Software Linux Cluster
HPCVL is operating a cluster of X86 based multicore machines running Linux.This page explains essential features of this cluster and is meant as a basic guide for its usage.
What's the Hardware ?
Our cluster consists of multiple X86 multicore nodes made by Sun Microsystems (AMD based), Dell (Intel x5670 based), and IBM (also based on Intel x5670). All nodes run CentOS Linux 5.8 and share a file system. Access is handled by Grid Engine. The server nodes are called sw0001...sw0050.
Presently, the workup node of the HPCVL "Software Cluster" is swlogin1. This is a Sun Fire X4440 Server with four sockets based on the 6-core AMD Opteron 8431 chip. These chips run at 2.4 GHz. The total physical memory of the node is 80 GB.
Most of the nodes of the SW cluster (sw0011-40) are Dell PowerEdge R410 Servers that have 2 sockets with a 6-core Intel® Xeon® processor (Intel x5670 / x5675) that runs at 2.9/3.07 GHz. These nodes offer a total of 12 cores that are 2-fold hyperthreaded, i.e. they support up to 24 threads. The scheduler is configured such that only 12 threads are executed at a time. These nodes have 64 Gbyte (sw0011-22, sw0035-40) or 32 Gbyte (sw0023-34) of physical memory.
A few nodes (sw0041-51) are IBM XServers 3850-X5 that are also based on the Intel® Xeon® processor (Intel E7-4860). These servers are -socket with cores per chip for a total of 40 cores per node and support for up to 80 threads (hyperthreading). The clock speed for these machines is 2.27GHz. Two of these servers (sw0050-51) have a 1 TB of physical memory, the others have 256 GB.
Why these Systems ?
The main emphasis in these systems is a high floating-point performance for a modest number of processes / threads. Since commercial software such as Fluent and Abaqus is increasingly focussed on support for Linux only, this cluster was acquired to continue to offer recent versions of these software packages. In addition, the higher single-core performance of these nodes (compared to the Sparc/Solaris based M9000 cluster, for instance), allows for a more efficient use of license seats which usually a priced per-core.
Who Should Use This Cluster ?
The software cluster runs on the Linux operating system, and should therefore only be used if the software cannot be compiled or run on the Sparc/Solaris platform. Runs that require more than 64 Gbyte of memory should be performed on the M9000 cluster unless the program is parallelized using MPI with distributed memory and very low communication requirements.
We suggest you consider using this compute cluster if
- Your application is very floating-point intensive and has little else to do, and you need modest amounts of memory. With few exceptions no more than 64 Gbyte are available in the form of shared memory.
- Your application is commercial or public-domain software that supports only Linux or poses considerable challenges to port to a Solaris platform.
- Your application is either explicitly parallel (for instance, using MPI) and has very low communication requirements, or is multi-threaded with a small number (typically no more than 12) of scaling threads.
- Your application uses a commercial license that is scaled per process; in such cases it is favourable to use machines with the maximum per-process performance.
The new cluster might not be suitable if
- You need to perform a large number of relatively short jobs, each serial or with very few threads. Jobs like this should be sent to the "Victoria Falls" cluster.
- Your application is memory intensive and/or compiles and runs well on the Solaris/Sparc platform. Such jobs should be sent to the default M9000 cluster.
- Your application is required to scale to a very large number of processes in a distributed-memory fashion and is communication intensive. Such jobs require a fast interconnect (Infiniband or similar) and should be run on a different system, for instance other Compute Canada installations.
If you think your application could run more efficiently on these machines, please contact us (firstname.lastname@example.org) to discuss any concerns and let us assist you in getting started.
Note that on these cluster (as on the M9000's), we have to enforce dedicated cores or CPUs to avoid sharing and context switching overheads. No "overloading" can be allowed.
How Do I Use This Cluster ?
... to access
The HPCVL Secure Portal at https://portal.hpcvl.queensu.ca offers a direct link called xterm (swlogin1). This link connect via a terminal to swlogin1 which is designated as a login/workup node for the cluster. If encounter issues with the portal login please let us know. Meanwhile, it is possible to "ssh" directly from sflogin0 to swlogin1 by typing ssh -X swlogin1 and re-typing your system password.
The file systems for all of our clusters are shared, so you will be using the same home directory as when you are using the M9000 servers or the standard login node sfnode0. swlogin1 node can be used for compilation, program development, and testing only, not for production jobs.
... to compile and link
Since the SW cluster has a completely different architecture than the M9000 Servers code must be re-compiled when migrating to this cluster. The compiler that we are using on this cluster is the Intel Compiler Suite. This includes compilers for Fortran, C, and C++, as well as MPI and OpenMP support, debuggers and development suite. This software resides in /opt/ics and is only visible to the Linux cluster. The versions are:
- Fortran (ifort): Intel(R) Fortran Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 12.1 Build 20110811
- C (icc): Intel(R) C Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 12.1 Build 20110811
- C++ (icpc): Intel(R) C++ Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 12.1 Build 20110811
We also supply the gnu C/C++/Fortran compilers (version 4.1.2 20080704 (Red Hat 4.1.2-52)) to enable compilation of some of the more "temperamental" public-domain software, but it is preferable to use the Intel suite if possible.
For applications that can not be re-compiled (for instance, because the source code is not accessible), a pre-compilerd Linux version (x64 for Redhat will do the trick) needs to be obtained.
... to run jobs
As mentioned earlier, program runs for user and application software on the login node are allowed only for test purposes or if interactive use is unavoidable. In the latter case, please get in touch to let us know what you need. Pruduction jobs must be submitted through the Grid Engine load scheduler. For a description of how to use Grid Engine, see the HPCVL GridEngine FAQ.
Grid Engine will schedule jobs to a default pool of machines unless otherwise stated. This default pool contains presently only the M9000 nodes m9k0001-8. Therefore, you need to add the following two lines to your script for your job to be scheduled to the Linux SW cluster exclusively:
#$ -q abaqus.q
#$ -l qname=abaqus.q
The abaqus name for the queue that is added here derives from the initial software Abaqus that was (and still is) run on this cluster.
Note that your jobs will run on dedicated threads, i.e. typically up to 12 processes can be scheduled to a single node. The Grid Engine will do the scheduling, i.e. there is no way for the user to determine which processes run on which cores.
...to find more information
General information about using HPCVL facilities can be found in our FAQ pages. We also supply user support (please contact us at email@example.com), so if you experience problems, we can assist you.