aboutsummaryrefslogtreecommitdiffstats
path: root/docs/release/results/os-nosdn-nofeature-ha.rst
diff options
context:
space:
mode:
Diffstat (limited to 'docs/release/results/os-nosdn-nofeature-ha.rst')
-rw-r--r--docs/release/results/os-nosdn-nofeature-ha.rst492
1 files changed, 492 insertions, 0 deletions
diff --git a/docs/release/results/os-nosdn-nofeature-ha.rst b/docs/release/results/os-nosdn-nofeature-ha.rst
new file mode 100644
index 000000000..9e52731d5
--- /dev/null
+++ b/docs/release/results/os-nosdn-nofeature-ha.rst
@@ -0,0 +1,492 @@
+.. This work is licensed under a Creative Commons Attribution 4.0 International
+.. License.
+.. http://creativecommons.org/licenses/by/4.0
+
+
+======================================
+Test Results for os-nosdn-nofeature-ha
+======================================
+
+.. toctree::
+ :maxdepth: 2
+
+
+apex
+====
+
+.. _Grafana: http://testresults.opnfv.org/grafana/dashboard/db/yardstick-main
+.. _POD1: https://wiki.opnfv.org/pharos?&#community_test_labs
+
+
+Overview of test results
+------------------------
+
+See Grafana_ for viewing test result metrics for each respective test case. It
+is possible to chose which specific scenarios to look at, and then to zoom in
+on the details of each run test scenario as well.
+
+All of the test case results below are based on 4 scenario test
+runs, each run on the LF POD1_ between August 25 and 28 in
+2016.
+
+TC002
+-----
+The round-trip-time (RTT) between 2 VMs on different blades is measured using
+ping. Most test run measurements result on average between 0.74 and 1.08 ms.
+A few runs start with a 0.99 - 1.07 ms RTT spike (This could be because of
+normal ARP handling). One test run has a greater RTT spike of 1.35 ms.
+To be able to draw conclusions more runs should be made. SLA set to 10 ms.
+The SLA value is used as a reference, it has not been defined by OPNFV.
+
+TC005
+-----
+The IO read bandwidth looks similar between different dates, with an
+average between approx. 128 and 136 MB/s. Within each test run the results
+vary, with a minimum 5 MB/s and maximum 446 MB/s on the totality. Most runs
+have a minimum BW of 5 MB/s (one run at 6 MB/s). The maximum BW varies more in
+absolute numbers between the dates, between 416 and 446 MB/s.
+SLA set to 400 MB/s. The SLA value is used as a reference, it has not been
+defined by OPNFV.
+
+TC010
+-----
+The measurements for memory latency are similar between test dates and result
+in approx. 1.09 ns. The variations within each test run are similar, between
+1.0860 and 1.0880 ns.
+SLA set to 30 ns. The SLA value is used as a reference, it has not been defined
+by OPNFV.
+
+TC011
+-----
+Packet delay variation between 2 VMs on different blades is measured using
+Iperf3. The reported packet delay variation varies between 0.0025 and 0.0148 ms,
+with an average delay variation between 0.0056 ms and 0.0157 ms.
+
+TC012
+-----
+Between test dates, the average measurements for memory bandwidth result in
+approx. 19.70 GB/s. Within each test run the results vary more, with a minimal
+BW of 18.16 GB/s and maximum of 20.13 GB/s on the totality.
+SLA set to 15 GB/s. The SLA value is used as a reference, it has not been
+defined by OPNFV.
+
+TC014
+-----
+The Unixbench processor test run results vary between scores 3224.4 and 3842.8,
+one result each date. The average score on the total is 3659.5.
+No SLA set.
+
+TC037
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The RTT results are similar throughout the different test dates and runs at
+approx. 15 ms. Some test runs show an increase with many flows, in the range
+towards 16 to 17 ms. One exception standing out is Feb. 15 where the average
+RTT is stable at approx. 13 ms. The PPS results are not as consistent as the
+RTT results.
