kuznet@ms2.inr.ac.ru wrote:
> Hello!
>
> > be wrong or does the 2514 not have enough horsepower?
>
> The test is invalid, unfortunately.
>
> You could dig out some 20MHz i386, to compare it to 25xx,
> and Linux still wins. 25xx cannot forward between 10Mbit ethernets,
> it was not designed to make this. It is good as serial line switch.
> [ Essentially, 2514 shipped with 2 ethernet ports is cheating
> of customers. ]
>
> Your Pentium boxes can be compared only to 7200 on 100Mbit interfaces.
> And Cisco will win in this case, if you do not use fastrouting
> drivers with Linux.
>
> Alexey Kuznetsov
I think everyone may be missing the point. The 2514 data is in no means a
comparison. The 2514 was designed for communications between ethernet and
WAN links and for that it is a very good solution. Including the data the
way I did may have been misleading. Assuming the data posted by Jason is
correct a pentium will beat a 7000 series router in PPS without fast
switching on the Linux box or the Cisco. The 7200 can pass 10k packets
and the 400Mhz Pentium can pass 70000 packets and at what cost savings.
Even if we were to discuss fast switching it would quickly become a
useless discussion if the amount of addresses rises to big for the
fast-switch tables in the Cisco. The entire discussion of pps becomes
useless when you realize a router is for changes in encapsulation or
policy. If you are connecting networks of similiar media you switch you
do not route anyways. The 2514 data was included for people like myself.
I was not aware until I did this testing what acceptable packet loss was.
Haveing such a widely deployed router tested in this manner gives me
confidence in a Linux based power PC/intel x86 router and plus I enjoy
using Linux. The comment about my tests being invalid make me cringe. I
made no conclusions based on the data. I presentted it and that is all.
To know that the 2514 has a 20Mhz processor and I used 166Mhz Pentium and
400Mhz PII should lead most network techs/admins to believe a comparison
would not be rational. I do beleive the data posted by Jason is far more
telling of the superior pps ability of a Linux PC router over what would
appear to be most of Cisco's product line.
The Smartbits tester had I think 7710 card in them, they could do 100/10
HD and FD. I do not have the config for the Smart Apps. In most test I
just used the two interfaces on the smart bits connected directly to the
21040 cogent NIC cards. I then selected those ports in Smart Apps and
filled in the NIC's MAC address they would be talking too and gave them
addresses on the same subnet as the NIC's they were conected to. I then
echoed "1" > /proc/sys/net/ipv4/ip_forward to start forwarding packets.
There were'nt any routing daemons running and I can send the boot/root
images to anyone who wants them.
The only thing I think stopping Linux and PC technology from dominateing
the networking (routing) world is support. Enterprises are very shy of
using Linux corporate wide because they do have the infrastructure to
support it themeselves. With Compact PCI, Hot Swappable cards, ATM OC-12
and OC-3 interfaces there has to be a company based on this stuff sooner
or later.
Anyways I hope I do not sound like I am being mean or that I am mad, I
love doing this stuff and am haveing fun. If I am wrong or have draw any
of the wrong conclusions I enjoy discussing this stuff and getting my
head on a little straighter.
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.rutgers.edu
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:26 EST