Skip to contentSkip to conversation

Can I reduce CPU use of conntrack/masq?

DISCUSSION Steve Snyder 3 participants 4 messages Nov 2, 2005

lore
Share & tools
lore
#0 lore raw

Hello.

I am working on what amounts to a Ethernet to Ultra-Wide Band (UWB) converter box. Packets are simply routed from 1 interface to another.

This box is based on an ARM7TDMI CPU, running Linux 2.4.26, and the network throughput of the box is CPU-limited. How limited? The 100Mbps/FD Ethernet can do no better than 35Mbps.

I've discovered that I can improve Ethernet throughput by about %20 by removing the the conntrack/masq support from the kernel. The removal is good only as a test, though, since I need this functionality to move the packets between interfaces.

This is the relevant config:

CONFIG_IP_NF_CONNTRACK=y
CONFIG_IP_NF_IPTABLES=y
CONFIG_IP_NF_NAT=y
CONFIG_IP_NF_NAT_NEEDED=y
CONFIG_IP_NF_TARGET_MASQUERADE=y

Enabled at boot time like this:

/sbin/iptables -t nat -A POSTROUTING -o uwb0 -j MASQUERADE echo "1" > /proc/sys/net/ipv4/ip_forward

I wonder if I can improve conntrack/masq performance at the expense of flexibility. This will be a closed system, with simple and static routing. Are there any trade-offs I can make to sacrifice unneeded flexibility in routing for reduced CPU utilization in conntrack/masq?

Thanks.

#1 lore raw
Quoted 25 lines · Steve Snyder

Hello.

I am working on what amounts to a Ethernet to Ultra-Wide Band (UWB) converter box. Packets are simply routed from 1 interface to another.

This box is based on an ARM7TDMI CPU, running Linux 2.4.26, and the network throughput of the box is CPU-limited. How limited? The 100Mbps/FD Ethernet can do no better than 35Mbps.

I've discovered that I can improve Ethernet throughput by about %20 by removing the the conntrack/masq support from the kernel. The removal is good only as a test, though, since I need this functionality to move the packets between interfaces.

This is the relevant config:

CONFIG_IP_NF_CONNTRACK=y
CONFIG_IP_NF_IPTABLES=y
CONFIG_IP_NF_NAT=y
CONFIG_IP_NF_NAT_NEEDED=y
CONFIG_IP_NF_TARGET_MASQUERADE=y

Enabled at boot time like this:

Quoted 7 lines

/sbin/iptables -t nat -A POSTROUTING -o uwb0 -j MASQUERADE echo "1" > /proc/sys/net/ipv4/ip_forward

I wonder if I can improve conntrack/masq performance at the expense of flexibility. This will be a closed system, with simple and static routing. Are there any trade-offs I can make to sacrifice unneeded flexibility in routing for reduced CPU utilization in conntrack/masq?

Hmmm... totally untested and don't know the details of UWB but... can't you simply ether-bridge the interfaces instead of masquerading? It should need less CPU

Quoted 7 lines

Thanks.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
Greetz, Antonio Vargas aka winden of network

http://wind.codepixel.com/
[email protected]
[email protected]

Every day, every year
you have to work
you have to study
you have to scene.

#2 lore raw
Quoted 1 line · Antonio Vargas

On 11/2/05, Steve Snyder <[email protected]> wrote:

[snip]

Quoted 8 lines
Quoted 4 lines

I wonder if I can improve conntrack/masq performance at the expense of flexibility. This will be a closed system, with simple and static routing. Are there any trade-offs I can make to sacrifice unneeded flexibility in routing for reduced CPU utilization in conntrack/masq?

Hmmm... totally untested and don't know the details of UWB but... can't you simply ether-bridge the interfaces instead of masquerading? It should need less CPU

Hmm... I'm not familiar with ether-bridge, and Google turns up only commercial products and BSD references.

Pointer to info, please?

Thanks.

#3 lore raw
Quoted 8 lines · On Wed, 2 Nov 2005 15:51:52 -0500 Steve Snyder <[email protected]> wrote:
Quoted 1 line · Antonio Vargas

On 11/2/05, Steve Snyder <[email protected]> wrote:

[snip]

Quoted 5 lines
Quoted 4 lines

I wonder if I can improve conntrack/masq performance at the expense of flexibility. This will be a closed system, with simple and static routing. Are there any trade-offs I can make to sacrifice unneeded flexibility in routing for reduced CPU utilization in conntrack/masq?

Quoted 6 lines
Quoted 3 lines

Hmmm... totally untested and don't know the details of UWB but... can't you simply ether-bridge the interfaces instead of masquerading? It should need less CPU

Hmm... I'm not familiar with ether-bridge, and Google turns up only commercial products and BSD references.

It in the kernel already! Look at
http://linux-net.osdl.org/index.php/Bridge
For more info

Signature · 2 lines
Stephen Hemminger <[email protected]>
OSDL http://developer.osdl.org/~shemminger