Message boards :
Benchmark Application :
Something For The Weekend
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
Today the jobs are running normal. Using more CPU and an athena.py process up to 99%. |
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
I noticed that the memory size for the created VM is not obeying the setting from the Benchmark_2016_08_05.xml file. The VM only gets 730 MB <vbox_job> <script/> <os_name>Linux26_64</os_name> <memory_size_mb>1024</memory_size_mb> <enable_network/> <enable_remotedesktop/> <copy_to_shared>init_data.xml</copy_to_shared> <completion_trigger_file>shutdown</completion_trigger_file> <enable_shared_directory/> <pf_host_port>7859</pf_host_port> <pf_guest_port>80</pf_guest_port> <job_duration>1800</job_duration> <enable_vm_savestate_usage/> <disable_automatic_checkpoints/> <heartbeat_filename>heartbeat</heartbeat_filename> <minimum_heartbeat_interval>1200</minimum_heartbeat_interval> </vbox_job> |
Send message Joined: 12 Sep 14 Posts: 1067 Credit: 329,589 RAC: 72 |
Every VM has now been instrumented to run the DB12 benchmark when it starts. This means that you should see the following two new lines in the task output for the Theory, CMS and LHCb applications. Guest Log: [INFO] Running the fast benchmark. Guest Log: [INFO] Machine performance xx.yy HEPSEC06 This makes this Benchmarking application redundant. |
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
... This means that you should see the following two new lines in the task output for the Theory, CMS and LHCb applications. I tested this on an i7 2600 with 1 core for the Theory-VM and the other 7 cores 100% busy with SixTrack, Climate and Microbiome Immunity. 2017-09-25 18:51:31 (13896): Guest Log: [INFO] Running the fast benchmark. 2017-09-25 18:53:19 (13896): Guest Log: [INFO] Machine performance 4.82 HEPSEC06 |
Send message Joined: 16 Aug 15 Posts: 966 Credit: 1,211,816 RAC: 0 |
How does this new benchmark affect the computation other than adding a minute or so to the startup time? |
Send message Joined: 12 Sep 14 Posts: 1067 Credit: 329,589 RAC: 72 |
It does not affect the computing but helps to understand and optimize the system. |
Send message Joined: 12 Sep 14 Posts: 1067 Credit: 329,589 RAC: 72 |
The average performance should be approximately 9.5 HEPSPEC06 per core. Here is a chart comparing some recent HEPSPEC06 measurements to the Dhrystone benchmark values that the BOINC client calculated for the host. Your host was measured to be 12.1 GIOPS so I would expect your HEPSPEC06 value to be higher. |
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
...Your host was measured to be 12.1 GIOPS so I would expect your HEPSPEC06 value to be higher. When BOINC's benchmark is measured all tasks are suspended, so I repeated a Theory task with 2-4 cores (out of 8) busy and now the machine performance is 11.64 HEPSEC06. Quite a difference with previously measured value of 4.82 HEPSEC06 where the Host-CPU had a 100% load. |
Send message Joined: 12 Sep 14 Posts: 1067 Credit: 329,589 RAC: 72 |
This suggests something is not optimized when the machine is busy. I now see an advantage of keeping the Benchmarking app as each task is short. You may want to run 4 Benchmarking tasks in parallel and see what the HEPSPEC value is. |
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
I could try that and I'm prepared - all other tasks suspended, no dev-tasks yet on my machine, but although Benchmark Application 218 Unsent tasks, I don't get 4 workunits. I get nothing. 5 times: 4954 lhcathome-dev 27 Sep 10:49:35 Scheduler request completed: got 0 new tasks 4955 lhcathome-dev 27 Sep 10:49:35 No tasks sent 4956 lhcathome-dev 27 Sep 10:49:35 No tasks are available for Benchmark Application |
Send message Joined: 13 Feb 15 Posts: 1185 Credit: 850,190 RAC: 1,017 |
I could not get any Benchmark task, so I tested with 4 concurrently running Theory tasks on an else idle system. Of course it's difficult to run the tests at exactly the same time. The results: 2017-09-27 16:30:48 (13616): Guest Log: [INFO] Running the fast benchmark. 2017-09-27 16:30:56 (14108): Guest Log: [INFO] Running the fast benchmark. 2017-09-27 16:31:02 (13712): Guest Log: [INFO] Running the fast benchmark. 2017-09-27 16:31:23 (17144): Guest Log: [INFO] Running the fast benchmark. 2017-09-27 16:31:47 (13616): Guest Log: [INFO] Machine performance 7.93 HEPSEC06 2017-09-27 16:31:54 (14108): Guest Log: [INFO] Machine performance 8.23 HEPSEC06 2017-09-27 16:32:04 (13712): Guest Log: [INFO] Machine performance 8.18 HEPSEC06 2017-09-27 16:32:27 (17144): Guest Log: [INFO] Machine performance 8.07 HEPSEC06 When running 1 VM the machine performance is 12.33 HEPSEC06 |
©2024 CERN