At 12:39 AM 6/7/01, Michael L. Williams wrote:
>Chuck...... I talked to a good friend of mine that knows more of this than
>I do...... and even HE wasn't clear on the line of switch -vs- router.....
>so my following comments are strickly my thinking out loud...... mostly to
>get feedback .....
>
>I don't think there is a difference in a layer 3 switch that does 100,000
>packets/sec -vs- a router that does 100,000 packets/sec. However,
>respectfully submit that I don't know of any routers that can keep up with
>Layer 3 switches because in multilayer switching the route processor only
>has to route the first packet in a flow.
Yes, but..... Isn't that true for a router with fast switching, silicon
switching, optimum switching, distributed switching, etc? And with NetFlow
switching, a router can take into account access control lists and QoS
features that need to be applied to a flow, much in the same way that these
new MLS switches can do this with their flow masks. (Thanks for your
earlier message that explained flow masks so well.)
I agree that MLS has great potential. It sounds complicated to configure
and hard to troubleshoot, though. I think I would keep a hub and a protocol
analyzer handy when first implementing it, so I could check traffic between
the MLS-RP and MLS-SE when things went wrong.
Regarding packets-per-second, we need to remember that this is a marketing
game. The enormous numbers come from the absolute maximum possible number
of packets on a Gigabit Ethernet, taking into account the inter-frame gap
and preamble. The test engineers pump frames of the smallest possible size
into the switch to make the numbers really look big. (I wonder if they take
into account the carrier extension with Gigabit Ethernet? That would make
the numbers less.) The PPS is based on this max number of packets coming in
one port and going out another Gigabit Ethernet port. They can increase the
numbers even more by using multiple ingress and egress ports and no
contention for an egress port.
The numbers for both switches and some routers are so astronomically high
these days that they stretch credibility. Do real-world traffic generators
really send that much traffic?
Regarding CPUs, the general-purpose CPU on the Cisco routers may not be
very fast, but the high-end routers also have Versatile Interface
Processors that help with high-speed switching.
One last point is that routers have features that switches don't have
today. We configure access lists on routers. (Although an MLS-SE can make
use of these access control lists, we still configure them on the router).
Routers run running protocols. Routers act as firewalls, policy servers,
handle RSVP and other QoS requests, connect telephones, act as DHCP
servers, connect modems both analog and cable, etc.
Just a few thoughts before this interesting discussion undergoes a
well-deserved retirement.
Priscilla
> At that point everything is
>switched as fast as the switch can switch them (tounge twister), and even
>the fastest routers would have trouble routing as speeds that switches can
>switch at. IMHO, any switch-router that can do 100,000 packets/sec can't
>reach it full potential unless Multilayer switching is enabled and
>configured. Further, I submit that there isn't a router that can handle
>the throughput of a switch-router utilizing Multilayer switching. With
>switch backplanes reaching 32 to 256Gbps, I don't know of a pure layer 3
>router than can keep up with this pace without using MLS.
>
>I disagree with the assertion that ASICs and processor are equal because
>they're both chips. Obviously we could compare the P4 chip to the 486 chip
>and see that not all chips are equal. And I realize that's not what you are
>claiming. =)
>
>However, there is a distict difference between a general purpose CPU and an
>ASIC and how fast they can do a specific job. General purpose CPUs have
>instruction sets and need to be programmed with code. ASICs are indeed
>chips, but they are chips whose entire circuitry is designed and dedicated
>to do one thing and one thing only. This give the ASIC a definite advantage
>over any general purpose CPU on a give task. Yes, I hate to make blanket
>statements because there are things like the speed of the respective CPUs,
>etc. But it is generally accepted that an ASIC performs a given task much
>faster than a general purpose CPU. That's why 3D graphics cards use
>specific processors (ASICs) instead of just another P3 on the video card,
>etc......
>
>Utilizing ASICs allows things like checking for MLS cache entries and
>rewriting Layer 2 info, etc "at wirespeed". I can't say 100%, but I believe
>this means that it can output a "routed" packet (assuming it's in the MLS
>cache) in a single "cycle". I know know if that's the correct terminology,
>but my point being that an ASIC is able to do this type of thing. A general
>purpose CPU running IOS code, etc would take many clock cycles to perform
>the same type operations and it's speed can't compare to an ASIC.
>
>I have to say though, that I'm not all that familiar with the top end
>routers (switch-routers, whatever) like the 12000 GSRs (or the new 10Gig
>SRs), etc..... so perhaps they have some ASICs to perform their
>routing...... dunno.........
>
>My 2 cents......
