Some suggestions on how to proceed on receive side:
1. support for distributing receive processing across multiple CPUs
(using NIC hw queues).
Multiple hw queues can be used to spread receive processing across CPUs;
this will eliminate main cpu% as a bottleneck for 10GbE performance.
Using a NI
Signed-off-by: Alexey Dobriyan <[EMAIL PROTECTED]>
---
net/TUNABLE | 11 ---
1 files changed, 11 deletions(-)
--- linux-vanilla/net/TUNABLE
+++ linux-net-TUNABLE/net/TUNABLE
@@ -15,30 +15,19 @@ AX25_DEF_N2
AX25_DEF_AXDEFMODE 8 = normal 128 is PE1CHL extended
AX25_DEF_IPDEFMODE
Many asked for it, so I spent the better part of today
putting it together, here it is:
http://vger.kernel.org/~davem/cgi-bin/blog.cgi/index.rss
to get the normal HTML version the new URL is
http://vger.kernel.org/~davem/cgi-bin/blog.cgi/index.html
If you try to go to the old URL, it'll just p
On Fri, Jul 29, 2005 at 11:46:13AM -0700, Max Krasnyansky ([EMAIL PROTECTED])
wrote:
> Evgeniy Polyakov wrote:
>
> >>I'm almost convinced that remap can brake even with ring buffer at 1500
> >>sizes. However we're talking about total per packet overhead here. At some
> >>point you have to _unmap_
On Jul 29, 2005, at 12:53 AM, Holger Eitzenberger wrote:
Another drawback of your approach is that it "cummulative", if I
set debug=13 I get in fact output for almost all flags, which might
be a lot.
Why not pass the flags directly (e. g. debug=0x11ff)?
Use ethtool -s ethX msglvl to set
Hi, its your favourite bug reporter again ;)
A Gentoo user (on CC) has reported many of these messages appearing:
skge ignoring bogus sensor interrups
They appear so frequently that 3GB of logs are generated in 1 hour of usage.
It looks like there is a problem in the skge_error_irq fun
The Linksys EG1032 uses Realtek's 8169 chipset.
Credit goes to Bob Wilson <[EMAIL PROTECTED]> for the report.
Signed-off-by: Francois Romieu <[EMAIL PROTECTED]>
diff -puN drivers/net/r8169.c~r8169-650 drivers/net/r8169.c
--- a/drivers/net/r8169.c~r8169-650 2005-07-30 00:19:32.552957434 +0200
On Fri, Jul 29, 2005 at 12:13:06PM -0500, Paul Vinson wrote:
> >> > Are you sure eth1 is an e1000? From e1000_main.c:
> >> >
> >> > #ifdef CONFIG_NET_POLL_CONTROLLER
> >> > netdev->poll_controller = e1000_netpoll;
> >> > #endif
> >> >
> >> > That's been there for 17 months.
> >>
> >> Yep, i
Andy Fleming wrote:
Here's the latest version of the patch, done against a cogito linux-2.6
branch. It has a handful of small changes, which I hope will speed
acceptance:
* Interrupts are no longer requested at interrupt time
* There is a function to print out PHY status
* The PHY drivers'
On Fri, Jul 29, 2005 at 12:43:34PM -0700, David S. Miller ([EMAIL PROTECTED])
wrote:
> From: Evgeniy Polyakov <[EMAIL PROTECTED]>
> Date: Fri, 29 Jul 2005 20:55:06 +0400
>
> > Unmapping is repeatedly being done in this driver - update_address() and
> > tlb_flush() does the thing.
> > Current code
From: Evgeniy Polyakov <[EMAIL PROTECTED]>
Date: Fri, 29 Jul 2005 20:55:06 +0400
> Unmapping is repeatedly being done in this driver - update_address() and
> tlb_flush() does the thing.
> Current code does skb freeing when it's slot needs to be used for the
> next skb.
