ALPS — Acer AR585 F1 Cluster

ALPS is short for “Advanced Large-scale Parallel Supercluster” (also known as “御風者” in Chinese). It is a supercomputer that offers an aggregate performance of over 177 TFLOPS. The system uses the AMDR Opteron 6100 processors, and has a total of 8 compute clusters, 1 large memory cluster, and over 25,600 compute cores.

ALPS is designed to support a wide range of applications — so many that NCHC could only begin to fathom the jobs that could potentially be run across it. APLS is also designed to function as a test bed for new application design and research; support for open source and academic-specific applications is a must. To offer the most robust level of support, the system was first planned with every angle in mind — not simply raw compute performance, and not necessarily best performance per watt.

> Configuration

1. System in Brief

   Hostname:

   alps.nchc.org.tw

   Login Node:

   alps.nchc.org.tw

   Interactive Nodes :

   alpi1.nchc.org.tw (211.73.70.21)

   alpi2.nchc.org.tw (211.73.70.22)

   alpi3.nchc.org.tw (211.73.70.23)

   alpi4.nchc.org.tw (211.73.70.24)

   alpi5.nchc.org.tw (211.73.70.25)

   Directories:

   /home (209 TB), for user's home

   /pkg (16 TB), for package

   /work (162 TB), for working scratch

2. Hardware Specification

   System Family:

   Acer Group Cluster

   System Model:

   Acer AR585 F1 Cluster

   Processors:

   AMD Opteron 6174, 12 cores, 2.2GHz (compute nodes)

   AMD Opteron 6136, 8 cores, 2.4GHz (fat nodes)

   Main Memory (per node):

   128 GB (compute nodes)

   256 GB (fat nodes)

3. Software Stack

   Operating System:

   Novell SuSE Linux Enterprise 11 SP1

   Job Scheduler & Queuing System:

   Platform LSF (Load Sharing Facility) 7.06

   Parallel Filesystem:

   Lustre

   Development Tools:

   Compiler:

   Intel Cluster Toolkit, PGI CDK/Server, x86 Open64, GCC

   Debugger:

   Allinea DDT

   Libraries:

   MPI/OpenMP:

   Platform MPI (formerly HP-MPI), Intel MPI, PGI MPI/OpenMP, MVAPICH/MVAPICH2

   Math:

   Intel MKL (Math Kernel Library), AMD ACML (AMD Coe Math Library)

> Queuing System

NCHC users specify one of the following submit classes to queue jobs. Upon submission the job is routed to the appropriate LSF class according to the following criteria.

Queue Name	CPU(s)/job	Max CPUs/user	(CPU/Run)time limit/job
serial	1	20	12 days
12cpu	2 ~ 12	120	8 days
48cpu	13 ~ 48	240	6 days
192cpu	49 ~ 192	768	5 days
384cpu	193 ~ 384	768	3 days
768cpu	385 ~ 768	1536	2 days
short	128	128	2 days
medium	128	128	6 days
long	128	128	10 days

All batch jobs must be submitted to a valid submit class. If the class doesn“t exist, job will be terminated and an error message will be issued.