I have read IOS documentation at CCO, it seems to me that Cisco do
encourage using "ip route-cache" in general. The document mentions a
terms "fast switching" to explain the "ip route-cache", which I don't
quite understand. Would someone please explain a bit details about how
exactly the "ip route-cache" behave? How about "ip route-cache same
interface"?
Thanks.
-raymond
It's on by default, and normally what you want.
>I have read IOS documentation at CCO, it seems to me that Cisco do
>encourage using "ip route-cache" in general. The document mentions a
>terms "fast switching" to explain the "ip route-cache", which I don't
>quite understand. Would someone please explain a bit details about how
>exactly the "ip route-cache" behave? How about "ip route-cache same
>interface"?
Fast switching means that the router maintains a high-speed cache of recent
destination addresses and the interfaces/next-hops they were sent to. This
means that it doesn't have to scan the routing table for every packet.
There are two potential problems with having the route cache enabled. The
first is that it takes up memory, which reduces the memory available for
packet buffers; we've seen performance problems on routers with very large
route caches. The second issue comes up if you have multiple paths to a
destination; the route cache will pick one interface for each destination
and stick with it, so you won't necessarily get even load balancing.
"ip route-cache same-interface" is useful if you're using secondary
addresses to implement multiple logical networks on the same segment. It
implements high-speed forwarding when a packet comes in an interface and
it's destined for an address that's reached through that same interface.
E.g. if you have
interface ethernet0
ip address 192.168.10.1 255.255.255.0
ip address 192.168.20.1 255.255.255.0 secondary
If a packet from 192.168.10.9 comes in, and its destination is
192.168.20.15, enabling "ip route-cache same-interface" will forward it
more quickly.
--
Barry Margolin, bar...@bbnplanet.com
GTE Internetworking, Powered by BBN, Burlington, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.
> There are two potential problems with having the route cache enabled. The
> first is that it takes up memory, which reduces the memory available for
> packet buffers; we've seen performance problems on routers with very large
> route caches. The second issue comes up if you have multiple paths to a
> destination; the route cache will pick one interface for each destination
> and stick with it, so you won't necessarily get even load balancing.
It's probably worth adding that if you have multiple paths to the same
logical destination (say Sprint's customer routes) via multiple transit
networks, you almost certainly do want to use the route-cache, because
the alternative is per packet load-balancing, which will completely
screw TCP performance as alternate packets get interleaved and arrive
at the destination out of sequence because of the likely variance in
delay characteristics across the different paths.
Of course, if you want to load-balance over two symmetric paths with
the same delay characteristics (e.g. two back-to-back HSSI connections
between two routers you control), per-packet load-balancing is probably
more useful.
M.
-raymond
Barry Margolin wrote:
>
> In article <37BB06FD...@iname.com>, Raymond <out...@iname.com> wrote:
> >Would someone sharing your experience (opinion) on enabling "ip
> >route-cache" on ethernet interface and serial interface (with T1 frame
> >relay) with a Cisco 2500 series routers?
>
> It's on by default, and normally what you want.
>
> >I have read IOS documentation at CCO, it seems to me that Cisco do
> >encourage using "ip route-cache" in general. The document mentions a
> >terms "fast switching" to explain the "ip route-cache", which I don't
> >quite understand. Would someone please explain a bit details about how
> >exactly the "ip route-cache" behave? How about "ip route-cache same
> >interface"?
>
> Fast switching means that the router maintains a high-speed cache of recent
> destination addresses and the interfaces/next-hops they were sent to. This
> means that it doesn't have to scan the routing table for every packet.
>
> There are two potential problems with having the route cache enabled. The
> first is that it takes up memory, which reduces the memory available for
> packet buffers; we've seen performance problems on routers with very large
> route caches. The second issue comes up if you have multiple paths to a
> destination; the route cache will pick one interface for each destination
> and stick with it, so you won't necessarily get even load balancing.
>
You can see the cache with "show ip cache", but I don't recall if it shows
the memory use. I think I once figured out how much was used for each
cache entry by doing "show mem" and "show ip cache", waiting a few minutes,
and then doing them again. I calculated the difference in memory use and
cache sizes between the two times, and divided that to get an approximate
size of the cache. (Actually, since many route cache entries also have ARP
entries, it was the combined size of them -- then I did the same thing with
fast switching turned off to find out how much the ARP cache contributes).
Josh
jo...@cisco.com
Raymond wrote in message <37BB06FD...@iname.com>...
>Would someone sharing your experience (opinion) on enabling "ip
>route-cache" on ethernet interface and serial interface (with T1 frame
>relay) with a Cisco 2500 series routers?
>
>I have read IOS documentation at CCO, it seems to me that Cisco do
>encourage using "ip route-cache" in general. The document mentions a
>terms "fast switching" to explain the "ip route-cache", which I don't
>quite understand. Would someone please explain a bit details about how
>exactly the "ip route-cache" behave? How about "ip route-cache same
>interface"?
