08-14-2008 12:40 AM - edited 03-06-2019 12:47 AM
Hello,
I have made several tests with my 3845 router and NM-16ESW and I found this problem :
I have PCs in main building connected in Gigabit to a switch which is connected to the 3845 in Gigabit too. The 3845 is linked to 4 other buildings via Hertzian Waves at 100Mb with its NM-16ESW.
PCs are on port Gi0/1 of 3845
Building A on port Fa1/0 of NM-16ESW
Building B on port Fa1/1 of NM-16ESW
Building C on port Fa1/2 of NM-16ESW
Building D on port Fa1/3 of NM-16ESW
When I generate a 100 Mb traffic from 1 PC in main building to 1 PC in bulding A --> bandwidth is 100Mb/s.
When I generate a 100 Mb traffic from 2 PCs in main building to 1 PC in bulding A and 1 PC in building B --> bandwidth is 50Mb/s for each connection.
When I generate a 100 Mb traffic from 3 PCs in main building to 1 PC in bulding A and 1 PC in building B and 1PC in building C --> bandwidth is 33Mb/s for each connection.
It seems that NM-16ESW shared 100Mb/s to all ports, we don't have 100Mb/s dedicated per port.
Is this problem located on my NM-16ESW card or it is a general one ? Is there a command to unlock this limitation and have a full 100Mb link dedicated by interface ?
Thanks a lot for your responses !
Best regards,
Florian
08-14-2008 03:24 AM
The NM slot in the 3800s is connected to a PCI bus, so one would think that would be adquate to support your four 100 Mbps ports. However, it's possible the particular module your using might have limited bandwidth to how it interfaces to the module slot. I'm unable to find any mention on that detail. Unless someone else knows, this might require information from Cisco.
Have you tried a throughput test between ports on the NM-16ESW? If not, you might try both L2 and L3 and see if there's any difference. In the former, traffic should stay within the switch module. With the latter, it would need to jump VLANs defined to the router. If you see a major difference between the two, it might further document an NM interface bandwidth bottleneck.
Something else to check, the 3845 can run short of performance at high forwarding rates much, much sooner than most L3 switches. What's the CPU utilization look like during your tests?
08-19-2008 08:38 AM
Thanks for your reply.
I have made more tests :
- I tried L2 and L3 mode --> same problem
- During tests CPU is loaded at 22% or 23%
- I made tests with PCs connected on NM card and PCs on LAN backbone. Simultaneous transfers were established :
1) In the way PCs on NM card to PCs on Backbone it's OK. Each PC on NM card can send at 100Mb.
2) But in the way PCs from Backbone to PCs on NM card it's NOK. Each PC on NM card received at the 50Mb rate.
To isolate the problem I have made the same test with a switch in place of 3845 with NM card. There is no problem, all is OK with 100 Mb dedicated per port and per PC.
Best regard,
Florian
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