IT merger tests
Introduction
We're running some tests to examine the various benefits of Meteorology and ESSC computer systems.
To make these tests I'm using a C++ program which I've written. Argo data is series of robotic probes that float with the Ocean currents and taken measurements of pressure, temperature and salinity at they decend and rise through the sea. Typically, every 10 days they re-surface and transmit data. For a given argo ID (one robotic probe) my program extracts data for this robotic device from many netCDF files and writes the data into one file. This test will use this program to look over almost all the argo IDs for 2010. The main tasks of this program are the read in of data and then writing out this data again in a combined file.
Timings
Machine | Description | Real | User | Sys |
ESSC Rocks compute cluster | 1.5 years old, started at 5:20pm on Tues 6 Dec (apparently, some of the Gluster performance features were turned off, same run below with performance issues on) | 4h21m8.18s | 5m59.63s | 2m0.18s |
ESSC Rocks compute cluster via GlusterFS client | Same run as above, with Gluster performance features on. Run by Dan over Christmas break | 3h15m | ||
ESSC Rocks compute cluster via Gluster-NFS | Run by Dan over Christmas break | 45m | ||
ESSC Rocks compute cluster via Gluster-NFS in April 2012 | Same as run above but by Marc at 6:30pm on 16 April 2012 | ? | ||
ESSC Rocks cluster and single server (triton) via convential NFS | Run by Dan over Christmas break. Load on triton was very low, because nothing else was accessing it at the time. | 15m | ||
master | Little newer than 2 years, 16Gb memory, 8 cores (2.13Ghz), running on local disk, started at 5:20pm on Tues 6 Dec | 12m14.95 | 7m17.81s | 0m46.22s |
niagara | 6 months old, 3.46Ghz, running on local disk, started at 5:20pm on Tues 6 Dec | 4m17.67s | 3m49.67s | 0m26.64s |
pilvista | About 2 years old, 24Gb memory, 6 cores (2.00Ghz), running on /export/radar-nas1, started at 10:09 on Wed 7 Dec, load of 17 | 22m15.72s | 9m45.48s | 2m35.38s |
pilvista in April 2012 | Same run as above but at 6pm on 16 April 2012, with load of 1 | 21m44.56s | 11m18.19s | 2m17.69s |
pilvista | Same as above, but running two years consequtively (2009 as well as 2010), load is 17 again before running these runs, started at 9:48 on Thurs 8 Dec | 28m20.43s and 29m6.38s | 7m53.90s and 9m15.66s | 2m49.90s and 2m45.27s |