UM version4.5 benchmarks - Revision history

GethinWilliams at 14:36, 24 May 2013

2013-05-24T14:36:52Z

2012-12-17T15:45:43Z

2012-12-17T15:45:17Z

Preamble

2012-12-17T15:43:54Z

FAMOUS

2012-12-17T15:42:48Z

Intel Westmere

2012-12-17T15:41:17Z

2012-12-17T15:39:43Z

Preamble

2012-12-17T15:37:22Z

Intel SandyBridge

2012-12-17T15:37:05Z

Intel Westmere

2012-12-17T15:31:47Z

Intel SandyBridge

@@ Line 10: / Line 10: @@
-=Intel Westmere=
+=Emerald=
-* Emerald.  Intel E5649 (2.53GHz)
+* Intel Westmere E5649 (2.53GHz)
 * QDR Infiniband (non-RoCE)
 * GCOMv3.1
@@ Line 53: / Line 53: @@
-=Intel SandyBridge=
+=Intel SandyBridge Test System=
 * Test system: Quad socket, 8-core E-4650L (2.60GHz) (L for Low power)
@@ Line 93: / Line 93: @@
 |-
 || 8x4 || 32 || ~65
 |-
 |}

@@ Line 8: / Line 8: @@
 * Only the master process writes output, this can lead to load-balance issues, which hinder scaling.
 * Worst case message latencies for a cohort of processors are what matter for scaling.  The vast majority of messages are either ~100 bytes or ~1KB in size.  Latencies are reported for these key message sizes.
 =Intel Westmere=
@@ Line 50: / Line 51: @@
 |-
 |}
 =Intel SandyBridge=

@@ Line 7: / Line 7: @@
 * If the model is not filesystem-bound, it is often (MPI massage) latency-bound.
 * Only the master process writes output, this can lead to load-balance issues, which hinder scaling.
-* Worst case message latencies for a cohort of processors are what matter for scaling.
+* Worst case message latencies for a cohort of processors are what matter for scaling.  The vast majority of messages are either ~100 bytes or ~1KB in size.  Latencies are reported for these key message sizes.
 =Intel Westmere=

@@ Line 80: / Line 80: @@
 |}
-* The last line of this table shows a real problem scaling beyond 16 cores.  Load balance?
+* The last line of this table shows a real problem scaling beyond 16 cores.  Load balance?  (Latencies are '''much''' better than QDR IB.)
 * Would like to try to improve file writing performance and re-run.

@@ Line 21: / Line 21: @@
 ||  || 0 bytes || 128 bytes || 1024 bytes
 |-
-|| between sockets || ~2.0us || ~2.4us || ~4.7us
+|| between nodes || ~2.0us || ~2.4us || ~4.7us
 |-
 |}

@@ Line 4: / Line 4: @@
 =Preamble=
-* Cluster/Parallel file systems are often a bottleneck.
+* Cluster/Parallel file systems are often a bottleneck.  Timings are for writing to local disk, unless specified otherwise.
 * If the model is not filesystem-bound, it is often (MPI massage) latency-bound.
 * Only the master process writes output, this can lead to load-balance issues, which hinder scaling.

@@ Line 57: / Line 57: @@
 * 20MB L3 cache
 * GCOMv3.1
 {| border="1" cellpadding="10"
-!colspan=4|MPI message latency
+!colspan=4|IMB ping-pong message latency
 |-
 ||  || 0 bytes || 128 bytes || 1024 bytes