Release notes. What’s new in MC#.MetaCluster v.0.30

You can download the latest version of MC#.MetaCluster Programming System

This is our first public release of MC#.MetaCluster system. This system is intended for those of you who want to run MC# programs in distributed mode on several Linux clusters. I.e. you can think that MetaCluster is the next level of abstraction:

Single Thread, One processor => Multiple Threads, One processor => Cluster => Several Clusters (MetaCluster) => GRID-network

Note that this system is much more complicated if you compare it to MC#.Cluster system and it is strongly recommended to use it only when you can’t achieve necessary performance within one cluster.

Please, read ReadMe file for installation instructions and other details.

We made measurements for couple of examples from HPCChallenge Class 2 Specification. Here they are…

EPStream triad

Source code: EPStream.mcs

In these measurements we used:

- 12 physical nodes of SKIF cluster with 2 processors AMD Athlon(TM) MP 1800+ each (1533.431 MHz).

- 4 physical nodes of ZOOT cluster (students’ cluster) with 2 processors Pentium III (Coppermine) each (601.371 MHz)

Gb/sec		Processors on ZOOT
Gb/sec		0	2	4	6	8
Processors on SKIF	0	-	0.48	0.86	1.47	1.90
	2	1.27	1.68	1.89	2.64	3.14
	4	2.36	3.09	3.43	3.68	4.09
	6	3.67	4.13	4.89	5.13	5.31
	8	4.57	5.31	5.35	6.21	6.30
	10	6.13	6.83	6.93	7.30	7.91
	12	7.13	7.44	7.67	8.26	8.83
	14	8.44	8.94	9.23	9.03	10.14
	16	9.33	10.00	10.33	10.59	11.60
	18	10.72	10.98	11.13	11.80	12.08
	20	11.14	12.21	12.61	12.87	13.21
	22	12.95	13.57	13.76	13.54	14.52
	24	14.01	14.65	15.30	15.42	16.24

GlobalFFT

Source code: GlobalFFT.mcs

In this example we used one physical cluster with 16 physical nodes and separated it into two clusters with 8 nodes each:

- Cluster 1: node-1x + node-2x, where x =[1..4]

- Cluster 2: node-3x + node-4x, where x=[1..4]

Each node had two processors AMD Athlon(TM) MP 1800+ each (1533.431 MHz)

mono GlobalFFT.exe 65536 /cluster $MCSGPATH/bin/nodes /withlog

Sec.		Cluster 1 (processors)
Sec.		0	2	4	6	8	10	12	14	16
Cluster 2 (processors)	0	-	388.86	192.51	129.00	96.83	78.32	65.69	56.88	50.01
	2	388.81	192.66	129.41	97.15	77.70	62.40	56.65	50.12	44.88
	4	193.05	129.04	97.34	78.15	65.18	56.51	50.10	45.14	40.83
	6	128.86	97.13	78.34	65.55	56.41	45.02	45.16	41.01	37.76
	8	97.22	78.14	65.78	56.65	49.91	41.13	40.93	37.49	34.92
	10	78.17	65.57	56.92	50.22	45.45	38.16	37.71	35.06	32.87
	12	65.75	56.79	50.32	44.92	40.64	35.33	34.94	33.07	31.00
	14	56.79	50.10	45.06	40.90	37.66	33.44	32.74	31.04	29.42
	16	50.15	44.88	40.91	37.73	35.16	32.89	31.07	29.32	28.19

You can download the latest version of MC#.MetaCluster Programming System

<<< here >>>