Computing Resources at the SSCC

Last Revised:8/21/2008

The SSCC makes a tremendous amount of computing power available to its members. This publication will help you find the computing resource that can do the job you need done. The first part is an overview of the SSCC's servers. But if you have a particular job in mind already, you may want to skip straight to Which Server Should I Use?

Winstat

Winstat is the SSCC's Windows Terminal Server farm. Windows Terminal Servers allow you to log in to a Windows server remotely with full access to both the SSCC network and the software installed on the server. Software installed on the Winstats include general purpose statistical software like Stata, SAS and SPSS, special purpose statistical software like HLM and NVivo, and general Windows programs like Microsoft Office, Adobe Acrobat, and Macromedia Dreamweaver. Logging in to Winstat is also the easiest way to transfer files between the SSCC and home or other locations.

The Winstat servers all have two Dual Core Intel Xeon processors (for a total of four physical processors) running at 2.0GHz and 4 gigabytes of RAM. However, the inherent limitations of 32-bit operating systems mean that at best 3 gigabytes can be allocated to any particular task. In reality it can be much less. Stata, for example, can claim only about 900 megabytes of memory on the Winstats.

Those looking to run long statistical jobs should be aware that "idle" sessions are terminated after 24 hours. Idleness is defined in terms of user activity, not processes, so you can't simply start a job and then let it run. If you need to run jobs longer than 24 hours, you should use our Linux servers or reserve a Lab PC.

For more information about Winstat, see Using Winstat, or simply log in.

Lab PCs

The SSCC's computer labs have more than 50 powerful PCs with a variety of statistical software installed, including most of what's on the Winstats and some additional programs. They also have client programs installed for accessing all the SSCC servers. This makes the labs an excellent place to do research.

Two PCs in the Group Projects Room inside the 3218 lab can be reserved for long-running jobs, or if you need to run more jobs overnight or over the weekend, talk to the consultant about using more of the PCs in 3218. While the Linux servers are a more convenient place for running such jobs, these PCs are the best place for running programs that are not available on Linux.

For more information about the SSCC's computer labs, see SSCC Computer Labs and Information for SSCC Instructional Lab Users.

Kite and Hal

Kite and Hal are the SSCC's primary Linux servers. Both have a variety of statistical software installed, including Stata/MP, SAS and R. They also have the usual Linux utilities and compilers.

Both have two 3.0GHz Quad Core Intel Xeon processors (for a total of eight physical processors) and 10 gigabytes of RAM. Kite and Hal both run 32-bit Linux and have the same 3 gigabyte per job limit as Winstat, but they usually get much closer to that limit. Stata can claim about 2.8 gigabytes of memory on Hal and Kite. Kite and Hal are much better than Winstat for computationally-intensive jobs.

Stata/MP is the multiprocessor version of Stata, which means it will use two processors to run a single job. This makes it much faster than the standard Stata installed on Winstat.

For more information about using Kite and Hal, see Connecting to SSCC Linux Computers and/or the SSCC Linux Handbook.

Falcon

Falcon is a 64-bit Linux server for jobs that require more memory than 32-bit operating systems can provide, or programs that require 64-bit Linux. Installed software includes Stata/MP, R and SPSS. Falcon has the same CPUs as Hal and Kite (two 3.0GHz Quad Core Intel Xeon processors) and will not run most jobs any faster than they will. However it has 14 gigabytes of RAM and can use temp space on disk as virtual memory if needed. Stata jobs can claim nearly 29 gigabytes of memory on Falcon. Falcon should be reserved for jobs that cannot be run anywhere else.

For more information about using Falcon, see Connecting to SSCC Linux Computers and/or the SSCC Linux Handbook.

Condor

Condor is a set of 20 virtual machines running 32-bit Linux. It runs as a sort of batch system: jobs are submitted to Condor, and then assigned to the first available machine. The job then gets exclusive use of that machine until the job is done. Condor cannot be used for interactive jobs, but it's the ideal place for very long jobs that will run for days or weeks. It's also the best place for multiple jobs: SSCC members are welcome to run up to six jobs at a time on Condor.

Four of the virtual machines have 3.0GHz Intel Xeon processors similar to Hal or Kite, with 4 gigaytes of RAM each. The others have Intel Xeon processors running at 3.4GHz, but they are of an older generation so that gives them performance comparable to more recent processors running at about 2GHz. They have 2 gigabytes of RAM each. Four of these virtual machines are set up with two processors and run Stata/MP, the others have one processor and run standard Stata. The Condor program itself takes up some RAM, so Stata jobs sent to Condor can only use about 2 gigabytes of memory.

Note that there is a trade-off when using Condor: your job may be assigned to a machine that is slower than Hal or Kite, but with Condor you get a machine to yourself and don't have to worry about Hal or Kite getting busy. Also, if you submit your job to Condor it won't make Hal or Kite busy for others.

For more information about Condor, see An Introduction to Condor. You can also submit Stata jobs to Condor from the web.

Beowulf

A Beowulf cluster is a group of Linux servers linked together so that they can all work on the same job in parallel. Programs for the Beowulf cluster must be written in Fortran or C/C++ and use either MPI (Message Passing Interface ) or PVM (Parallel Virtual Machine). The SSCC has two Beowulf clusters. PALAH is used for learning Beowulf and testing code, while FLASH is the production cluster.

For more information about Beowulf, see Using the SSCC's Beowulf Clusters.

Which Server Should I Use?

Obviously a server is of no use to you if it doesn't have the software you need. So first check the Software page to see which servers have the software you want to use. Then if, like most SSCC members, you're comfortable with Windows but haven't used Linux, the best place to start is probably Winstat or the Lab PCs (assuming your program is available). Winstat is easy to get to and you already know how to use it.

The following are some situations you might run into which suggest another server might work better for your job than Winstat. If so, the chief obstacle may be learning Linux. Linux's command-line interface can be intimidating because it just waits for you to type something without giving you any icons or menus to suggest what you can do. But it won't take long for you to learn enough to run your jobs. Our publication Running Linux Programs Using Windows (Mostly) also has some handy shortcuts.

Here are the situations (though this is by no means a complete list):

I can't get enough memory!

If you're running a job on the Lab PCs or Winstat and can't get enough memory, you may need to switch to Kite or Hal.

I still can't get enough memory!

If Hal or Kite can't provide enough memory, your job may be a candidate for Falcon.

My job is taking forever to run!

If you're running a job on Winstat or the Lab PCs and it's taking a long time, running it on Kite or Hal will be much quicker. Condor will also be faster than Winstat or a Lab PC, and you won't have to worry about Kite or Hal getting busy.

No, it really will take forever to run. Weeks. Months. Years.

Jobs that take weeks to complete are not unusual at the SSCC, though it does require students with deadlines to plan well ahead (assume your program will go wrong at least once). On the other hand, jobs can be simply too big to be feasible.

If you're using software the SSCC supports, the first thing you should do is visit with the consultant and make sure your program is as efficient as possible. Then submit it to Condor.

But if it still takes too long, your job may be a candidate for Beowulf. You'll need to write the program in C/C++ or FORTRAN, and figure out how to use MPI or PVM to split the job into small parts that the many Beowulf processors can all work on it at the same time. But that may be the only way to run your job.

I need to run multiple jobs at the same time!

Condor is absolutely ideal for running multiple jobs. Feel free to submit up to six jobs at a time to Condor. You are also welcome to run one job each on Kite and Hal, or to reserve some of the Lab PCs in 3218. However, please do not submit more than one job per server--see the SSCC CPU Usage Policy.