Check if over-booking of CPU makes sense

[willirath]

In

https://github.com/ExaESM-WP4/dask-jobqueue-configs/blob/8318f739d97abc5064151121a2fd95c3120f4e6d/nesh/nesh-dask-jobqueue-config.yaml#L13

we might decide to go for cores higher than the requested CPU count and make use of multi threading of the CPUs. To test if this makes sense, run two cases:

1. Use default config with cores=4, draw, average, compute lots of random numbers with dask array (dask.array.random.normal(size=(100e9, 1), chunks=(100e6, 1)).mean().compute()), and check worker CPU usage in the dash board.

2. Rerun 1. but with cores=8 (not changing the job specs themselves), repeat same computation and see if you can substantially lower execution time. If so, we could / should have cores higher than cpunum_job.

[kathoef]

The behaviour might be different across the batch classes because of execution host hardware differences, e.g. CPUs for clexpress, clmedium, cllong, clbigmem, and feque are Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz and for clfo are Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz, respectively. It might be necessary to come up with batch class specific defaults. (Just as a quick note to myself.)

[willirath]

On neshcl218 (from the clbigmem):

$ lscpu
Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                32
On-line CPU(s) list:   0-31
Thread(s) per core:    1  # <-- Single threaded!
Core(s) per socket:    16
Socket(s):             2
NUMA node(s):          2
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 85
Model name:            Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz
Stepping:              4
CPU MHz:               2798.931
CPU max MHz:           3700.0000
CPU min MHz:           1000.0000
BogoMIPS:              4200.00
Virtualization:        VT-x
L1d cache:             32K
L1i cache:             32K
L2 cache:              1024K
L3 cache:              22528K
NUMA node0 CPU(s):     0-15
NUMA node1 CPU(s):     16-31
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 cdp_l3 intel_ppin intel_pt ssbd mba ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req pku ospke spec_ctrl intel_stibp flush_l1d

[kathoef]

Did this systematically on all the available execution hosts. Turns out that nodes associated with the clexpress, clmedium, cllong, clbigmem and clfo2 batch class all have multithreading disabled and we can/should keep the current default. (The host of the feque has multithreading, though.)

hosts=`qstat -E -F ehost | grep nesh`
for host in ${hosts}; do
 ssh -o StrictHostKeyChecking=no username@${host} -t "hostname; lscpu | grep -e 'CPU(s):' -e 'Thread(s) per core:' | grep -v NUMA" >> host_thread_info.txt
done

[kathoef]

Playing around with the above code it turned out that (arbitrary) overbooking of requested CPU resources is possible... We should therefore probably add a section in 02_customize_your_cluster.ipynb indicating that users should never use a Dask worker core number larger/different from the requested batch job CPU number. (It would actually be nice to come up with a way to automatically handle this, I think.)