Fail Cisco Marathon (1)

Bkjia.com exclusive Article] We have tested 13 manufacturers' 10 GB Ethernet switches. In addition to six mainstream vendors, Arista, Blade, Cisco, Dell, Extreme, HP ProCurve ), we will also test 3Com/H3C, Brocade, Enterasys, Force10, Fujitsu, Juniper, and Raptor products.

3Com/H3C has been acquired by HP during this test.) You will learn about the lack of samples and we will obtain relevant samples through HP) and announce that you will not participate in this test. Then Brocade, Enterasys, Force10, And Juniper declared a lack of samples. Fujitsu and Raptor did not respond to our test invitation. We are somewhat disappointed with the absence of these network device giants, hoping to get their support in future tests. Note: H3C has cut off all external media publicity recently. I don't know if it is caused by embarrassment.

We will evaluate the performance of these switches in 10 aspects: appearance, management and ease of use, power consumption, MAC address capacity, unicast and multicast throughput, unicast and multicast latency and jitter, fairness of link aggregation, multicast group throughput, and latency of adding/exiting a stream, and forwarding tolerance capability forward pressure ). BKJIA editing Note: for the performance evaluation of vswitches under normal circumstances, note the following: features, application-level QoS Assurance, VLAN support, network management, and link aggregation, supports VRRP.

In addition to the first two aspects, we have used the Spirent Test Center's fault generation/analyzer with 24 10 GB Ethernet filtering CV modules and 10 gbase-sr wireless transceiver) to test these switches.

We asked the supplier to complete a detail questionnaire to evaluate the appearance of the vswitch. However, we did not verify the answers to each question in the questionnaire.

The evaluation results of management and ease of use are evaluated through the use results of the same functions of the vswitch and the feedback from the appearance questionnaire during the test. For example, whether the network management function is supported)

We use Fluke 322 and Fluke 335 clamp table clamp meter) to test the power consumption of each switch. This test includes three aspects: Ac-line voltage, active-line current at idle time, and active-line current at full load. We configure the Spirent test center to provide communication traffic at the current frequency for each port of the switch, so that the control line and data line of the switch can run at full capacity. We get the power value by the product of voltage and current, and then we take 12 wireless receivers from the switch and repeat this method (calculate the power at idle time ).

We tested the MAC address capacity in the Spirent test center according to RFC 2889. This standard allows you to perform a binary search to determine the maximum number of MAC addresses that a switch can learn when it is not under a flood attack flooding. During the test of this data item, the MAC address survival time of the Spirent data center is set to twice the MAC address survival time of the tested switch.

To test unicast throughput, latency, and jitter, we set up the Spirent data center to provide communication traffic to all switch ports in a full mesh mode. In each data test, we conducted a 60-second run test of 64, 65, 108, 256, 1518, and 9216 bytes per frame, the binary search is used to determine the throughput frequency of the vswitch. We tested the throughput, average, maximum latency, and average and maximum jitter of the vswitch when using a frame test with a specific number of bytes.

For multicast throughput, latency, and jitter tests, we use frames with the same number of bytes as unicast tests. Here, we configure a separate Spirent Data Center Port to send multi-wave communication traffic and add the remaining 23 ports to the same 989 multicast groups.

For the Fairness Test of Link Aggregation, We have configured the Sprient data center to use it as the LACP Protocol Link Aggregation Control Protocol and Link Aggregation Control Protocol) one member of the link aggregation competition for data transmission with the other seven ports on the host. Initially, we introduced eight port Link aggregation groups (LAG) and provided the other seven ports with a single communication traffic direction from competing ports on the test center to the switch, then, go to the data center's LAG member, and then to the competition host of the test center ). To avoid communication traffic competition among other LAG members, we only send 10% linear frequency communication traffic. Then, we close a LACP on the Data Center Port and send the same communication traffic to the LAG test with eight members. In both cases, we will record the data packets received from each LAG Member, compare the number of received data frames with the standard number, and then perform a fair computing analysis.

To test the multicast group capacity, we will use the RFC 3918 standard in the Spirent data center. This standard is added to a fixed number of groups and then attempts to send communication traffic to all groups. The test instrument uses a binary search to determine the maximum number of successfully added groups. If the communication traffic of the vswitch is successfully sent to all groups, the vswitch passes this test. If one or more groups are not added, the vswitch fails this test. We use the most challenging test environment where receivers on 23 ports are added to all groups at the same time.

To test the addition/departure latency of multicast groups, we use the RFC 3918 standard in the data center again. This test is the opposite of throughput and latency tests: even if the IGMP listening table is empty, we still send multicast communication traffic to the switch. Then, we send an IGMP addition message to all receiver ports. "Join latency" refers to the time difference between "sending and adding messages on the specified port" and "receiving the first multicast frame from the Working Group on the port. For latency tests, there are also other evaluation methods: We will calculate the time difference from "sending the first exit message" to "the switch stops sending messages to the port workgroup.

Once again, we use the RFC 2889 standard in the data center for the evaluation of forwarding tolerance capabilities. This standard allows users to set an illegal data frame sending interval-a 12-bit interval less than the 802.3 Ethernet standard. Because the communication traffic sent from the Spirent data center to the switch is faster than the normal speed, therefore, the switch will lose part of the communication traffic-but its sending frequency will reveal more or less what its clock rate is set.