>
>Thanks.
>
>-raymond
first packet we process switch and then build an entry in cache table
and switch subsequent packets based on that cache entry. Fast switching is
only done on transit traffic through the router.
ip route-cache same-interface means that packet enters and leaves out the
same interface. e.g packet comes in e 0 and goes out e0. This usually happens
when you have multiple Ip subnets on the same ethernet int.
Hope this helps
Zaheer
>
> Thanks.
>
> -raymond
>
For me this isn't an issue as we're only using 2-4 B channels (ie 128-256K)
so CPU impact is a non-issue.
For others it might be.
Might have been an issue with that particular software on that particular
model (no really?) - but now you know.
Enjoy,
Phil P
-----------------------------------------------------------------------
People who say it cannot be done
should not interrupt those who are doing it.
-----------------------------------------------------------------------
Quokka Sports, Inc. Phil Pierotti
Ground Floor, Senior Network Engineer
525 Brannan St, phil.p...@quokka.com
San Francisco CA 94107 Ph: +1 415 977 2390
Fx: +1 415 908 1841
o_o
http://www.quokka.com/ / )o
(_/
I have wondered if this includes hub and spoke type frame infrastructures wherein
you have multiple sub-ints, instead of secondary addresses (i.e. would this be
useful configured on the main interface (Serial 0) for traffic between Network 1
on Serial 0.1 and Network 2 on Serial 0.2?
Barry Margolin wrote:
> "ip route-cache same-interface" is useful if you're using secondary
> addresses to implement multiple logical networks on the same segment. It
> implements high-speed forwarding when a packet comes in an interface and
> it's destined for an address that's reached through that same interface.
> E.g. if you have
>
> interface ethernet0
> ip address 192.168.10.1 255.255.255.0
> ip address 192.168.20.1 255.255.255.0 secondary
>
> If a packet from 192.168.10.9 comes in, and its destination is
> 192.168.20.15, enabling "ip route-cache same-interface" will forward it
> more quickly.
>
> --
> Barry Margolin, bar...@bbnplanet.com
> GTE Internetworking, Powered by BBN, Burlington, MA
> *** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
> Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.
--
Jason M. Czaplewski
Shands HealthCare
System Analyst
(352) 395-0680 x85775
On Mon, 23 Aug 1999, Phil Pierotti wrote:
> >From a machine connected to FA0/0 through the router, through an ISDN BRI
> and into my AS5200.
> The TAC recommended solution was "no ip route-cache" and "no ip
> mroute-cache".
>
> Might have been an issue with that particular software on that particular
> model (no really?) - but now you know.
Hmm, must have been a specific issue with your software rev or a combo of
the BRI interface and the image..
I have a 2621 doing:
Two FrameRelay T1s (via WIC-2T)
Full BGP feed on each T1, with prefix-lists
RIP & OSPF on LAN side
a NM-1E along with the 2 x 100bt
Bridging of two of the ports (long story, dont ask)
'ip route-cache flow' on all interfaces (not exporting it tho)
ACLs on most interfaces (in and out)
Traffic-shape via groups, used on 3 interfaces(FastE, both T1s)
Route map for HTTP traffic redirection on E1/0 to proxy server
xxx uptime is 7 weeks, 2 days, 16 hours, 34 minutes
Running:
IOS (tm) C2600 Software (C2600-IS-M), Version 12.0(4)T, RELEASE SOFTWARE
(fc1) Copyright (c) 1986-1999 by cisco Systems, Inc.
(yes, a T release!)
xxx>show ip cache flow
IP packet size distribution (2504M total packets):
[...]
With this setup, the router CPU avg is around %40..spikes to %70 happen
now and then..
I have been very happy with it!
cheers,
--
jason
-raymond
Does this suggest that 24KB is used for the cache?
-raymond
Barry Margolin wrote:
>
> In article <37BC5BD5...@iname.com>, Raymond <out...@iname.com> wrote:
> >Thanks for the detailed explanation about "fast switching/ip
> >route-cache". Is there a way to find out how much memory has been used
> >for caching route (ip route-cache)? Thanks.
>
> You can see the cache with "show ip cache", but I don't recall if it shows
> the memory use. I think I once figured out how much was used for each
> cache entry by doing "show mem" and "show ip cache", waiting a few minutes,
> and then doing them again. I calculated the difference in memory use and
> cache sizes between the two times, and divided that to get an approximate
> size of the cache. (Actually, since many route cache entries also have ARP
> entries, it was the combined size of them -- then I did the same thing with
> fast switching turned off to find out how much the ARP cache contributes).
>
Traffic going through the router, as opposed to traffic originated by the
router itself.