hadi
2003-06-10 2:45 ` Ralph Doncaster
2003-06-10 3:23 ` Ben Greear
2003-06-10 3:41 ` Ralph Doncaster
2003-06-10 18:10 ` Ralph Doncaster
2003-06-10 18:21 ` Ben Greear
2003-06-10 4:34 ` Simon Kirby
2003-06-10 11:01 ` Jamal Hadi
2003-06-10 11:28 ` Jamal Hadi
2003-06-10 13:18 ` Ralph Doncaster
2003-06-10 16:10 ` David S. Miller
2003-06-10 10:53 ` Jamal Hadi
2003-06-10 11:41 ` chas williams
2003-06-10 16:27 ` David S. Miller
2003-06-10 16:57 ` chas williams
2003-06-10 11:41 ` Pekka Savola
2003-06-10 11:58 ` John S. Denker
2003-06-10 12:12 ` Jamal Hadi
2003-06-10 16:33 ` David S. Miller
2003-06-10 12:07 ` Jamal Hadi
2003-06-10 15:29 ` Ralph Doncaster
2003-06-11 19:48 ` Florian Weimer
2003-06-11 19:40 ` CIT/Paul
2003-06-11 21:09 ` Florian Weimer
2003-06-10 13:10 ` Ralph Doncaster
2003-06-10 13:36 ` Jamal Hadi
2003-06-10 14:03 ` Ralph Doncaster
2003-06-10 16:38 ` David S. Miller
2003-06-10 16:39 ` David S. Miller
2003-06-10 18:41 ` Florian Weimer
2003-06-11 11:47 ` Was (Re: " Jamal Hadi
2003-06-11 18:41 ` Real World Routers 8-) Florian Weimer
2003-06-10 15:53 ` Route cache performance under stress David S. Miller
2003-06-10 16:15 ` 3c59x (was Route cache performance under stress) Bogdan Costescu
2003-06-10 16:20 ` Andi Kleen
2003-06-10 16:23 ` Jeff Garzik
2003-06-10 17:02 ` 3c59x David S. Miller
2003-06-10 17:16 ` 3c59x Jeff Garzik
2003-06-10 17:14 ` 3c59x David S. Miller
2003-06-10 17:25 ` 3c59x Jeff Garzik
2003-06-10 17:30 ` 3c59x David S. Miller
2003-06-10 19:20 ` 3c59x Jeff Garzik
2003-06-10 19:21 ` 3c59x David S. Miller
2003-06-10 17:18 ` 3c59x Andi Kleen
2003-06-10 17:29 ` 3c59x chas williams
2003-06-10 17:31 ` 3c59x David S. Miller
2003-06-10 17:39 ` 3c59x chas williams
2003-06-10 17:43 ` 3c59x David S. Miller
2003-06-11 17:52 ` Route cache performance under stress Robert Olsson
2003-06-10 1:53 ` Simon Kirby
2003-06-10 3:18 ` Ralph Doncaster
2003-06-10 16:06 ` David S. Miller
2003-06-10 15:56 ` David S. Miller
2003-06-10 16:45 ` 3c59x (was Route cache performance under stress) Bogdan Costescu
2003-06-10 16:49 ` Andi Kleen
2003-06-11 9:54 ` Robert Olsson
2003-06-11 10:05 ` Andi Kleen
2003-06-11 10:38 ` Robert Olsson
2003-06-11 12:08 ` Jamal Hadi [this message]
2003-06-10 17:12 ` 3c59x David S. Miller
2003-06-10 17:19 ` Route cache performance under stress Ralph Doncaster
2003-06-10 15:49 ` David S. Miller
2003-06-10 17:33 ` Ralph Doncaster
2003-06-10 17:32 ` David S. Miller
2003-06-10 18:34 ` Robert Olsson
2003-06-10 18:57 ` David S. Miller
2003-06-10 19:53 ` Robert Olsson
2003-06-10 21:36 ` CIT/Paul
2003-06-10 21:39 ` Ralph Doncaster
2003-06-10 22:20 ` David S. Miller
2003-06-10 23:58 ` Ralph Doncaster
2003-06-10 23:57 ` David S. Miller
2003-06-11 0:41 ` Ralph Doncaster
