[TriLUG] OT - gigabit switches

Ryan Leathers rleathers at americanri.com
Wed Sep 20 17:22:05 EDT 2006


To add a bit to Greg's comments... 
(the term "L3 switching" used to bug me too) 

The clear distinction between L2 and L3 forwarding decisions is that L2
decisions are based on L2 information (example MAC address) while L3
decisions are based on L3 information (example IP address)

A distinction can also be made between switching and routing.  This is
not nearly so clear a distinction.  What it ultimately comes down to is
speed through efficiency in making the forwarding decision.  Let me
explain.

Here is the traditional L3 (routing) process, rather simplified.  The
router reads the payload of each L2 frame for which its own L2 address
is a match. It passes the L3 packet header data to the active routing
process which then compares the L3 destination for the best match in a
routing table.  This result is used to create a new L2 frame to
encapsulate the L3 Packet and drop the frame on the correct egress
buffer.  Each of these steps requires processor time and suffers the
pains of interrupt and slice availability along the way which makes
traditional routing slow compared to switching.

Traditional L2 (switching) is more streamlined since there is just one
table lookup resulting in the arriving frame being dropped on the
correct egress interface without any additional mangling.  This is
comparatively fast.

So there is a distinction to be made about the process by which a
forwarding decision is arrived at and executed.  So we don't get mired
in the details of particular vendors or products I will risk some
generalization here.  The switching process likely involves a single
interrupt, lookup, and execution.  The routing process involves all of
the switching steps plus at least one additional table lookup and
decision, followed by a new frame encapsulation.  The routing process
takes longer not expressly because of the L3 data involved, but because
of the process steps employed in traditional methods of arriving at the
L3 decision. This distinction between switching and routing processes is
completely separate from that of the contrast between L2 and L3 data.

Given an understanding of the distinctions between both process and
data, one can see that applying a switching process to the task of
making a L3 forwarding decision could reap performance benefits.  This
is what L3 switching is in a nutshell.  In short, if a way can be
devised to use L3 data in order to populate the table used by the
switching process, then it is possible to achieve the performance
benefits inherent in the process while retaining the value of decisions
based on L3 hierarchical addressing.  To go deeper into how this is done
we really need to talk about specific implementations since there is
more than one way to peel this onion, but this is the gist of it.   L3
switches are applying a switching process to make decisions via table
lookup against tables which have the benefit of being created with
knowledge of L3.  It is still a switching process no matter how much the
result smells like routing.  

>From a practical standpoint, to me, unless we're in a seriously deep
network design discussion, making the distinction between routing and L3
switching is splitting hairs.
  


On Wed, 2006-09-20 at 14:43 -0400, Greg Brown wrote:
> Good questions!  Here is a basic run-down of Layers 1, 2, and 3 from a
> network standpoint (where I spend most of my time).
> 
> Layer 1 refers to the physical wiring or lack there of.  Fiber optic and
> Cat6 copper cables are most popular for Gig speeds though Cat5e is
> technically rated for Gig though I have never used it for such a purpose.
> 
> Layer 2 regards information at the MAC address layer and it is called the
> "Data Link Layer".  Switching occurs at Layer 2.  Therefore a layer2 switch
> is a device that uses the unique MAC address that is burned into every card
> to determine if a packet should or should not be sent to a particular port.
> Broadcasts and multicasts are sent to every port within a single collision
> domain (layer 2 switch or groups of switches).
> 
> Layer 3 refers to network protocol information and is called the "network
> layer".  Routing occurs at Layer 3.  Packets are sent to and from ports
> based on things like IP Address ranges, Appletalk Zones, and whatever other
> kinds of protocols in the mix.  Thankfully today it is almost all TCP/IP
> which makes the world a much more simple place to play.
> 
> So what is "Layer 3 switching"?  It is a term that makes my skin crawl.  It
> used to be said, back in the days of the Cisco AGS and other old routers,
> that routers were slow and you didn't want them in the core of the network
> (at least some people said that).  Routers got faster and they wanted a term
> to reflect this so someone put forth "Layer 3 switching" and it suck.  Layer
> 3 is routing, not switching.  The closest pure thing to L3 switching was
> Virtual Link State Protocol buried in Cabletron gear and nobody uses that
> anymore.  Whenever you hear "layer 3 switch" think router.
> 
> I don't have much experience with Dell gig switch gear as I mostly use Cisco
> at my present employer.  We have 48 port POE Gig blades in our 65xx big guns
> and a few gig blades in our 45xx devices.  In addition we have 48 port gig
> switches with 10 gig uplinks.  They are hot stuff and we use them to
> interconnect some or our blade servers.
> 
> You mentioned that some of your gear will have dual ports and that is a
> whole other ball of wax.  We have many servers in the data center with a
> primary gig port and a 10/100 control network port on different ranges.  We
> also have F5 load balancers with multiple gig connections, etc.
> 
> Gig is sweet.  Deploy wisely and upgrade the switch-to-server cables to Cat6
> if possible.  Some of the more inexpensive gig switches give you a lot of
> bang for the buck but avoid the commodity stuff as even though they say they
> are gig they have little more then 250 meg of throughput.
> 
> Greg
> 
> 
> On 9/20/06, Christopher L Merrill <chris at webperformance.com> wrote:
> >
> > We would like to upgrade our testlab to a gigabit switch.  I'm not an
> > expert in this area (I spend most of my time at level 7). Maybe some of
> > you network gurus can set me straight.
> >
> > In our lab we'll have 10-25 machines (some with dual NICs) mostly fast
> > ethernet but we're moving machines to GigE. We currently have a Dell 2324
> > with 2 GigE ports.  Performance between the machines connected directly
> > to the switch is the only real goal.  Management functions are not
> > important, beyond the desire for simple plug-n-play management that we
> > have today (in other words...no management - just plug them in).  Note
> > that we do load-testing in the lab, so for a network with only a handful
> > of machines, they run a LOT of traffic between them - frequently to the
> > CPU/NIC limit of each machine.
> >
> > I've read a little on layer-2 and layer-3 switches and think I grasp the
> > differences, but it is not clear to me if there would be any performance
> > benefit for US with a layer-3 switch.  Budget is obviously a concern so
> > a $2000 switch will take some selling.  Under what kind of situations
> > would
> > we see a difference in the total throughput rates on, for example, a Dell
> > 2724 (layer 2, <$300), a Dell 5324 (layer 2/3, $750) and a 6024 (layer 3,
> > $2000)???  What about an older Dell 5224 (layer2/3, ebay $300)?  What does
> > "layer 2/3" mean?
> >
> > Any recommendations or pointers for some good reading would be much
> > appreciated.
> >
> > p.s. We have lots of Linux boxes in the lab  :>
> >
> > TIA,
> > C
> >
> >
> >
> > --
> > ------------------------------------------------------------------------ -
> > Chris Merrill                           |  Web Performance, Inc.
> > chris at webperformance.com                |  http://webperformance.com
> > 919-433-1762                            |  919-845-7601
> >
> > Website Load Testing and Stress Testing Software & Services
> > ------------------------------------------------------------------------ -
> > --
> > TriLUG mailing list        : http://www.trilug.org/mailman/listinfo/trilug
> > TriLUG Organizational FAQ  : http://trilug.org/faq/
> > TriLUG Member Services FAQ : http://members.trilug.org/services_faq/
> >




More information about the TriLUG mailing list