4

What is the multicast routing latency of Cisco C891F-K9 ISR? Please cite references for your answer. I think it's store-forward type and isn't one of the fastest models (given its price!).

I've read a datasheet from Netgear claiming its L2 VLAN switch has the latency of 4.5 us. If Cisco ISR does routing by CPU, should we expect it is much slower, (eg. 10--20 us?)

  • I'm going to take the liberty of editing this a bit... please roll back if this isn't what you want. – Mike Pennington Nov 7 '14 at 11:54
  • 5
    BTW, I'm strugging to understand why this even matters... 4.5 microseconds versus presumably 20 microseconds. Nobody should expect to do things like high frequency trading or real-time control on a small office product (i.e. what the Cisco 891 was designed for). Furthermore, you're comparing a layer-2 switch with the CPU latency of a layer-3 router. This question might benefit from explaining what you hope to do with this information. You might find that this is a flawed strategy – Mike Pennington Nov 7 '14 at 12:01
  • @MikePennington One reasons is, we want to know the effective bandwidth of routing multicast packets w/o dropping any packets. If we assume the router forwards packets one by one, the effective bandwidth should not limited only by the interface speed (eg. Fast Ethernet, Gigabit Ethernet, ...) but also by the latency of routing. 20 us => 50k packets/sec => 26 Mpbs for 64 byte packets. 4.5 us => 114 Mpbs – nodakai Nov 8 '14 at 9:06
  • @MikePennington Perhaps my reference to Netgear L2 switch was rather confusing? Let me rephrase the second paragraph as: "Netgear could indeed publicize what we wanted to know. Isn't it natural for us to search for the same for Cisco as well? (But we admit the datasheet we got from Netgear was for their L2 switch, not for their router)" – nodakai Nov 8 '14 at 9:19
1

What is the multicast routing latency of Cisco C891F-K9 ISR? Please cite references for your answer. I think it's store-forward type and isn't one of the fastest models (given its price!).

... from the comments: we want to know the effective bandwidth of routing multicast packets w/o dropping any packets.

What you're looking for is the RFC 2544 No Drop Rate (NDR), which is shown in Table 1 of Cisco's Portable Product Sheet, I've included a copy of it inline below...

Portable Product Sheet

In short, the fastest a Cisco 891 can forward packets without dropping them is 100Kpps, which translates to 10 microseconds of latency; however, realize this is the fastest you can forward traffic. The performance will be lower if you turn on interface-level features, such as netflow, CBWFQ, or security access-lists. Unfortunately you can't know how much lower without testing the performance yourself; you can use tools such as iperf3 for this kind of testing.

Usually it's a safe assumption to take 60% of the maximum NDR and use that number as your forwarding rate (with a lot of features); 60% of 100Kpps is 60Kpps, or 17 microseconds of latency. There should not be a meaningful difference between unicast and multicast forwarding rates.

... from the comments: the Cisco 891 seems to be limited to 20 us => 50k packets/sec => 26 Mpbs for 64 byte packets. 4.5 us => 114 Mpbs

This is where your assumptions are leading you sideways. A router's NDR is measured with 64-byte frames, but you should not use that to build the maximum possible bandwidth of the device, because no sane person builds services that inefficiently; we all use much larger frames than 64-bytes. You're the only person who can tell us what your average server's packets will be, but let's assume they are 300 bytes:

  • Cisco 891 @ 60Kpps (many features) and 300 byte IP packets: 144Mbps
  • Cisco 891 @ 100Kpps and 300 byte IP packets: 240Mbps

Final notes:

Keep in mind that the PPS forwarding rate of the router is a unidirectional measurement. Bidirectional traffic will consume part of those numbers... so if we use 60Kpps / 300-byte packets, and you require 90Mbps downstream, you should plan on no more than 54 Mbps of upstream traffic.

Also account for the reality that network engineers do not want to operate routers at 100% CPU all the time. Usually we start looking for upgrades around 70 or 80% CPU.

So... all that said, your best approach to finding out whether a Cisco 891 is enough is to do some testing with the real services and features that you're going to pass in production. If your IT staff won't turn on many features, plan on 100Mbps bi-directionally. If your IT staff needs to turn on a lot of features, you really should test the router with the combination of features you think you'll need.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.