Difference between revisions of "HPC engineering TODO List"
Jump to navigation
Jump to search
| Line 50: | Line 50: | ||
== Completed section == | == Completed section == | ||
* PCM cluster | * PCM cluster | ||
| − | ** | + | ** Certified |
* K20 | * K20 | ||
| − | ** | + | ** Up and Running |
| − | |||
| − | |||
| − | |||
| − | |||
* Rocks Cluster | * Rocks Cluster | ||
** Get ACML installed in /share/apps/acml/5.x | ** Get ACML installed in /share/apps/acml/5.x | ||
| + | ** AMD System added to cluster | ||
| + | |||
| + | *Provisioning tests | ||
| + | ** PCM to Node downstairs - OK | ||
| + | ** Rocks to Node in Lab - OK | ||
Revision as of 16:00, 22 October 2012
Benchmarking
- Openfoam benchmarking (ARM vs Sandy bridge)#
- Single node tests
- OpenMPI for full 12/48 calx nodes
- Blender, open source rendering. To get a feel for ARM vs Intel on media projects
- jbench (see what the JVM performance is like on a system, ejre vs openjdk)
- Itaru (BWA genome stuff) pBWA doesnt scale (needs 12 more systems)
- justonedb guys, mysql vs postgres vs justonedb with their data sets (bloom analytics software) Peter
- spec cpu and hepspec on arm
- DP to email spec
- openmpi 1.6 results for all tests
- JH IMB tests, pingpong np=2, all tests 2 nodes, 4 nodes, 8 nodes
- linpack HPL
Software testing
- Hadoop terasort on v1.0.3 apache
- Install / tests CDH (Cloudera distribution of hadoop)
- JH Intel e3-1220L-v2 ivy bridge, sysbench power/performance tests
- JH cxmanage on the headnode, and on wiki
- VDI and spice for ubuntu / fedora, test on ARM, instructions on wiki
- Openstack, newest release with dashboard
- ScaleIO to port software to ARM (source and rebuild)
- Zettaset
- DP put in touch with UK partner #
- Ubuntu stuff to do:
- MaaS
- JuJu
- Landscape
- apt-proxy
Hardware Bits
- Server range for Viglen
- Stage 1: Hardware on SB similar to current range
- Stage 2: Additional storage products for schools (cheap NAS / iSCSI solution) to compete against hitachi
- Stage 3: Solutions, cloud, vmware?
Mike TODO List
- Benchmarks
- Get used to building codes, here are some of the standard ones we use as part of HPC testing
- HPL (Linpack)
- Stream (memory bandwidth)
- IMB (Network throughput and latency)
- Iozone (disk performance)
- Give them a try when you get a chance
- ARM Benchmarks
- Oil and Gas: Seismic processing benchmarks: http://geocomputing.narod.ru/benchmark.html (give a go on an ARM box and intel box)
Completed section
- PCM cluster
- Certified
- K20
- Up and Running
- Rocks Cluster
- Get ACML installed in /share/apps/acml/5.x
- AMD System added to cluster
- Provisioning tests
- PCM to Node downstairs - OK
- Rocks to Node in Lab - OK