Pentium II only doing 20Mbps

Hi, I have a Pentium II 350MHz, 64 MB RAM, HD 8GB, with linux kernel
2.4

I want to use it as a router/nat/firewall.

My problem is that I am getting only 20Mbps in the internal LAN, is
there any kernel limit that stop me getting near to 100Mbps?
TIA,

Nico

nicolas.patik@gmail.com wrote in
news:1102364656.769790.191850@f14g2000cwb.googlegr oups.com:

> Hi, I have a Pentium II 350MHz, 64 MB RAM, HD 8GB, with linux kernel
> 2.4
>
> I want to use it as a router/nat/firewall.
>
> My problem is that I am getting only 20Mbps in the internal LAN, is
> there any kernel limit that stop me getting near to 100Mbps?
> TIA,
>
> Nico
>

Redundancy is the best way to teach idiots.
Redundancy is the best way to teach idiots.
Redundancy is the best way to teach idiots.
Redundancy is the best way to teach idiots.
Redundancy is the best way to teach idiots.

Exactly why are convinced that your situation is a kernel limit?

I read something about it in slashdot and in an MIT paper:
http://developers.slashdot.org/devel...tid=126&tid=95

http://developers.slashdot.org/comme...98&cid=9661019
Re:Performance is pretty reasonable (Score:4, Informative)
by Fzz (153115) on Saturday July 10, @10:34AM (#9661019)
Take a look at figures 17 and 18 of this paper:
Eddie Kohler et al, "The Click modular router [mit.edu]". ACM
Transactions on Computer Systems 18(3), August 2000, pages 263-297.
These experiments are a few years old now, but 32-bit PCI hasn't
changed in that time, so they should still be valid on non-server-class
PCs. Vanilla Linux topped out at around 80Kpps, whereas polling gets
you over 300Kpps, and the Click optimizations get you nearer 400Kpps.

Similar experiments on FreeBSD with device polling give results in the
same ballpark.

- Mark

"Nico" <nicolas.patik@gmail.com> wrote in
news:1102417473.133552.200330@z14g2000cwz.googlegr oups.com:

> I read something about it in slashdot and in an MIT paper:
> http://developers.slashdot.org/devel...3.shtml?tid=12
> 6&tid=95
>
> http://developers.slashdot.org/comme...98&cid=9661019
> Re:Performance is pretty reasonable (Score:4, Informative)
> by Fzz (153115) on Saturday July 10, @10:34AM (#9661019)
> Take a look at figures 17 and 18 of this paper:
> Eddie Kohler et al, "The Click modular router [mit.edu]". ACM
> Transactions on Computer Systems 18(3), August 2000, pages 263-297.
> These experiments are a few years old now, but 32-bit PCI hasn't
> changed in that time, so they should still be valid on
> non-server-class PCs. Vanilla Linux topped out at around 80Kpps,
> whereas polling gets you over 300Kpps, and the Click optimizations get
> you nearer 400Kpps.
>
> Similar experiments on FreeBSD with device polling give results in the
> same ballpark.
>
> - Mark
>

I have no idea what a Kpps is.
On more than one occasion I have seen cases where "something bad"
happens to one of the two NIC's connected by a cable where one
NIC goes into single "duplex" at 10Mbs; which degrades throughput
below 100Mbs.

Your original "conclusion" that the problem must be in the kernel
may be in error; then again it may not. But without checking other
alternatives, you won't know for sure.

HTH & YMMV

IANAL_VISTA wrote:
> "Nico" <nicolas.patik@gmail.com> wrote in
8<
>>These experiments are a few years old now, but 32-bit PCI hasn't
>>changed in that time, so they should still be valid on
>>non-server-class PCs. Vanilla Linux topped out at around 80Kpps,
>>whereas polling gets you over 300Kpps, and the Click optimizations get
>>you nearer 400Kpps.
>>
>>Similar experiments on FreeBSD with device polling give results in the
>>same ballpark.

this stuff is old and 2.6 kernels seem to be alot better.

> I have no idea what a Kpps is.

kilo packets per second, IIRR

> On more than one occasion I have seen cases where "something bad"
> happens to one of the two NIC's connected by a cable where one
> NIC goes into single "duplex" at 10Mbs; which degrades throughput
> below 100Mbs.

this is often a problem on some NIC chipsets.

use "mii-tools" and force to 100baseTx-FD for full duplex
or 100baseTx-HD half duplex, if you see transmission errors...

> Your original "conclusion" that the problem must be in the kernel
> may be in error; then again it may not. But without checking other
> alternatives, you won't know for sure.

also depends on the chipset of the NICs you are using.

which kernel and which NICs are you using?

cheers,

Kim

Kernel 2.4.27 and 2 nics 3c905B 100BaseTX with 3c59x modules.

I tried to force 100Mbps but here is what I've got:

# mii-diag -F 100baseTx-FD eth0
Setting the speed to "fixed", Control register 2100.
Basic registers of MII PHY #0: c000 c000 c000 c000 c000 c000 c000
c000.
The autonegotiated capability is 0000.
No common media type was autonegotiated!
This is extremely unusual and typically indicates a configuration
error.
Perhaps the advertised capability set was intentionally limited.
Basic mode control register 0xc000: Auto-negotiation disabled, with
Speed fixed at 10 mbps, half-duplex.
Transceiver in loopback mode!
Transceiver currently being reset!
Basic mode status register 0xc000 ... c000.
Link status: not established.
Your link partner advertised c000:.
End of basic transceiver informaion.

"Nico" <nicolas.patik@gmail.com> wrote in
news:1102506877.505049.90290@c13g2000cwb.googlegro ups.com:

> Kernel 2.4.27 and 2 nics 3c905B 100BaseTX with 3c59x modules.
>
> I tried to force 100Mbps but here is what I've got:
>
> # mii-diag -F 100baseTx-FD eth0
> Setting the speed to "fixed", Control register 2100.
> Basic registers of MII PHY #0: c000 c000 c000 c000 c000 c000 c000
> c000.
> The autonegotiated capability is 0000.
> No common media type was autonegotiated!
> This is extremely unusual and typically indicates a configuration
> error.
> Perhaps the advertised capability set was intentionally limited.
> Basic mode control register 0xc000: Auto-negotiation disabled, with
> Speed fixed at 10 mbps, half-duplex.
> Transceiver in loopback mode!
> Transceiver currently being reset!
> Basic mode status register 0xc000 ... c000.
> Link status: not established.
> Your link partner advertised c000:.
> End of basic transceiver informaion.
>
>

Speed fixed at 10 mbps, half-duplex; says it all.

txs