aboutsummaryrefslogtreecommitdiffstats
path: root/docs/testing/user/userguide/opnfv_yardstick_tc001.rst
diff options
context:
space:
mode:
Diffstat (limited to 'docs/testing/user/userguide/opnfv_yardstick_tc001.rst')
-rw-r--r--docs/testing/user/userguide/opnfv_yardstick_tc001.rst133
1 files changed, 133 insertions, 0 deletions
diff --git a/docs/testing/user/userguide/opnfv_yardstick_tc001.rst b/docs/testing/user/userguide/opnfv_yardstick_tc001.rst
new file mode 100644
index 000000000..b53c508a6
--- /dev/null
+++ b/docs/testing/user/userguide/opnfv_yardstick_tc001.rst
@@ -0,0 +1,133 @@
+s work is licensed under a Creative Commons Attribution 4.0 International
+.. License.
+.. http://creativecommons.org/licenses/by/4.0
+.. (c) OPNFV, Ericsson AB and others.
+
+*************************************
+Yardstick Test Case Description TC001
+*************************************
+
+.. _pktgen: https://www.kernel.org/doc/Documentation/networking/pktgen.txt
+
++-----------------------------------------------------------------------------+
+|Network Performance |
+| |
++--------------+--------------------------------------------------------------+
+|test case id | OPNFV_YARDSTICK_TC001_NETWORK PERFORMANCE |
+| | |
++--------------+--------------------------------------------------------------+
+|metric | Number of flows and throughput |
+| | |
++--------------+--------------------------------------------------------------+
+|test purpose | The purpose of TC001 is to evaluate the IaaS network |
+| | performance with regards to flows and throughput, such as if |
+| | and how different amounts of flows matter for the throughput |
+| | between hosts on different compute blades. Typically e.g. |
+| | the performance of a vSwitch depends on the number of flows |
+| | running through it. Also performance of other equipment or |
+| | entities can depend on the number of flows or the packet |
+| | sizes used. |
+| | |
+| | The purpose is also to be able to spot the trends. |
+| | Test results, graphs and similar shall be stored for |
+| | comparison reasons and product evolution understanding |
+| | between different OPNFV versions and/or configurations. |
+| | |
++--------------+--------------------------------------------------------------+
+|test tool | pktgen |
+| | |
+| | Linux packet generator is a tool to generate packets at very |
+| | high speed in the kernel. pktgen is mainly used to drive and |
+| | LAN equipment test network. pktgen supports multi threading. |
+| | To generate random MAC address, IP address, port number UDP |
+| | packets, pktgen uses multiple CPU processors in the |
+| | different PCI bus (PCI, PCIe bus) with Gigabit Ethernet |
+| | tested (pktgen performance depends on the CPU processing |
+| | speed, memory delay, PCI bus speed hardware parameters), |
+| | Transmit data rate can be even larger than 10GBit/s. Visible |
+| | can satisfy most card test requirements. |
+| | |
+| | (Pktgen is not always part of a Linux distribution, hence it |
+| | needs to be installed. It is part of the Yardstick Docker |
+| | image. |
+| | As an example see the /yardstick/tools/ directory for how |
+| | to generate a Linux image with pktgen included.) |
+| | |
++--------------+--------------------------------------------------------------+
+|test | This test case uses Pktgen to generate packet flow between |
+|description | two hosts for simulating network workloads on the SUT. |
+| | |
++--------------+--------------------------------------------------------------+
+|traffic | An IP table is setup on server to monitor for received |
+|profile | packets. |
+| | |
++--------------+--------------------------------------------------------------+
+|configuration | file: opnfv_yardstick_tc001.yaml |
+| | |
+| | Packet size is set to 60 bytes. |
+| | Number of ports: 10, 50, 100, 500 and 1000, where each |
+| | runs for 20 seconds. The whole sequence is run twice |
+| | The client and server are distributed on different hardware. |
+| | |
+| | For SLA max_ppm is set to 1000. The amount of configured |
+| | ports map to between 110 up to 1001000 flows, respectively. |
+| | |
++--------------+--------------------------------------------------------------+
+|applicability | Test can be configured with different: |
+| | |
+| | * packet sizes; |
+| | * amount of flows; |
+| | * test duration. |
+| | |
+| | Default values exist. |
+| | |
+| | SLA (optional): max_ppm: The number of packets per million |
+| | packets sent that are acceptable to loose, not received. |
+| | |
++--------------+--------------------------------------------------------------+
+|usability | This test case is used for generating high network |
+| | throughput to simulate certain workloads on the SUT. Hence |
+| | it should work with other test cases. |
+| | |
++--------------+--------------------------------------------------------------+
+|references | pktgen_ |
+| | |
+| | ETSI-NFV-TST001 |
+| | |
++--------------+--------------------------------------------------------------+
+|pre-test | The test case image needs to be installed into Glance |
+|conditions | with pktgen included in it. |
+| | |
+| | No POD specific requirements have been identified. |
+| | |
++--------------+--------------------------------------------------------------+
+|test sequence | description and expected result |
+| | |
++--------------+--------------------------------------------------------------+
+|step 1 | Two host VMs are booted, as server and client. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 2 | Yardstick is connected with the server VM by using ssh. |
+| | 'pktgen_benchmark' bash script is copyied from Jump Host to |
+| | the server VM via the ssh tunnel. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 3 | An IP table is setup on server to monitor for received |
+| | packets. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 4 | pktgen is invoked to generate packet flow between two server |
+| | and client for simulating network workloads on the SUT. |
+| | Results are processed and checked against the SLA. Logs are |
+| | produced and stored. |
+| | |
+| | Result: Logs are stored. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 5 | Two host VMs are deleted. |
+| | |
++--------------+--------------------------------------------------------------+
+|test verdict | Fails only if SLA is not passed, or if there is a test case |
+| | execution problem. |
+| | |
++--------------+--------------------------------------------------------------+