Cipres Science Gateway

 

MAFFT Benchmarks:

The Table below shows the speedup obtained for MAFFT as additional processors are brought into play on Trestles. Rationale: All jobs are run on 32 processors, with the goal of giving the best possible wall clock time. Although there is some loss of efficiency, the goal for MAFFT is to achieve maximum wall time performance for the user. For this reason MAFFT is always run on a single dedicated node (32 processors). This is possible because relatively few MAFFT jobs are run on the Gateway, and these jobs are not heroic in nature, . If submissions (or the magnitude of individual submissions) increase, it may be necessary to increase the efficiency by running the jobs over fewer processors..

MAFFT-6.843 results for Test Data
     
Problem
OpenMP Threads
Fast Wall Time (s)
Slow Wall Time (s)
Speedup
Efficiency
Test Data
1
137
1.00
1.00
linsi
2
75
82
1.83
0.91
4
41
46
3.34
0.84
8
20
44
6.85
0.86
16
17
27
8.06
0.50
32
13
17
10.54
0.33
Trestles info:
2.4-GHz AMD Magny-Cours processors
Quad-socket, oct-core (32-way) nodes
QDR InfiniBand interconnect
GNU 4.1.2 compiler
 -O3

 

 

 

  
Return to Login Page  

If there is a tool or a feature you need, please let us know.