+In some test runs when running with less than approx. 10000 flows the PPS
+throughput is normally flatter compared to when running with more flows, after
+which the PPS throughput decreases. Around 20 percent decrease in the worst
+case. For the other test runs there is however no significant change to the PPS
+throughput when the number of flows are increased. In some test runs the PPS
+is also greater with 1000000 flows compared to other test runs where the PPS
+result is less with only 2 flows.
+
+The average PPS throughput in the different runs varies between 414000 and
+452000 PPS. The total amount of packets in each test run is approx. 7500000 to
+8200000 packets. One test run Feb. 15 sticks out with a PPS average of
+558000 and approx. 1100000 packets in total (same as the on mentioned earlier
+for RTT results).
+
+There are lost packets reported in most of the test runs. There is no observed
+correlation between the amount of flows and the amount of lost packets.
+The lost amount of packets normally range between 100 and 1000 per test run,
+but there are spikes in the range of 10000 lost packets as well, and even
+more in a rare cases.
+
+CPU utilization statistics are collected during UDP flows sent between the VMs
+using pktgen as packet generator tool. The average measurements for CPU
+utilization ratio vary between 1% to 2%. The peak of CPU utilization ratio
+appears around 7%.
+
+TC069
+-----
+Between test dates, the average measurements for memory bandwidth vary between
+22.6 and 29.1 GB/s. Within each test run the results vary more, with a minimal
+BW of 20.0 GB/s and maximum of 29.5 GB/s on the totality.
+SLA set to 6 GB/s. The SLA value is used as a reference, it has not been
+defined by OPNFV.
+
+TC070
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The RTT results are similar throughout the different test dates and runs at
+approx. 15 ms. Some test runs show an increase with many flows, in the range
+towards 16 to 17 ms. One exception standing out is Feb. 15 where the average
+RTT is stable at approx. 13 ms. The PPS results are not as consistent as the
+RTT results.
+In some test runs when running with less than approx. 10000 flows the PPS
+throughput is normally flatter compared to when running with more flows, after
+which the PPS throughput decreases. Around 20 percent decrease in the worst
+case. For the other test runs there is however no significant change to the PPS
+throughput when the number of flows are increased. In some test runs the PPS
+is also greater with 1000000 flows compared to other test runs where the PPS
+result is less with only 2 flows.
+
+The average PPS throughput in the different runs varies between 414000 and
+452000 PPS. The total amount of packets in each test run is approx. 7500000 to
+8200000 packets. One test run Feb. 15 sticks out with a PPS average of
+558000 and approx. 1100000 packets in total (same as the on mentioned earlier
+for RTT results).
+
+There are lost packets reported in most of the test runs. There is no observed
+correlation between the amount of flows and the amount of lost packets.
+The lost amount of packets normally range between 100 and 1000 per test run,
+but there are spikes in the range of 10000 lost packets as well, and even
+more in a rare cases.
+
+Memory utilization statistics are collected during UDP flows sent between the
+VMs using pktgen as packet generator tool. The average measurements for memory
+utilization vary between 225MB to 246MB. The peak of memory utilization appears
+around 340MB.
+
+TC071
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The RTT results are similar throughout the different test dates and runs at
+approx. 15 ms. Some test runs show an increase with many flows, in the range
+towards 16 to 17 ms. One exception standing out is Feb. 15 where the average
+RTT is stable at approx. 13 ms. The PPS results are not as consistent as the
+RTT results.
+In some test runs when running with less than approx. 10000 flows the PPS
+throughput is normally flatter compared to when running with more flows, after
+which the PPS throughput decreases. Around 20 percent decrease in the worst
+case. For the other test runs there is however no significant change to the PPS
+throughput when the number of flows are increased. In some test runs the PPS
+is also greater with 1000000 flows compared to other test runs where the PPS
+result is less with only 2 flows.
+
+The average PPS throughput in the different runs varies between 414000 and
+452000 PPS. The total amount of packets in each test run is approx. 7500000 to
+8200000 packets. One test run Feb. 15 sticks out with a PPS average of
+558000 and approx. 1100000 packets in total (same as the on mentioned earlier
+for RTT results).
+
+There are lost packets reported in most of the test runs. There is no observed
+correlation between the amount of flows and the amount of lost packets.
