[nsp] Cisco 2500 Traffic Limit and high cpu utilization.
Mehmet Ali Suzen
msuzen at mail.north-cyprus.net
Mon Mar 8 03:55:22 EST 2004
Hi,
Thanks for the tips, indeed. I have already tried NetFlow before,
but when I enable cef, router begins to malfunction. I don't
know what is wrong? Are there any conflict with the other
switching mechanisms or service?
-Mehmet
On Sat, Mar 06, 2004 at 10:28:21AM -0600, Terry Baranski wrote:
> > Hi All,
> > I am wondering what is the traffic limit that a cisco
> > 2500 series routers can handle? And why do we get
> > quite often high cpu load and IP load? How can I
> > prevent this? I tried WRED queuing on the interfaces
> > it helps a bit but not sufficient.
>
> Depends on average packet size and what performance-reducing features
> are enabled. Make sure CEF switching is enabled and working, first and
> foremost:
> http://www.cisco.com/en/US/products/sw/iosswrel/ps1828/products_tech_not
> e09186a00801e1e46.shtml
>
> The 2500s are rated at 4400 packets-per-second (using CEF) with 64 byte
> packets, which comes to around 2.2Mbps. As packet size increases the
> achievable packets-per-second tends to decrease, but the maximum
> bandwidth one can push through a router tends to go up because the
> packets are bigger. A fairly limited number of tests with other router
> models show bandwidth increasing by around 3-4x as packet size goes from
> 64 bytes to 1500 bytes (Ethernet traffic). That would put a 2500 at
> around 6-8Mbps with 1500 byte packets, but note that this is simply an
> estimate based on the performance profiles of some other routers.
> Others here may have done tests on the 2500 series specifically that
> they can share.
>
> Note also that these numbers are best case; i.e., CEF enabled, large
> packets, no performance-reducing features enabled, and so forth.
> Average packet size is usually much lower than 1500 bytes on real
> networks. And anytime you enable a feature that increases the work the
> router has to do on each packet (e.g., ACLs, NAT, policy routing, etc),
> performance will often suffer.
>
> -Terry
>
More information about the cisco-nsp
mailing list