>
>Mike W.
>
>"Chuck Larrieu" wrote in message
>[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> > Very good, and thanks.
> >
> > But... to quote a sage, who made this point last time this topic came up,
> > what exactly is the difference between a router that routes 100,000
>packets
> > per second, and a layer three switch that switches 100,000 packets per
> > second?
> >
> > Cisco can talk about ASIC's versus processors all they want. Both are
>chips.
> > High end routers also have ASIC's as well as other means to optimize
>traffic
> > flow.
> >
> > Truth be told, layer 3 switch is a marketing concept, plain and simple.
> >
> > Chuck
> >
> > -----Original Message-----
> > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of
> > Michael L. Williams
> > Sent: Wednesday, June 06, 2001 3:56 PM
> > To: [EMAIL PROTECTED]
> > Subject: Re: Layer3 switch vs Router [7:7406]
> >
> > For the record: Layer 3 switch = Multilayer switch (I say this because
I
> > like to use the term Multilayer switch rather than Layer 3 switch...
dunno
> > why... I guess because in the switch/routers, you actually use and
>configure
> > Multilayer switching)
> >
> > In reply to some of the other posts on this topic:
> >
> > >This is actually covered in under the switching portion of the CCNP..
> > >Here is a link from Cisco for you to reference:
> > >http://www.cisco.com/warp/public/cc/so/neso/lnso/cpso/l3c85_wp.htm
> >
> > Layer 3 switching (Multilayer switching) *is* covered on the CCNP
>Switching
> > exam and I don't know how one could pass it without an understanding of
>how
> > it works. (although that link that was provided pointed to a page that
> > didn't explain MLS very well at all)
> >
> > > One differance is that a layer 3 switch does wire-speed switching
> >
> > AFAIK, *all* switches perform wire-speed switching, as long as the
>backplane
> > isn't oversubscribed (even then what gets switched is done at wirespeed
>and
> > other stuff is dropped). Can someone make sure I'm not fibbing or
>confused
> > on this?
> >
> > >Okay, two things -
> > >
> > >One - abduct a Cisco marketing rep, tie them to a chair, shine a bright,
> > hot
> > >light in their face and ask *them* what the difference is.
> > >
> > >Two - while they're tied up, ask them what "wire-speed" is supposed to
> > mean.
> >
> > Wire-speed simply means that the data is switched across the backplane to
> > the destination port ASAP, as fast as the wires can carry the data thru
>the
> > ASICs and to the destination port. I.E. the speed of light (minus a
small
> > fraction because the wires do actually have a non-zero resistance =)
> >
> > >Someone may have a product based answer for you but literally a router
is
>a
> > >layer 3 switch. Just think of all of the functionality that a switch
> > offers
> > >you and add on the route switch module to sweeten the pot.
> > >A router either bridges or has separate subnets on each of its
> > >interfaces.(simplistic answer of course).
> >
> > This is very misleading. Although a router has a switching process
within
> > it, and that switching process can take on many forms, a router is NOT
> > simply a Layer 3 switch. Although a router can bridge (including
bridging
> > VLANS using Integrated Routing and Bridging), even then it is not the
> > equivalent to a switch because of the way it performs the process (in
> > software on a CPU instead of with an ASIC). The router doesn't keep a
CAM
> > table like a switch, etc. and without something like a NetFlow Feature
>Card
> > or MLS processor, a router can't bridge (switch) at wirespeed like a true
> > Layer 2 switch...... Even on a switch/router with a NFFC or the like,
> > without Multilayer switching enabled, it's like a router with a ton of
> > ethernet (or whatever) interfaces. The packets are NOT routed and
>forwarded
> > out at wirespeed like they are (after the first packet) when Multilayer
> > switching is enabled and configured.
> >
> > I by no means am the expert on Multilayer switching, and I'm not trying
to
> > flame anyone for their answers. But there were alot of things being said
> > that didn't answer the original poster's question (actually the one post
>by
> > Bob Salazar was correct, but he didn't mention that you have to actually
> > turn on the Multilayer switching fuctions on the hardware he listed) or
>that
> > were misleading. I'm sure there are some things that I messed up or just
> > don't have a full understanding of (like "wire-speed hehe), so comments
>are
> > welcome.
> >
> > Mike W.
> >
> > "Denton, Jason" wrote in message
> > [EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> > > Can anyone tell me what the REAL difference is between a layer3 switch
>and
> > a
> > > router?
> > >
> > > Jason
________________________
Priscilla Oppenheimer
http://www.priscilla.com
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=7591&t=7406
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