+The lost amount of packets normally range between 100 and 1000 per test run,
+but there are spikes in the range of 10000 lost packets as well, and even
+more in a rare cases.
+
+Cache utilization statistics are collected during UDP flows sent between the
+VMs using pktgen as packet generator tool. The average measurements for cache
+utilization vary between 205MB to 212MB.
+
+TC072
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The RTT results are similar throughout the different test dates and runs at
+approx. 15 ms. Some test runs show an increase with many flows, in the range
+towards 16 to 17 ms. One exception standing out is Feb. 15 where the average
+RTT is stable at approx. 13 ms. The PPS results are not as consistent as the
+RTT results.
+In some test runs when running with less than approx. 10000 flows the PPS
+throughput is normally flatter compared to when running with more flows, after
+which the PPS throughput decreases. Around 20 percent decrease in the worst
+case. For the other test runs there is however no significant change to the PPS
+throughput when the number of flows are increased. In some test runs the PPS
+is also greater with 1000000 flows compared to other test runs where the PPS
+result is less with only 2 flows.
+
+The average PPS throughput in the different runs varies between 414000 and
+452000 PPS. The total amount of packets in each test run is approx. 7500000 to
+8200000 packets. One test run Feb. 15 sticks out with a PPS average of
+558000 and approx. 1100000 packets in total (same as the on mentioned earlier
+for RTT results).
+
+There are lost packets reported in most of the test runs. There is no observed
+correlation between the amount of flows and the amount of lost packets.
+The lost amount of packets normally range between 100 and 1000 per test run,
+but there are spikes in the range of 10000 lost packets as well, and even
+more in a rare cases.
+
+Network utilization statistics are collected during UDP flows sent between the
+VMs using pktgen as packet generator tool. Total number of packets received per
+second was average on 200 kpps and total number of packets transmitted per
+second was average on 600 kpps.
+
+Detailed test results
+---------------------
+The scenario was run on LF POD1_ with:
+Apex
+OpenStack Mitaka
+OpenVirtualSwitch 2.5.90
+OpenDayLight Beryllium
+
+Rationale for decisions
+-----------------------
+Pass
+
+Tests were successfully executed and metrics collected.
+No SLA was verified. To be decided on in next release of OPNFV.
+
+
+Joid
+====
+
+.. _Grafana: http://testresults.opnfv.org/grafana/dashboard/db/yardstick-main
+.. _POD5: https://wiki.opnfv.org/pharos?&#community_test_labs
+
+
+Overview of test results
+------------------------
+
+See Grafana_ for viewing test result metrics for each respective test case. It
+is possible to chose which specific scenarios to look at, and then to zoom in
+on the details of each run test scenario as well.
+
+All of the test case results below are based on 4 scenario test runs, each run
+on the Intel POD5_ between September 11 and 14 in 2016.
+
+TC002
+-----
+The round-trip-time (RTT) between 2 VMs on different blades is measured using
+ping. Most test run measurements result on average between 1.59 and 1.70 ms.
+Two test runs have reached the same greater RTT spike of 3.06 ms, which are
+1.66 and 1.70 ms average, but only one has the lower RTT of 1.35 ms. The other
+two runs have no similar spike at all. To be able to draw conclusions more runs
+should be made. SLA set to be 10 ms. The SLA value is used as a reference, it
+has not been defined by OPNFV.
+
+TC005
+-----
+The IO read bandwidth actually refers to the storage throughput and the
+greatest IO read bandwidth of the four runs is 173.3 MB/s. The IO read
+bandwidth of the four runs looks similar on different four days, with an
+average between 32.7 and 60.4 MB/s. One of the runs has a minimum BW of 429
+KM/s and other has a maximum BW of 173.3 MB/s. The SLA of read bandwidth sets
+to be 400 MB/s, which is used as a reference, and it has not been defined by
+OPNFV.