> So this numbers already in
[EMAIL PROTECTED] wrote:
We want to extract our LAN card driver from tulip core driver and make a
new file uli526x.c at tulip folder,
because we have added some ethtool interface support and non-eprom support
in our driver and may be other change
in the futher. If our controllers support are stil
Evgeniy Polyakov wrote:
I'm almost convinced that remap can brake even with ring buffer at 1500
sizes. However we're talking about total per packet overhead here. At some
point you have to _unmap_ the thing once user-space app is done with it.
With ring buffer it's "memcpy()/kfree_skb()" where w
On Fri, Jul 29, 2005 at 09:41:09AM -0600, Christopher Friesen ([EMAIL
PROTECTED]) wrote:
> Evgeniy Polyakov wrote:
> >Couple of numbers...
> >Remapping of the physical page took about 25-50% less time than 1500
> >bytes copying using memcpy().
>
> Presumably as packet size decreases, at some poin
On Fri, Jul 29, 2005 at 09:34:56AM -0700, Max Krasnyansky ([EMAIL PROTECTED])
wrote:
> Evgeniy Polyakov wrote:
> >Couple of numbers...
> >Remapping of the physical page took about 25-50% less time than 1500
> >bytes copying using memcpy().
> >And 15 times faster just after reboot, i.e. without any
Evgeniy Polyakov wrote:
Couple of numbers...
Remapping of the physical page took about 25-50% less time than 1500
bytes copying using memcpy().
And 15 times faster just after reboot, i.e. without anything in the
cache.
CPU is Xeon with HT enabled:
cpu family : 15
model : 2
model n
Evgeniy Polyakov wrote:
Couple of numbers...
Remapping of the physical page took about 25-50% less time than 1500
bytes copying using memcpy().
Presumably as packet size decreases, at some point the memcpy() becomes
cheaper. With your hardware, where is this crossover point?
Chris
-
To unsu
[linux-kernel dropped, netdev added]
On Fri, Jul 29, 2005 at 08:15:20PM +0800, [EMAIL PROTECTED] wrote:
> We want to extract our LAN card driver from tulip core driver and make a
> new file uli526x.c at tulip folder,
> because we have added some ethtool interface support and non-eprom support
> in
David S. Miller wrote:
> From: "David S. Miller" <[EMAIL PROTECTED]>
> Date: Thu, 28 Jul 2005 20:54:24 -0700 (PDT)
>
>>Patrick, I think some parts of this change:
>>
>>[IPV4/6]: Check if packet was actually delivered to a raw socket to
>> decide whether to send an ICMP unreachable
>>
>>might
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
Hi Patrick,
Attached you'll find two patches (against 2.6.12.3) fixing a problem i
had with the combination of ipsec and bridging.
First I'll describe the two patches, then I'll describe the problem.
The first patch updates the nf_reset function to i
Bcc: Evgeniy Polyakov <[EMAIL PROTECTED]>
Subject: [RFC] Netlink subscription/multicasting.
Reply-To: [EMAIL PROTECTED]
Hello, developers.
Here is some thoughts about netlink subscription/multicasting/group
number.
Attached [only compile tested] patch implements two socket options
for netlink so
Just simple spelling mistake fixes.
Signed-Off-By: Baruch Even <[EMAIL PROTECTED]>
diff -Nurp 2.6.13-rc4-orig/include/net/tcp.h 2.6.13-rc4/include/net/tcp.h
--- 2.6.13-rc4-orig/include/net/tcp.h 2005-07-29 11:17:25.0 +0100
+++ 2.6.13-rc4/include/net/tcp.h2005-07-29 11:14:28.
Couple of numbers...
Remapping of the physical page took about 25-50% less time than 1500
bytes copying using memcpy().
And 15 times faster just after reboot, i.e. without anything in the
cache.
CPU is Xeon with HT enabled:
cpu family : 15
model : 2
model name : Intel(R) Xeon(T
Jesse Brandeburg wrote:
Please don't apply this patch, we need the NETIF_MSG_LINK messages to
be printed because most everyone likes to know when their link came
up/went down.
Then your driver need to be more explicit with it. Some other people
will think this is a bug, even more because cur
24 matches
Mail list logo
