Virtual Switch Layer 2 Support
The OSA-Express features can support two transport modes of the OSA model; Layer 2 (Link Layer or MAC Layer) and Layer 3 (Network layer). Both the virtual switch and Linux then are configured to support the desired capability (Layer 2 or Layer 3).
The virtual switch, introduced in z/VM V4.4 supports Layer3 mode, is designed to improve connectivity to a physical LAN for hosts coupled to a guest LAN. It eliminates the need for a routing virtual machine by including the switching function in CP to provide IPv4 connectivity to a physical LAN through an OSA-Express Adapter.
With the PTF for APAR VM63538 and PQ98202, z/VM V5.1 will support Layer 2 mode. In this mode, each port on the virtual switch is referenced by its Media Access Control (MAC) address instead of by Internet Protocol (IP) address. Data is transported and delivered in Ethernet frames, providing the ability to handle protocol-independent traffic for both IP (IPv4 or IPv6) and non-IP, such as IPX, NetBIOS, or SNA. Coupled with the Layer 2 support in Linux for zSeries and the OSA-Express and OSA-Express2 support for the z890 and z990, Linux images deployed as guests of z/VM can use this protocol-independent capability through the virtual switch.
This section summarizes measurement results comparing IPv4 over virtual switch with Layer 3 to IPv4 over virtual switch with Layer 2. It also compares IPv4 (Layer 2) with IPv6 (Layer 2). Measurements were done using OSA-Express Gigabit Ethernet cards.
Methodology: An internal version of the Application Workload Modeler (AWM) was used to drive request-response (RR) and streaming (STR) workloads with IPv4 (Layer 3), IPv4 (Layer 2) and IPv6 (Layer 2). The request-response workload consisted of the client sending 200 bytes to the server and the server responding with 1000 bytes. This interaction was repeated for 200 seconds. The streaming workload consisted of the client sending 20 bytes to the server and the server responding with 20MB. This sequence was repeated for 400 seconds.
A complete set of runs, consisting of 3 trials for each case, for 1, 10, and 50 client-server pairs, was done with the maximum transmission unit (MTU) set to 1492 (for RR and STR) and 8992 (for STR only).
The measurements were done on a 2084-324 with 2 dedicated processors in each LPAR used. Connectivity between the two LPARs was over an OSA-Express card to OSA-Express card. The OSA level was 6.26. The software used includes:
- z/VM 5.1.0 with APAR VM63538
- TCP/IP 5.1.0 with PQ97436 and PQ98202 (the virtual switch controller stack)
- Linux SuSe SLES8 kernel levels 2.4.21-251 with qeth module dated 20041104
Results: The following tables compare the average of 3 trials for each measurement between IPv4 over a virtual switch configured for Layer 3 (noted as v4 in the tables) and IPv4 over a virtual switch configured for Layer 2 (noted as v5 in the tables), and between IPv4 and IPv6 (noted as v6 in the tables) over virtual switch configured for Layer 2. The numbers shown are the percent increase (or decrease). A positive number for throughput (either MB/sec or trans/sec) is good and a negative number for CPU time is good.
In general, the larger the MTU and/or the more activity, the smaller the difference between IPv4 over Layer 3 versus Layer 2.
Number of clients | 1 | 10 | 50 |
MTU 1492 runid V4 (Layer3) trans/sec Total CPU msec/trans Emul CPU msec/trans CP CPU msec/trans |
|
|
|
runid V5 (Layer2) trans/sec Total CPU msec/trans Emul CPU msec/trans CP CPU msec/trans | vl5rn01 1414.05 0.0650 0.0270 0.0380 | vl5rn10 10883.51 0.0460 0.0230 0.0230 | vl5rn50 25957.55 0.0357 0.0210 0.0147 |
runid V6 (Layer2) trans/sec Total CPU msec/trans Emul CPU msec/trans CP CPU msec/trans | vl6rn01 1389.10 0.0687 0.0290 0.0397 | vl6rn10 10642.39 0.0483 0.0260 0.0223 | vl6rn50 25196.77 0.0390 0.0233 0.0157 |
% diff V4 to V5 |
|
|
|
trans/sec Total CPU msec/trans Emul CPU msec/trans CP CPU msec/trans | 2% 3% 8% -1% | 4% 5% 3% 6% | 5% 2% 0% 5% |
% diff V5 to V6 |
|
|
|
trans/sec Total CPU msec/trans Emul CPU msec/trans CP CPU msec/trans | -2% 6% 7% 4% | -2% 5% 13% -3% | 0% 9% 11% 7% |
Note: 2084-324; z/VM 5.1.0; TCP/IP 510; V5 denotes IPv4 over virtual switch configured for Layer2 |
Throughput is slightly higher for MTU 1492 for IPv4 over Layer 2
and slightly lower for IPv6.
Number of clients | 1 | 10 | 50 |
MTU 1492 runid V4 (Layer3) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB |
|
|
|
runid V5 (Layer2) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | vl5sn01 42.87 9.80 4.26 5.54 | vl5sn10 71.30 8.34 4.12 4.22 | vl5sn50 89.57 7.93 3.95 3.97 |
runid V6 (Layer2) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | vl6sn01 41.40 11.03 5.33 5.70 | vl6sn10 69.20 9.25 5.09 4.16 | vl6sn50 82.57 8.67 4.83 3.84 |
%diff V4 to V5 |
|
|
|
MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | 4% 7% 7% 7% | -0% 5% 4% 6% | -1% 4% 4% 4% |
%diff V5 to V6 |
|
|
|
MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | -3% 13% 25% 3% | -3% 11% 23% -1% | -8% 9% 22% -3% |
MTU 8992 runid V4 (Layer3) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB |
|
|
|
runid V5 (Layer2) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | vl5sj01 30.93 5.75 2.26 3.49 | vl5sj10 111.50 4.73 2.11 2.62 | vl5sj50 115.50 4.85 2.19 2.67 |
runid V6 (Layer2) MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | vl6sj01 30.50 5.68 2.36 3.32 | vl6sj10 111.03 4.75 2.22 2.53 | vl6sj50 115.10 4.91 2.29 2.61 |
% diff V4 to V5 |
|
|
|
MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | 1% 3% 2% 4% | 0% 3% 1% 5% | 0% 2% 1% 3% |
% diff V5 to V6 |
|
|
|
MB/sec Total CPU msec/MB Emul CPU msec/MB CP CPU msec/MB | -1% -1% 4% -5% | 0% 0% 5% -3% | 0% 1% 5% -2% |
Note: 2084-324; z/VM 5.1.0; TCP/IP 510; V5 denotes IPv4 over virtual switch configured for Layer 2 |
While throughput for MTU 1492 was almost the same for IPv4 over Layer 3 compared to Layer 2, the CPU cost is higher for Layer 2. However, the cost does decrease as the load increases. IPv6 compared to IPv4 gets less throughput and costs more in CPU time. For MTU 8992 the results are almost the same for all three cases.