Howard C. Berkowitz wrote: > > At 7:29 PM +0000 1/31/03, Chuck Church wrote: > >I got into this discussion kind of late, but here's my take: > > > >Functionally, you can configure either to do what you want. > But a 1 armed > >router has a couple major limitations that a layer 3 switch > doesn't. A > >layer 3 switch has ASICs (application specific integrated > chip/circuit) that > >can perform MAC re-writes, RIB/FIB lookups, rate-limiting, > QOS, and ACL at > >wire speed without bothering the CPU of the device. > > As you point out, a little indirectly with the footnote about > the VIP > in the table below, so do the 7500 and up. Even the 7000 can > have > separate routing and switching processors. > > Above the 7500, there's extensive use of ASICs and distributed > forwarding processors. > > MAC rewrites are normal functions in commercial L2 chips, so > that is an issue. > > The question really comes in the more sophisticated QoS and > routing > functions, which, after all, tend to be more needed in the WAN. > > > A 1 armed router needs > >to use the CPU for some of these functions, and will quickly > become a > >bottleneck after a certain level of traffic is passing through. > > Agreed, if it is a single CPU router. ASICs even appear on > some > lower-end devices for things like encryption.
I think the real-world comparison that the original poster probably had in mind was a rather low-end, inexpensive one-armed router compared to a L3 switch. It sounds like we all agree that in general the L3 switch is going to have better performance in that case. > > >Also, a 1 > >armed router is limited by it's 1 arm :) That link will be > limited to 100 > >mb/sec (unless you move up to a 72xx or higher router, where > gig is > >possible). > > True. That is a good point. Another option, by the way, is a 2-armed router. Seriously, instead of doing inter-VLAN routing on one interface on the router, why not just add an Ethernet interface to the router? Duh. :-) This would be a solution somewhere in the middle with regards to cost and performance. Well, it's been a great discussion! Thanks everyone. Priscilla > > > So for instance if you're copying a large file between VLANs, > >it'd be pretty easy to use up all the bandwidth of that 100 > mbit full duplex > >link, even if the CPU wasn't working hard on the 1 armed > router. > > Looking at the broader picture, it isn't necessarily > route-versus-switch. A heavily used server can have multiple > NICs in > multiple VLANs, with full speed on each (including GE). A L2 > switch > can handle intra-VLAN switching. > > > Moving to > >a layer 3 switch typically bumps that layer 3 device to layer > 2 backplane a > >multi-gigabit speed connection. So if your traffic between > vlans will ever > >exceed 100 mbit, you can either shell out huge bucks for a > 72xx, or get a > >real QOS-friendly 3550 that is both faster and cheaper. Of > course if you > >need WAN modules in the device that's another story. I was > sent this chart > >a while ago listing speeds of various routers and switches: > > > >> Router Performance Specs > >> > >> Router Switching Performance - Performance based on 64 Byte > packets > > Obviously, there's a tremendous difference based on which > switching > path is used. It can be very feature-, release-, and > platform-dependent if enabling a given feature drops you out of > CEF, > fast switching, etc. > > It's been my experience this is more likely to happen in an > L3 > switch or low-end router. This isn't necessarily bad design. > Many > of these features are more critical in WANs than LANs, and > switches > are, reasonably enough, optimized for LANs. > > > > > >> Platform Process Fast Fast > >> Switching Switching Switching > >> (PPS) (Mb/S) > > I suspect >>> is bps > > > > ------------------------------------------------------- > >> 1400 600 4,000 2,048,000 > >> 1600 600 4,000 2,048,000 > >> 1700 1,500 8,400 4,300,800 > >> 2500 800 4,400 2,252,800 > >> 261X 1,500 15,000 7,680,000 > >> 262X 1,500 25,000 12,800,000 > >> 265X 2,000 37,000 18,944,000 > >> 3620 2,000 40,000 20,480,000 > >> 3640 4,000 80,000 40,960,000 > >> 3660 12,000 120,000 61,440,000 > >> MC3810 2,000 10,000 5,120,000 > >> 4000 1,800 14,000 7,168,000 > >> 4500 5,000 40,000 20,480,000 > >> 4700 7,000 50,000 25,600,000 > >> 7120 13,000 175,000 89,600,000 > > > 7140 20,000 300,000 153,600,000 > >> 7200-NPE100 7,000 100,000 51,200,000 > >> 7200-NPE150 10,000 150,000 76,800,000 > >> 7200-NPE175 9,000 175,000 89,600,000 > >> 7200-NPE200 13,000 200,000 102,400,000 > >> 7200-NPE225 13,000 225,000 115,200,000 > >> 7200-NPE300 20,000 300,000 153,600,000 > >> 7200-NPE400 20,000 400,000 204,800,000 > >> 7200-NSE-1 20,000 300,000 153,600,000 > >> uBR-NPE150 10,000 100,000 51,200,000 > >> uBR-NPE200 13,000 150,000 76,800,000 > >> 7000-RP 2,500 30,000 15,360,000 > >> 7500-RSP2 5,000 220,000 112,640,000 > >> 7500-RSP4 8,000 345,000 176,640,000 > >> 7500-RSP8 22,000 470,000 240,640,000 > >> Cat 2948G-L3 N/A 10,000,000 5,120,000,000 > >> Cat 4908G-L3 N/A 12,000,000 6,144,000,000 > >> Cat 4232-L3 N/A 6,000,000 3,072,000,000 > >> Cat -RSM 14,000 175,000 89,600,000 > >> Catalyst-RSFC 170,000 87,040,000 > >> Catalyst-RSFC/NFFCII 2,000,000 1,024,000,000 > >> Catalyst-MSFC (IP,IPX) 15,000,000 7,680,000,000 > >> Catalyst-MSFC (Other) 170,000 87,040,000 > >> Catalyst-MSFC2 (IP,IPX) 15,000,000 7,680,000,000 > >> Catalyst-MSFC2 (Other) 680,000 348,160,000 > >> Catalyst-MSFC (X-bar) 30,000,000 15,360,000,000 > >> > >> NOTE: VIP2 Distributed Switching significantly increases > > > the performance on RSP platforms. > > Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=62292&t=62273 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
