The pros and cons of Layer 2 routing assist
 
|
|||
 | |||
|
Sign up to receive this and other networking newsletters in your inbox.
In last week's issue, we discussed switching at various layers in the protocol stack - Layers 2, 3, 4 and above. This week, we'll focus on a specific type of switching that combines Layer 2 and 3 switching - which we at The Tolly Group call Layer 2 routing assist - and how this can impact your network.
Layer 2 routing assist forwards Layer 3 packets that would typically be routed as if they were Layer 2. If you have, say, an enterprise router such as a Cisco 7500 with a single Fast Ethernet link to three or four high-density switches that are aggregating your workgroups, Layer 2 routing assist monitors and learns traffic flows routed between subnetworks by the Cisco device. A switch can (if connected to both subnets) bypass the router and forward the packets directly without the router even seeing them. Thus, the router does all the brain work and the switches do the grunt work. This scenario also applies if you're using a routing function that is internal to the switch.
There are several pros to Layer 2 routing assist. First, the router can cool off and not route every single packet that must be routed. Performance is exponentially improved because the switches can do this "pseudo routing" at all ports (for example, 50 ports at 100M bit/sec have a potential 500M bit/sec of throughput). The router may have only one or two physical connections to the switch; if the router had to do all the routing, the maximum possible throughput would be maybe 100M bit/sec. Latency is reduced because the packet need only hit the switch and can avoid traversing the router.
But there are some cons to this method. Pseudo routing is not real routing. And the packet is forwarded at Layer 2, which means that little or no Layer 3 information is processed. Checksums are not checked, so bad packets can propagate the network. Also, IP time-to-live fields (and similar fields in other protocols) are not assessed, so the possibility exists that a packet that has reached its hop limit will not be "killed." In addition, there can be no Layer 3-based security, and other sophisticated routing functions would not be implemented.
However, most of the drawbacks won't have much of an impact on a well-designed network. Considering the performance gains, most can live with the loss of Layer 3 field checking, some of which is done on session initiation before the Layer 2 switches can learn the flow and take over. If you have a similar scenario that can benefit from this technology, make sure you discuss the trade-offs with your vendor.
RELATED LINKS
The many uses of the word "switch"
The newsletter referred to at left. High Speed LANs, 4/14/98.
Switching grows up
Detailed look at where switching is heading. Network World, 5/4/98.
Layers, layers and more layers
Mary Petrosky uncovers the mysteries of switching and vendor hype in her weekly column. Network World, 11/3/97.
Psst, hey buddy, wanna buy a network switch?
Everybody's selling switches today, or so they say. Network World, 8/25/97.
Foundry Networks FastIron Backbone Switch Layer 2 Performance and Router Assist Test Summary
From The Tolly Group.
OSI Reference Model
Defines each of the seven layers of switching.
Back to the High Speed LANs archive
Get a free subscription to the High-Speed LANs and other newsletters
