Why would you want to install this many ethernet cards? Would it not be
easier to install some cards that used 802.1q trunking and just use one
interface. Then you could take an ethernet switch and divide it up to
segment the network as so. I would think that this would be faster and
cleaner than dozens of interfaces. So now I must ask, does LRP support
802.1q on the network card if the nic supports it also. How would you set
that up in LRP if it does?
-----Original Message-----
From: Matti Aarnio [mailto:matti.aarnio@sonera.fi]
Sent: Friday, April 14, 2000 1:20 AM
To: kiran keshava
Cc: Linux Router Question Forum; linux-net@vger.rutgers.edu;
linux-router@linuxrouter.org
Subject: [LRP] Re: Multiport Ethernet Cards(Plenty of em...)
On Fri, Apr 14, 2000 at 02:44:17AM -0400, kiran keshava wrote:
> Hi all,
>
> I heard a few days back that someone is Installing Many many NICs on a
> linux machine. I am also working on benchmarking a Linux Box as a router
> and am planing on installing six 4-port 10/100Mbps Ethernet cards (total
> 24 ports) on it. I was wondering which card would be a good pick,
> presently I'm thinkin of using Matrox NS-FNIC/4 multiport 10/100Mbps
> ethernet cards. This card is rather new and seems to have a problem with
> the drivers. Is there any other 4-port ethernet card that anyone else has
> used and would recommend for usage at 100Mbps ???
Propably better approach is to use an external switch with VLAN
trunk capability (of IEEE 802.1q kind), and VLAN trunk driver
at Linux. ( This in general sense, benchmarking is other story
with 'performance at any cost' kind of thinking.. )
Each load port on switches must be configured with its own VLAN
number (from 2 to 4094), and corresponding VLAN numbers must
be enabled at Linux driver, too.
E.g. have some (single!) GigaEther card at Linux, and cisco
Catalyst 2980G-EN switch with GBIC to Linux card. That way
you have fast-ethers at load side, and each port can be
multiplexed to the trunk. If you need more ports, just
add (chaining or in a tree) more switches on that core switch,
perhaps have e.g 3508 as core switch and hook 2980Gs on it.
(Or if you want to go head-to-head with "big iron" routers,
there are cisco 12000 series monsters with gigaethers, and
5500/6500/8500 series big switches with VLAN capability...)
Single 2980G gives you 80 load ports, hooking them on a 3508
gives you 7 x 80 load ports: 560 ports...
Granted those switches cost your arm, leg, and first born,
but pick then something cheaper, e.g. load ports are 10BaseT
FDX, trunk ports are 2x 100BaseT-FDX (or possibly 2x MM Fiber)
You get easily 24 ports of 10BaseT, and 1 load-port of 100BaseT.
The other 100baseT is your VLAN trunk to your router.
A thing from somewhere in between previous two is using
2924XL-EN switches, or 3548XL-EN switches. Perhaps in form
of using two FE cards and hooking them as ether-channel to
the core switch. (Such *may* give you over 100 Mbit/sec
capacity, but it is never guaranteed with etherchannels.)
I can buy 2924XL-EN switches at port prices comparable as
1.5-2 times a good single port FE card price, however dual-
and quad-port FE cards are way more expensive per port, and
not so efficient. A good GE card is *very* good at doing
packet processing and (especially) interrupt mitigation, and
can thus drop your system IRQ load considerably while data
throughput increases dramatically.
For routing benchmarks.. load-ports are FE, trunk a GE, and
and you won't want any capacity overbooking.. Thus you can
route with such setup 10 load ports thru the trunk.
(This is presuming you can get them all saturated -- which is
likely. Benchmarking is the name of the game...)
To increase the trunk capacity, you might add second GE card
to your machine, and run them as separate routing domains,
not necessarily as Ether-Channel pair. Say first card gets
load ports 1 thru 10, second gets ports 11 thru 20.
To do this you pick 3524XL switch (or 3548XL if you want to
try cases where e.g. two ports belong to same VLAN, and you
allow trunk capacity saturation.) If you can get GE cards
with GBICs, you may consider Copper based short-haul (like
circa 0.5 meters) cheapish GBICs, otherwise you must go and
pick (likely) short-haul multimode fiber GBICs for the switch.
> Thanks in Advance
> Awaiting a reply
> Kiran
/Matti Aarnio <matti.aarnio@sonera.fi>
_______________________________________________
linux-router maillist - linux-router@linuxrouter.org
http://www.linuxrouter.org/mailman/listinfo/linux-router
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Apr 15 2000 - 21:00:24 EST