+
+TC010
+-----
+The tool we use to measure memory read latency is lmbench, which is a series of
+micro benchmarks intended to measure basic operating system and hardware system
+metrics. The memory read latency of the four runs is 1.1 ns on average. The
+variations within each test run are different, some vary from a large range and
+others have a small change. For example, the largest change is on September 14,
+the memory read latency of which is ranging from 1.12 ns to 1.22 ns. However,
+the results on September 12 change very little, which range from 1.14 ns to
+1.17 ns. The SLA sets to be 30 ns. The SLA value is used as a reference, it has
+not been defined by OPNFV.
+
+TC011
+-----
+Iperf3 is a tool for evaluating the pocket delay variation between 2 VMs on
+different blades. The reported pocket delay variations of the four test runs
+differ from each other. The results on September 13 within the date look
+similar and the values are between 0.0087 and 0.0190 ms, which is 0.0126 ms on
+average. However, on the fourth day, the pocket delay variation has a large
+wide change within the date, which ranges from 0.0032 ms to 0.0121 ms and has
+the minimum average value. The pocket delay variations of other two test runs
+look relatively similar, which are 0.0076 ms and 0.0152 ms on average. The SLA
+value sets to be 10 ms. The SLA value is used as a reference, it has not been
+defined by OPNFV.
+
+TC012
+-----
+Lmbench is also used to measure the memory read and write bandwidth, in which
+we use bw_mem to obtain the results. Among the four test runs, the memory
+bandwidth within the second day almost keep stable, which is 11.58 GB/s on
+average. And the memory bandwidth of the fourth day look similar as that of the
+second day, both of which remain stable. The other two test runs relatively
+change from a large wide range, in which the minimum memory bandwidth is 11.22
+GB/s and the maximum bandwidth is 16.65 GB/s with an average bandwidth of about
+12.20 GB/s. Here SLA set to be 15 GB/s. The SLA value is used as a reference,
+it has not been defined by OPNFV.
+
+TC014
+-----
+The Unixbench is used to measure processing speed, that is instructions per
+second. It can be seen from the dashboard that the processing test results
+vary from scores 3272 to 3444, and there is only one result one date. The
+overall average score is 3371. No SLA set.
+
+TC037
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The mean packet throughput of the four test runs is 119.85, 128.02, 121.40 and
+126.08 kpps, of which the result of the second is the highest. The RTT results
+of all the test runs keep flat at approx. 37 ms. It is obvious that the PPS
+results are not as consistent as the RTT results.
+
+The No. flows of the four test runs are 240 k on average and the PPS results
+look a little waved since the largest packet throughput is 184 kpps and the
+minimum throughput is 49 K respectively.
+
+There are no errors of packets received in the four runs, but there are still
+lost packets in all the test runs. The RTT values obtained by ping of the four
+runs have the similar average vaue, that is 38 ms, of which the worest RTT is
+93 ms on Sep. 14th.
+
+CPU load of the four test runs have a large change, since the minimum value and
+the peak of CPU load is 0 percent and 51 percent respectively. And the best
+result is obtained on Sep. 14th.
+
+TC069
+-----
+With the block size changing from 1 kb to 512 kb, the memory write bandwidth
+tends to become larger first and then smaller within every run test, which
+rangs from 22.3 GB/s to 26.8 GB/s and then to 18.5 GB/s on average. Since the
+test id is one, it is that only the INT memory write bandwidth is tested. On
+the whole, when the block size is 8 kb and 16 kb, the memory write bandwidth
+look similar with a minimal BW of 22.5 GB/s and peak value of 28.7 GB/s. SLA
+sets to be 7 GB/s. The SLA value is used as a a reference, it has not been
+defined by OPNFV.
+
+TC070
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The network latency is measured by ping, and the results of the four test runs
+look similar with each other. Within each test run, the maximum RTT can reach
+more than 80 ms and the average RTT is usually approx. 38 ms. On the whole, the
+average RTTs of the four runs keep flat.
+
+Memory utilization is measured by free, which can display amount of free and
+used memory in the system. The largest amount of used memory is 268 MiB on Sep
+14, which also has the largest minimum memory. Besides, the rest three test
+runs have the similar used memory. On the other hand, the free memory of the
+four runs have the same smallest minimum value, that is about 223 MiB, and the
+maximum free memory of three runs have the similar result, that is 337 MiB,
+except that on Sep. 14th, whose maximum free memory is 254 MiB. On the whole,
+all the test runs have similar average free memory.
