Merge pull request #15 from beorn-/doc

documentation: Add converted from pptx overview to documentation/
criteo · May 28, 2024 · 38b79ea · 38b79ea
2 parents 5a1cd3e + ec1a6ce
commit 38b79ea
Show file tree

Hide file tree

Showing 16 changed files with 157 additions and 0 deletions.
diff --git a/documentation/img/Hwbench-intro10.png b/documentation/img/Hwbench-intro10.png
diff --git a/documentation/img/Hwbench-intro12.png b/documentation/img/Hwbench-intro12.png
diff --git a/documentation/img/Hwbench-intro13.png b/documentation/img/Hwbench-intro13.png
diff --git a/documentation/img/Hwbench-intro14.png b/documentation/img/Hwbench-intro14.png
diff --git a/documentation/img/Hwbench-intro15.png b/documentation/img/Hwbench-intro15.png
diff --git a/documentation/img/Hwbench-intro16.png b/documentation/img/Hwbench-intro16.png
diff --git a/documentation/img/Hwbench-intro17.png b/documentation/img/Hwbench-intro17.png
diff --git a/documentation/img/Hwbench-intro18.png b/documentation/img/Hwbench-intro18.png
diff --git a/documentation/img/Hwbench-intro19.png b/documentation/img/Hwbench-intro19.png
diff --git a/documentation/img/Hwbench-intro20.png b/documentation/img/Hwbench-intro20.png
diff --git a/documentation/img/Hwbench-intro21.png b/documentation/img/Hwbench-intro21.png
diff --git a/documentation/img/Hwbench-intro22.png b/documentation/img/Hwbench-intro22.png
diff --git a/documentation/img/Hwbench-intro4.png b/documentation/img/Hwbench-intro4.png
diff --git a/documentation/img/Hwbench-intro5.png b/documentation/img/Hwbench-intro5.png
diff --git a/documentation/img/Hwbench-intro6.png b/documentation/img/Hwbench-intro6.png
diff --git a/documentation/overview.md b/documentation/overview.md
@@ -0,0 +1,157 @@
+# HWbench origins
+
+Hwbench the benchmark orchestrator was  initially written by the Criteo Hardware Team.
+
+# Specifications
+
+- Predictive
+- Modular
+- Automated
+- Hardware driven
+
+---
+
+We have not integrated the device in prod yet
+Siena is installed on a devbox that is not representative to a final server. Fan speeds and other chassis specific metrics will be ignored to focus on the processor itself.
+
+Micro benchmarks are used as a reference point for app performance with our internal customers.
+
+Time-limited evaluation
+
+# Scripted
+
+![](img/Hwbench-intro4.png)
+
+
+
+
+- __runtime__ : absolute time in seconds per executed benchmark
+
+- __monitor__ : enable system monitoring via BMC \(redfish\)\, system \(turbostat\, sysfs\, …\)
+
+    - __Objects__:
+        - _Thermal_
+        - _Fan_
+        - _PowerConsumption_
+        - _PowerSupplies_
+        - _Frequencies_
+
+    - __Supports__: Generic redfish \+ OEM \(HPE / DELL\)
+
+- __engine__ : micro\-benchmark\, hwbench plugin to wrap features & command lines
+
+- __hcc__ : selects cores from the system to be eligible for benchmarking\.
+
+    - __Objects__:
+      - Logical cores \( _\<int>_ \)
+      - Physical cores \( _CORE\<int>\)_
+      - NUMA \( _NUMA\<int>_ \)
+      - Quadrant\( _QUADRANT\<int>_ \)
+      - _Simple_  \(1\, 2\, 4\, 8\, 16\, 32\.\.\.\)
+
+    Usage: interval \(0\-63\)\, list \(1\,31\)
+
+    Example: CORE0\, CORE1\, QUADRANT0\, QUADRANT0\-3
+
+- __hccs__ : selects how hosting cpu cores are associated with jobs
+
+    - __Objects__:
+      - _none _ \(all cores at once\)\
+      - _iterate _ \(each hcc value will create a new job\)
+
+- __stressor\_range__ : selects how many engine instance per job
+
+    - __Objects__:
+      - \<int>\
+      - interval \(\<x\-y>\)
+      - list \(\<x>\,\<y>\)\
+      - _auto _ \(same value as hccs\)
+
+- __\[<__  _job\_name_  _>_  __\]__ : defines one top level job
+
+- __engine\_module__ : what module from selected  _engine_
+
+- __engine\_module\_parameter__ : a list of  _engine\_module_  tests to execute
+
+# Execution (example with AMD 8434P)
+
+Using this config:
+
+![](img/Hwbench-intro5.png)
+
+Output:
+
+![](img/Hwbench-intro6.png)
+
+
+collected info:
+* System info
+* Performance
+* Monitoring
+
+
+# Render (example with AMD 8434P)
+
+Compare & graph several traces file
+
+![](img/Hwbench-intro10.png)
+
+__The following graphs are:__
+
+a  _very small_  subset of the complete graph series
+
+fully automatically generated
+
+# Same chassis graph
+
+![](img/Hwbench-intro12.png)
+
+# Individual Fan behavior
+
+![](img/Hwbench-intro13.png)
+
+# All fans vs power
+
+![](img/Hwbench-intro14.png)
+
+# All fans vs thermal
+
+![](img/Hwbench-intro15.png)
+
+# Compared max perf
+
+![](img/Hwbench-intro16.png)
+
+# Compared max perf per core
+
+![](img/Hwbench-intro17.png)
+
+# Compared max perf per watt
+
+![](img/Hwbench-intro18.png)
+
+# Compared max perf per core per watt
+
+![](img/Hwbench-intro19.png)
+
+# Compared scaling performance
+
+![](img/Hwbench-intro20.png)
+
+# Compared scaling performance per watt
+
+![](img/Hwbench-intro21.png)
+
+# Custom spike engine to study fans
+
+![](img/Hwbench-intro22.png)
+
+# What's next ?
+
+__In random order:__
+
+- Add more metrics: PDU & other power sensors\, thermal sensors
+- Add more engines: storage\, applicative sample loads
+- Add parallelism to combine several engines: i\.e\.:  _stress\-ng_  \+  _fio_
+- Add more graph types?
+- Get contributions\, ideas?