2003-06-11 0:58 ` David S. Miller
2003-06-11 0:58 ` David S. Miller
2003-06-11 0:51 ` Ben Greear
2003-06-11 1:01 ` David S. Miller
2003-06-11 1:15 ` Ben Greear
2003-06-11 1:22 ` David S. Miller
2003-06-11 1:51 ` Ben Greear
2003-06-11 3:33 ` David S. Miller
2003-06-11 11:54 ` gettime: Was (Re: " Jamal Hadi
2003-06-11 12:08 ` Andi Kleen
2003-06-12 3:30 ` David S. Miller
2003-06-12 6:32 ` Ben Greear
2003-06-12 8:46 ` David S. Miller
2003-06-11 15:57 ` Ben Greear
2003-06-12 3:29 ` David S. Miller
2003-06-11 1:17 ` Ralph Doncaster
2003-06-11 1:23 ` David S. Miller
2003-06-11 7:28 ` Andi Kleen
2003-06-11 7:25 ` Andi Kleen
2003-06-11 17:40 ` Robert Olsson
2003-06-13 5:38 ` David S. Miller
2003-06-13 10:22 ` Robert Olsson
2003-06-13 17:15 ` Robert Olsson
2003-06-12 6:45 ` David S. Miller
2003-06-12 13:56 ` Robert Olsson
2003-06-12 21:35 ` David S. Miller
2003-06-13 10:50 ` Robert Olsson
2003-06-10 0:56 ` Ralph Doncaster
2003-06-09 11:38 ` Jamal Hadi
2003-06-09 11:55 ` David S. Miller
2003-06-09 12:18 ` Jamal Hadi
2003-06-09 12:32 ` David S. Miller
2003-06-09 13:22 ` Jamal Hadi
2003-06-09 13:22 ` David S. Miller
2003-06-09 8:56 ` David S. Miller
2003-06-09 22:39 ` Robert Olsson
2003-06-09 6:25 ` David S. Miller
2003-06-09 6:59 ` Simon Kirby
2003-06-09 7:03 ` David S. Miller
2003-06-09 13:04 ` Ralph Doncaster
2003-06-09 13:26 ` Jamal Hadi
2003-06-09 5:44 ` David S. Miller
2003-06-09 5:51 ` CIT/Paul
2003-06-09 6:03 ` David S. Miller
2003-06-09 6:52 ` Simon Kirby
2003-06-09 6:56 ` David S. Miller
2003-06-09 7:36 ` Simon Kirby
2003-06-09 8:18 ` Simon Kirby
2003-06-09 8:22 ` David S. Miller
2003-06-09 8:31 ` Simon Kirby
2003-06-09 9:01 ` David S. Miller
2003-06-09 9:47 ` Andi Kleen
2003-06-09 10:03 ` David S. Miller
2003-06-09 10:13 ` Andi Kleen
2003-06-09 10:13 ` David S. Miller
2003-06-09 10:40 ` YOSHIFUJI Hideaki / 吉藤英明
2003-06-09 10:40 ` David S. Miller
2003-06-09 14:14 ` David S. Miller
2003-06-09 6:47 ` Simon Kirby
2003-06-09 6:49 ` David S. Miller
2003-06-09 13:28 ` Ralph Doncaster
2003-06-09 16:30 ` Simon Kirby
2003-06-17 20:58 ` Florian Weimer
2003-06-09 5:38 ` David S. Miller
2003-06-10 3:05 ` Steven Blake
2003-06-12 6:31 ` David S. Miller
2003-06-08 17:58 ` Pekka Savola
2003-05-21 0:09 ` Simon Kirby
2003-05-21 0:13 ` David S. Miller
2003-05-26 9:29 ` Florian Weimer
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20030611075703.R39786@shell.cyberus.ca \
--to=hadi@shell.cyberus.ca \
--cc=Robert.Olsson@data.slu.se \
--cc=ak@suse.de \
--cc=bogdan.costescu@iwr.uni-heidelberg.de \
--cc=davem@redhat.com \
--cc=fw@deneb.enyo.de \
--cc=linux-net@vger.kernel.org \
--cc=netdev@oss.sgi.com \
--cc=ralph+d@istop.com \
--cc=sim@netnation.com \
--cc=xerox@foonet.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).