10-19-2010 02:49 AM - edited 03-06-2019 01:35 PM
Hi,
I have a 3750G being used as a router with the default desktop SDM template configured. In the logs there are multiple entries for:
"Oct 19 09:18:53: %PLATFORM_UCAST-6-PREFIX: One or more, more specific prefixes
could not be programmed into TCAM and are being covered by a less specific prefix"
...and when issuing the below command a fairly large number of routes seem to be impacted, for instance:
sh platform ip unicast failed route
[ouput omitted]
Entries covered by 10.73.0.0/16 Tbl:0:
10.73.11.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.9.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.27.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.17.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.21.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.4.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
10.73.16.0/24 Tbl:0 : Cover:10.73.0.0/16 Tbl:0
Total of 7 entries covered by 10.73.0.0/16 Tbl:0
[output omitted]
...yet if I look at one of the routes that the switch doesn't seem to have a specific routing entry for, it appears to have it:
PPFX_X37_274#sh ip ro 10.73.16.251
Routing entry for 10.73.16.0/24
Known via "ospf 1", distance 110, metric 25201, type inter area
Last update from 10.4.112.234 on Vlan900, 17:00:20 ago
Routing Descriptor Blocks:
* 10.4.112.246, from 10.200.1.216, 17:00:20 ago, via Vlan900
Route metric is 25201, traffic share count is 1
Q. Why does it still know a precise route if it says it is covered by 10.73.0.0/16 in the above command?
Q. What is the exact impact of packets being routed to this subnet via the 3750 given that it is stating it doesn't have enough resources to hold this route?
I'm hoping it doesn't mean that every packet is effectively process switched each time by the switch CPU (adding latency)....
I understand the resolution is to change the SDM template to routing to allow for extra resources to be assigned for routing, but want to understand the exact current impact this causes.
Rgds
Phil
10-19-2010 03:10 AM
Hello Phil,
the multilayer switch has not enough resources in the forwarding plane, with the current SDM template it can install in the TCAM table up to N (6k) routes.
in your case the device is attempting to make an intelligent decision on what routes to leave out of TCAM table and some component routes are not installed in TCAM.
This does not mean that the route is not known as you have noted.
You can do the following:
you can apply appropriate route summarization on upstream devices in order to make this device happy with this current setup
AND/OR
you can move to a different SDM template but this requires a switch reload
see
moving to routing template can accomodate 11000 routes.
but you can see reduced other TCAM resources
You need to consider what you really need.
prefixes that are not installed in the TCAM and are not covered are process switched for sure causing high cpu usage.
If you don't need the component routes find a way to get only the aggregate routes
what can happen is that component route has a different next-hop and the device is using the next-hop of the aggregate route this can even cause connectivity problems if the major network is partiotioned. (just an example I don't know your environment...)
So more then a perfomance issue you can face suboptimal routing or even connectivity problems if passing traffic to some device that wll try to route it back to this device.
Hope to help
Giuseppe
10-19-2010 03:51 AM
Hi Guiseppe,
The 3750 is in Area0 and we are already performing some summarisation, but can not improve this further.
PPFX_X37_274#sh ip ro summary
IP routing table name is Default-IP-Routing-Table(0)
IP routing table maximum-paths is 32
Route Source Networks Subnets Overhead Memory (bytes)
connected 0 2 232 304
static 0 0 0 0
ospf 1 65 2155 32768 337440
Intra-area: 122 Inter-area: 1308 External-1: 630 External-2: 160
NSSA External-1: 0 NSSA External-2: 0
internal 22 25784
Total 87 2157 233000 363528
My understanding is the desktop default allows for upto 2k unicast routes (indirect), and above shows 2,157 subnets suggesting I've just hit the threshold... SDM prefer routing allows for 8k indirect routes (I guess this means non-attached networks??). So a change of template should resolve this problem.
FYI - I am seeing packet loss and high latency for only a subset of subnets across this switch, so I guess this is the cause. The switch is also running at around 35% CPU utilisation.
thanks
Phil
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide