BandwidthSaturation - Revision history

Daniel-schurhoff-de23@rwth-aachen.de at 14:15, 3 September 2019

2019-09-03T14:15:51Z

Thomas-gruber-5537@uni-erlangen.de at 17:17, 12 March 2019

2019-03-12T17:17:20Z

Thomas-gruber-5537@uni-erlangen.de at 16:58, 12 March 2019

2019-03-12T16:58:51Z

Thomas-gruber-5537@uni-erlangen.de at 16:43, 12 March 2019

2019-03-12T16:43:46Z

Thomas-gruber-5537@uni-erlangen.de: Created page with "== Description == The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which da..."

2019-03-06T10:06:02Z

Created page with "== Description == The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which da..."

New page

== Description ==
The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which data path can be saturated. Inside CPU packages and whole compute nodes, the main resource for saturation is the memory subsystem but also the last level cache is a candidate. At cluster level, the shared data path is the interconnect network shared by all compute nodes.

== Symptoms ==
* Saturating speedup across cores sharing a data path

== Detection ==
At system level use a hardware-counter tool like:
* LIKWID with performance groups MEM and L3
* Same information can be provided by perf or PAPI
At cluster level you need special tools for the specific network technology

== Possible optimizations and/or fixes ==
* Reorganize access to data with methods like spacial or temporal blocking

← Older revision		Revision as of 14:15, 3 September 2019
Line 1:		Line 1:
		+	[[Category:Performance Pattern]]
	== Description ==		== Description ==
	The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which data path can be saturated. Inside CPU packages and whole compute nodes, the main resource for saturation is the memory subsystem but also the last level cache is a candidate. At cluster level, the shared data path is the interconnect network shared by all compute nodes.		The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which data path can be saturated. Inside CPU packages and whole compute nodes, the main resource for saturation is the memory subsystem but also the last level cache is a candidate. At cluster level, the shared data path is the interconnect network shared by all compute nodes.

@@ Line 9: / Line 9: @@
 == Detection ==
 === Node-level ===
-At system level use a hardware-counter tool like:
+In order to detect saturated bandwidth, you need some reference value. For memory bandwidth on node-level, a common benchmark is [https://www.cs.virginia.edu/stream/ STREAM].
   <nowiki>
 $ ./stream_c.exe
@@ Line 31: / Line 28: @@
   <nowiki>
 $ likwid-perfctr -C N:0-71 -g MEM -m ./jacobi N=10000 M=10000 iters=400
 +----------------------------------------+-------------+-----------+------------+-----------+
 |                 Metric                 |     Sum     |    Min    |     Max    |    Avg    |
@@ Line 50: / Line 48: @@
 === Cluster-level ===
-At cluster level you need special tools for the specific network technology
+At cluster level you need special tools for the specific network technology if you want to track saturation in your network.

@@ Line 1: / Line 1: @@
 == Description ==
 The pattern "Bandwidth Saturation" describes the performance limitation caused by fully utilizing a shared data path. It depends on the system level which data path can be saturated. Inside CPU packages and whole compute nodes, the main resource for saturation is the memory subsystem but also the last level cache is a candidate. At cluster level, the shared data path is the interconnect network shared by all compute nodes.
 == Symptoms ==

@@ Line 6: / Line 6: @@
 == Detection ==
 At system level use a hardware-counter tool like:
 * LIKWID with performance groups MEM and L3
 * Same information can be provided by perf or PAPI
 At cluster level you need special tools for the specific network technology
 == Possible optimizations and/or fixes ==
-* Reorganize access to data with methods like spacial or temporal blocking
+In general, it is beneficial to saturate the bandwidth as your code runs at the highest performance possible but maybe you can do better.