Hi,
This is my confusion. Anyone help me clear it.
Why should anyone buy ccr1036 ?
ccr1036…1000usd approx practical thru 2Mpps practically
Layer3 Switch for example brocade 6610 price approx 4000 usd thruput 100s of mpps
then what is the reason for layer 3 switch being so cheap ?
Why should we buy ccrs (I’d purchased quite a few all models from 1036-1009)
suggestion appreciated.
The L3 switch couldn’t ‘route’ 100’s of mpps. It’d switch-forward that much. Ultimately, on the L3 switches, once you turn on routing features you are then limited by CPU which is significantly smaller in switches than in routers, as switches mostly rely on the switch chips to do the forwarding.
CCR1036 can actually route its advertised rate.
Similar to Mikrotik CRS range - switching will be at full line rate but turn any port into a router port and start actually routing you are limited by the mipsbe 400/600mhz CPU, significantly less throughput.
This is only partially true. While the switches of the larger vendors do have smaller CPUs just like the CRS they also don’t rely on it nearly as much as Mikrotik. The larger vendors Cisco, Brocade and such have built hardware based L3 forwarding engines into their switches. In Cisco land its called Cisco Express Forwarding. If the L3 forwarding is able to be handled by CEF it is done at or near wirespeed. As an example on some older model (talking EOS and EOL here) Cisco switches if you hit IPv6 packets the hardware forwarding engine in the switch could not handle it so it routed via CPU. This meant that instead of routing at wire speed on the GigE switch it could only do about 80Mbps.
If you look at the Brocade website you will see three different speed designations switching and forwarding. Switching is the L2 switch speed in this case its the speed of the backplane. The forwarding speed is the L3 speed assuming it hits the ASICs instead of CPU. Cisco also lists the same kind of stats on their data sheets as well
http://www.brocade.com/products/all/switches/product-details/icx-6610-switch/specifications.page
http://www.cisco.com/c/en/us/products/collateral/switches/catalyst-3650-series-switches/data_sheet_c78-729449.html
The down side to these (granted they are LAN switches and not marketed as WAN/Carrier routers) is that their routing feature sets are not as comprehensive. In many cases they don’t support many carrier features such as VPLS, MPLS, etc… The buffers are typically smaller and the QOS options are much more limited than their router lines which typically operate a little more like a CCR. Even the larger “router” lines still implement the hardware L3 forwarding of the lower end lines they just add more memory for large routing tables and larger CPUs and such. That said most have HW offload for VPLS, GRE, and MPLS as well these days.
For use in a LAN environment for routing they typically work great.
Brocade specs says 6610 has forwarding capacity of 432Mpps.
Try loading a full-route BGP table on a L3 switch (aprox. 550k prefixes/routes)…
With the Brocade line mentioned it would be a issue. It only supports 16k routes.
If you move to something bigger it can certainly handle it with the same HW offload mentioned before. Right box for the right job. Just take a look at the Sup 2t on the Cisco side or something in the MLX line from brocade. Both can handle routing tables upwards of 1024k IPv4 routes.
The technical differences between a “L3 switch” and a “router” these days is not much and in many vendors lines there really is no difference.
For 1000 bucks as mentioned on the initial post?
No it isn’t and my intent was never to say it was. It was only to debunk that L3 switches are incapable at out routing a “router”.
The CRR and the Cisco 6500 don’t really live in the same performance or feature realm I would not expect them to be the same price. All the extra chips needed to do the HW offload come at a price both for development and pure parts cost.
That’s the point: While there are a bunch of surprisingly fast and surprisingly cheap L3 switches available, they usually offer only a very limited amount of maximum route entries - most of them can only handle around 16-64k routes, which is not enough to use them in any true multihomed BGP scenario as they cannot fit the full IP routing table.
Also, they have extremely limited featuresets, stuff like GRE tunnels (if available at all) are done in CPU, not in ASIC, so that’s where they become slow again. Some devices have support for hardware-processed ACLs, but those are “stupid” rulesets that do not use connection tracking/inspection and usually do not support NAT. If NAT works at all, it only does so in CPU/Software.
As many before me already said - choose the right tool for the right job.
A L3-Switch or Router that can handle 1 million routes in hardware costs several times as much as a CCR, but easily outperforms it on plain forwarding duties - not so much on anything else though that it has to process in CPU.