Development for LHC@home

Here is the new benchmarking application that can used to measure the performance of your machine. It runs the ATLAS Kit Validation which is what has recently been used as a benchmark for the cloud procurement activities. It supports multi-core and essentially the faster it runs the better. The results are sent to our servers and we will try to expose these to you at a later date.

Thanks, Laurence.

However, i fail to see a benchmarking tool of any kind, just a report about it.

Not, where it is, which one to run and how.

It is a new application in this project. Take a look at my results.

It is a new application in this project. Take a look at my results.

No access to the last link. This one works.

Didn't you compress the Benchmark_2016_08_05.vdi file (869856,94 K)? It was! the uncompressed file was 3575808kB.
I ran 1 task. http://lhcathomedev.cern.ch/vLHCathome-dev/result.php?resultid=232381 It created a VM with 8 processors, 1024MB RAM and swap 1024MB too and all swap is used.

The memory scaling with cores hasn't been configured. After experimenting, we should know what it should be set to.

How can you have run time less that CPU time! Result 232372.

How can you have run time less that CPU time! Result 232372.

you = Steve Hawker* (owner of that task)

His VM has 4 cores, so in theory the cpu time could be 4 times the elapsed time.

Here is an example of the message that we get back after running the benchmark.

{"message":{"_timestamp":"2016-08-05T13:37:59Z","_id":"2-501_2016-08-05T13:37:59Z","profiles":{"rkv":{"evt":{"speed_real":["0.948"],"speed_cpu":["0.992"],"cpu_values":[1.00791],"real_values":[1.05503],"entries":99},"initialization":{"cpu_values":[96.759],"real_values":[332.8],"entries":1},"finalization":{"cpu_values":[0.013],"real_values":[0],"entries":1}},"kv":{"start":"1470397080","thr_1":{"finalization":{"real":{"unit":"s","value":0},"vmem":{"unit":"MB","value":1355.043},"cpu":{"unit":"s","value":0.013},"entries":1},"initialization":{"real":{"unit":"s","value":332.8},"vmem":{"unit":"MB","value":1291.977},"cpu":{"unit":"s","value":96.759},"entries":1},"evt":{"real":{"unit":"s","value":1.05503,"error":0.44877},"vmem":{"unit":"MB","value":1353.246},"cpu":{"unit":"s","value":1.00791,"error":0.44521},"entries":99},"vmem-leak":{"evt 2-20":{"unit":"kb/evt","value":0},"evt 21-50":{"unit":"kb/evt","value":0},"first-evt":{"unit":"Mb","value":1353.246},"last -evt":{"unit":"Mb","value":1353.246},"evt 51+":{"unit":"kb/evt","value":0},"10th -evt":{"unit":"Mb","value":1353.246}},"memory":{"vm_exe":{"unit":"kB","value":1400},"VmHWM":{"unit":"kB","value":1016876},"VmSwap":{"unit":"kB","value":1387568},"VmPeak":{"unit":"kB","value":2696},"VmPTE":{"unit":"kB","value":32},"VmSize":{"unit":"kB","value":0},"VmLck":{"unit":"kB","value":0},"vm_data":{"unit":"kB","value":899564},"VmLib":{"unit":"kB","value":215624},"VmStk":{"unit":"kB","value":1016800},"VmRSS":{"unit":"kB","value":1387568}}},"kv_tag":"KV-Bmk-BOINC","end":"1470397697"}},"metadata":{"benchmark_target":"machine","meminfo":2044680,"UID":"2-501","classification":"i6_1_f6m42s7_mhz2494.316","freetext":"Dev","cpuname":"Intel(R) Core(TM) i5-2520M CPU @ 2.50GHz","ip":"10.0.2.15","osdist":"Scientific Linux release 6.7 (Carbon)","bogomips":4988.63,"VO":null,"cpunum":1,"pnode":"","mp_num":"1","pyver":"2.6.6","cloud":"BOINC"}},"@version":"1","@timestamp":"2016-08-05T11:48:30.616Z"}

I got one, only seems to use about 25% on my 10core 20 thread machine.

It seems, it uses whatever memory is allocated plus some swap.

It seems, I've had two crashes out of two so far...

---

Me too. But all multicore tasks fail on my Opteron 1210 running SuSE Linux because it has only 2 cores, while CMS tasks run nicely. Also my 32-bit Linux laptop never fails and is contributing to MCPLots while the 84554 Windows 10 PC wth its AMD A10-6700 and 4 cores is not detected.
Tullio

All Benchmark Application tasks for my computer (ID 37) are succeeding.
I changed the configuration, so that only single core VM's are created.
I advise to increase the RAM to 1408MB for a single core to reduce the swapping to a minimum.

ivan wrote:

It seems, I've had two crashes out of two so far...

On both machines:
2016-08-06 10:45:18 (26871): VM Heartbeat file specified, but missing.

I got one that went through OK, with 20cores

ivan wrote:

It seems, I've had two crashes out of two so far...

On both machines:
2016-08-06 10:45:18 (26871): VM Heartbeat file specified, but missing.

I'm getting that on some Theory tasks as well, usually after quite a few successful jobs. Then that happens and the task is written off, with zero credit. :-(

---

ivan wrote:

It seems, I've had two crashes out of two so far...

On both machines:
2016-08-06 10:45:18 (26871): VM Heartbeat file specified, but missing.

I'm getting that on some Theory tasks as well, usually after quite a few successful jobs. Then that happens and the task is written off, with zero credit. :-(

You got it with a benchmark job too. You probably noticed.
The 'heartbeat' file is written every minute by the VM into the shared folder in BOINC slots directory belonging to the task.
The minimum heartbeat interval is set to 20 minutes.
If the task is running and the heartbeat is not refreshed within 20 minutes, the VM is treated as dead, erroring out and cleaned.

As far as I noticed I've seen this happening only on Linux hosts.
Should be investigated.

If the task is running and the heartbeat is not refreshed within 20 minutes, the VM is treated as dead, erroring out and cleaned.

Maybe in that case, there should be a secondary check(of a different kind), if the VM is running, before terminating the task.

The heartbeat mechanism, by itself, does not seem to be reliable enough.



EDIT: Maybe it is a caching issue. The file length or content is not changed, so it might not always be updated.

Here is a plot with some initial benchmarking results. It shows events per second against processors ordered by rank. You may be able to find your machines there if you know your CPU. I will try to log the interesting values in the stderr_txt.

I have the impression due to short run time and low cpu-usage, that the current VM's don't get a benchmark job.

http://lhcathomedev.cern.ch/vLHCathome-dev/results.php?userid=38&offset=0&show_names=0&state=4&appid=8