2901 Router Throughput Increase

[Brandon James]

I have a point to point connection with an off-site location that we recently upgraded our bandwidth from 100 to 500 Mbps. I had a 2901s on both sides and discovered that 167 Mbps was the max throughput we could achieve.

I recently migrated the routing to a 3750G switch, on the local side, with plans to replace the remote side for a 3750G also. My thoughts for this is that I can take advantage of the hardware routing and overcome the 167 Mbps hurdle.

Question:

Access to the remote site is via coordination only and my schedule changes a lot....   I wanted to ask the group if there might be a way to get beyond 167 Mbps with the 2901s, such as CEF/ASIC? Something I can configure remotely and perhaps prevent my travel to the off-site for hardware replacement.

The only thing I am doing special over the link is EIGRP and potentially migrating to OSPF to accommodate a SonicWALL firewall. I have a 6 Mbps backup link that also terminates at the off-site datacenter.

Any thoughts or advice will be greatly appreciated.

Thanks,

Brandon

[Leo Laohoo]

I had a 2901s on both sides and discovered that 167 Mbps was the max throughput we could achieve.

The CPU can't do more than that.  Look at this way:  If this tiny router can achieve 500 Mbps (and more), what is to stop everyone from buying the bigger models?

3945 or the 3945E can achieve 500 Mbps.  The newer 4431 (Data Sheet, Table 4) is designed for >500 Mbps.

[Brandon James]

Leo, thanks for the response. I had considered the 4431 but from a budget standpoint decided to go with Layer 3 switching to achieve the speed as I didn't need all the other services such as NAT, advanced QoS, and what ever else might be required to run in software.

I was really hoping to find out if there was a way or trick to get the port on the 2901 to process in hardware.

Thanks,

Brandon

[Joseph W. Doherty · Accepted solution]

I had a 2901s on both sides and discovered that 167 Mbps was the max throughput we could achieve.

Discovered by your own testing or going by Cisco's (old - 2009) performance paper?

Cisco, in a 2010 white paper, documents the 2901 as capable of forwarding 3,114 Mbps (yes, that's 3 Gbps).  This is the device's best possible performance, i.e. just forwarding max size Ethernet packets.

Conversely, in that same paper, Cisco recommends the 2901 for only 25 Mbps of WAN (i.e. duplex) bandwidth.  The latter is a conservative recommendation, i.e. you're likely to almost always be able to provide that much performance regardless of your traffic or configuration.

What this tells us, a 2901's performance is very variable - much depends on your config and the nature of your traffic.

To obtain the maximum forwarding performance out of a ISR you'll want to use the minimal possible configuration.

For example, if you're using a dynamic protocol like EIGRP, replacing it with static routes would save a few CPU cycles, that can be used for forwarding packets.  (Oh, and the fewer routes, the better to.  For example, using a default for all off-site routes.)

Not using any ACLs, also saves CPU cycles.

Deactivating CDP should save a few CPU cycles.

Deactivating NTP should save a few CPU cycles.

Etc., etc., etc.

However, before you go down the road trying to saving a microsecond of CPU here or there, yes, you'll want to insure items like CEF are enabled and/or your buffer management is optimal, i.e. all hits without buffer list trims/creates, etc.  And/or, you might want to minimize hitting up this router with SNMP queries, etc., etc.

Again, the closer the config is to only doing basic packet forwarding, you'll maximize the ISR's performance.

Not withstanding the foregoing, for 500 Mbps link, you'll want a more capable device.

[Brandon James]

Joseph, The 167 came from the cisco documentation. I was seeing sustainable speeds of around 105 with two 2901s. When I replaced one side with the 3750G, I noticed a push closer to 120 Mbps. These are sustainable speeds, I have noticed spikes up closer to 200 Mbps.

I appreciate your insight on the CPU cycles, I will go back and reduce my SNMP monitoring and try out static routes to see what kind of bump I get. In the end I will replace with a switch, I just want to squeeze out as much bandwidth as possible prior to the swap.

As far as buffer management, do you have a link to any documents or articles that you would recommend as a good read? 

Thanks,

Brandon

[Joseph W. Doherty]

I appreciate your insight on the CPU cycles, I will go back and reduce my SNMP monitoring and try out static routes to see what kind of bump I get. In the end I will replace with a switch, I just want to squeeze out as much bandwidth as possible prior to the swap.

Yea, although keep in mind most ISRs don't devote much of their CPU to other stuff, but every CPU not used for anything not absolutely required, is CPU that can be used for packet forwarding.

Your mileage may vary.

As far as buffer management, do you have a link to any documents or articles that you would recommend as a good read? 

Yes, there's at least one paper on tuning buffers (http://www.cisco.com/c/en/us/support/docs/routers/10000-series-routers/15091-buffertuning.html), there may be more.  However, more importantly, the later IOSs support an "autotune" global setting, buffers tune automatic, which you might try.