+
+Network throughput and packet loss can be measured by pktgen, which is a tool
+in the network for generating traffic loads for network experiments. The mean
+network throughput of the four test runs seem quite different, ranging from
+119.85 kpps to 128.02 kpps. The average number of flows in these tests is
+24000, and each run has a minimum number of flows of 2 and a maximum number
+of flows of 1.001 Mil. At the same time, the corresponding packet throughput
+differ between 49.4k and 193.3k with an average packet throughput of approx.
+125k. On the whole, the PPS results seem consistent. Within each test run of
+the four runs, when number of flows becomes larger, the packet throughput seems
+not larger in the meantime.
+
+TC071
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The network latency is measured by ping, and the results of the four test runs
+look similar with each other. Within each test run, the maximum RTT can reach
+more than 94 ms and the average RTT is usually approx. 35 ms. On the whole, the
+average RTTs of the four runs keep flat.
+
+Cache utilization is measured by cachestat, which can display size of cache and
+buffer in the system. Cache utilization statistics are collected during UDP
+flows sent between the VMs using pktgen as packet generator tool.The largest
+cache size is 212 MiB in the four runs, and the smallest cache size is 75 MiB.
+On the whole, the average cache size of the four runs is approx. 208 MiB.
+Meanwhile, the tread of the buffer size looks similar with each other.
+
+Packet throughput can be measured by pktgen, which is a tool in the network for
+generating traffic loads for network experiments. The mean packet throughput of
+the four test runs seem quite different, ranging from 119.85 kpps to 128.02
+kpps. The average number of flows in these tests is 239.7k, and each run has a
+minimum number of flows of 2 and a maximum number of flows of 1.001 Mil. At the
+same time, the corresponding packet throughput differ between 49.4k and 193.3k
+with an average packet throughput of approx. 125k. On the whole, the PPS results
+seem consistent. Within each test run of the four runs, when number of flows
+becomes larger, the packet throughput seems not larger in the meantime.
+
+TC072
+-----
+The amount of packets per second (PPS) and round trip times (RTT) between 2 VMs
+on different blades are measured when increasing the amount of UDP flows sent
+between the VMs using pktgen as packet generator tool.
+
+Round trip times and packet throughput between VMs can typically be affected by
+the amount of flows set up and result in higher RTT and less PPS throughput.
+
+The RTT results are similar throughout the different test dates and runs at
+approx. 32 ms. The PPS results are not as consistent as the RTT results.
+
+Network utilization is measured by sar, that is system activity reporter, which
+can display the average statistics for the time since the system was started.
+Network utilization statistics are collected during UDP flows sent between the
+VMs using pktgen as packet generator tool. The largest total number of packets
+transmitted per second differs from each other, in which the smallest number of
+packets transmitted per second is 6 pps on Sep. 12ed and the largest of that is
+210.8 kpps. Meanwhile, the largest total number of packets received per second
+differs from each other, in which the smallest number of packets received per
+second is 2 pps on Sep. 13rd and the largest of that is 250.2 kpps.
+
+In some test runs when running with less than approx. 90000 flows the PPS
+throughput is normally flatter compared to when running with more flows, after
+which the PPS throughput decreases. For the other test runs there is however no
+significant change to the PPS throughput when the number of flows are
+increased. In some test runs the PPS is also greater with 1000000 flows
+compared to other test runs where the PPS result is less with only 2 flows.
+
+There are lost packets reported in most of the test runs. There is no observed
+correlation between the amount of flows and the amount of lost packets.
+The lost amount of packets normally differs a lot per test run.
+
+Detailed test results
+---------------------
+The scenario was run on Intel POD5_ with:
+Joid
+OpenStack Mitaka
+OpenVirtualSwitch 2.5.90
+OpenDayLight Beryllium
+
+Rationale for decisions
+-----------------------
+Pass
+
+Conclusions and recommendations
+-------------------------------
+Tests were successfully executed and metrics collected.
+No SLA was verified. To be decided on in next release of OPNFV.
+
+