I’m very surprised by the performance of Cisco 2901 router in comparison to his fellow Cisco 2951 router. I do not know, maybe it’s got such an IOS or the developers forgot something somewhere limit, but the equipment was pleased. Now, let’s share tests results on performance of the both routers to make your choice.First, we start with the test on a clean routing between interfaces on Cisco 2901. By connecting the two generators to Gi0 / 0 and Gi0 / 1, putting ip addresses from different subnets, and let the traffic through iperf, received around 550Mbit / c of live traffic. Test sub-one of the physical interfaces showed almost similar results. The only difference was in the boot CPU router on physical occupancy was less in one and half times. Then, let’s try to enter the equipment in MPLS network and do EoMPLS on it. It was possible to pump in both directions about a 230Mbit / c (for comparison with 28xx series was about 12 Mbps / s) for ip vpn received in the same order of test 500Mbit / c, but the asymmetry of the traffic seen on the ip vpn Cisco introduced in MPLS. When traffic goes from Cisco 2901 router to the remote PE, it is like writing a 500Mbit / c and CPU load was 36%, but when, on the contrary from the remote PE to the s2901 rate has dropped and the order of 320Mbit / c and 26% CPU load. A similar behavior has been reported on the 2911th. The processor is not like one head taller previous series.
If observed from remote client traffic of the same order of 230Mbit / c and CPU load under a hundred, then in the direction of the client is going to 55Mbit / c and a load of about 70 percent with a periodic load jumping out to a hundred. What caused it, and found none. Tried to create polisery by MQC and hang them. 1200 pieces hung, do not try, expecting that the 1024th will swear, but all went well. On individuals and on sub-interfaces could very well cut, but on l2tpv3 the input of non-limiting speed, but managed to limit and that the good old rate-limit `th. All of course have done with the license data till now.
Now, let’s turn to tests of NAT. At its default 512MB of RAM, managed to get almost 400tysch broadcasts. Graphics memory and processor speak for themselves. The load is true for interfaces in the test did not give.
Generally speaking, if you do not plan on MPLS, IP SLA, IPv6’s troubles, no license is necessary, it is enough to open a functional – BGP, OSPF, speed limit, and netflow. But since I want to see what the equipment is capable of, you can activate a temporary license at the time of its expiration functionality should be shut down. To activate the temporary license visit the portal is required. All tests have 60 days. We have activated the license data.The first thing that came to mind, to see how much actual traffic can be routed through it. No ACL, no speed limiters, no dynamic routing processes or collection of statistics – nothing. Only test the traffic generators and their prikonekchenye subnet to different physical interfaces, then see what happened.CPU usage with clean routing does not exceed 15%, the memory was abundantly. The second that I wanted to see what if download Full-View cisco in 2951, if she can handle. Raised BGP, leaked about 340,000 routes, swallowed and though that, there you are already loaded with the full route table with all the same testing iperf.CPU utilization of approximately 15% of all cover those, but the memory will eat almost the entire free number is around 16MB.
Where a ful-twist, there is a second, everyone is interested. Well aware that there will, then it’s interesting! Began to pump the second ful twist in cisco 2951 and here she stumbled from the lack of memory. By default, it is 512MB, slightly more than half eaten unfolded IOS. Deliver back the memory and you can try one more time. Just for fun, decided to take two full view and restart the cisco 2951, loaded, climbed both peer, swayed both full view, but when the treatment started disconnected cef, – output two full loaded tables with disabled cef and 5Mbit / c by cisco 2951. No wonder deliver memory, let’s see what happens.
What’s next? Of course ACL and polisery, sub-interfaces. Quickly Generate Script 700 sabinterfeysov, 1500 policy-map’s, each hung sabinterfeys speed limit on the input and output, access-list’s on the input and output, for each sub prikonekchenoy on its own subnet. Of this size – configuration file did not want to save, had to include compression. We test and enjoy the performance by roughly 20% worse than the last schedule, load on the processor has risen to 36%, sabinterfeysy not eaten little memory. While the results of the tests for such equipment costs are happy.