Compute Servers

Applications such as Mathematica and cplex require lots of memory and cpu cycles so that they often impede other, usually interactive, activities on your desktop or even cannot run. And the weekly reboot/upgrade procedures on desktop systems may terminate your compute job prematurely.

In those cases, revert to one of the remote compute systems below.

Computing facilities at SURFsara

Procedures to provide access to computing facilities at SURFsara have started as a pilot in September 2016. To participate in the pilot or follow the procedure that may already have succeeded it, try the following links:

Computing facilities at the department

They can be accessed with your Linux account unless indicated otherwise. They do not take part in the weekly reboot/upgrade procedures, and a notice will be sent to in case of planned maintenance.

Status

These machines were purchased with money from projects that finished years ago. New budget for continuation or replacement still needs to be negotiated. We keep them going on a best-effort basis.

Hardware maintenance contracts for these systems expired in 2015 or earlier, for all hardware except the ngrid chassis, which allows the ngrid machines to operate.

This means we won't be replacing any other hardware: if any of these machines goes down due to hardware failure, it may never be back! Several ngrid machines have already disappeared due to hard disk failures, and more will follow.

The operating systems on mammoth and elephant are severely outdated. Due to their unusual architecture and their end-of-life status, we can't afford to upgrade them. As a consequence, we can't install or upgrade any software on them, either.

The operating systems on the ngrid01-ngrid31 systems are out of date as well. We intend to replace them in the coming months (Q2 of 2016).

Disk space

Disk space is available on /home and /scratch as usual on linux systems. We may impose limits also on /scratch if we get too many disk full problems because on filesystems with hundreds of GB, it may take hours only to find out who is using disk space. You are welcome to use the space and speed that /scratch provides but in case you plan to use a considerable amount of that space, please also plan a date to clear disk space. If you need to store compute results or other precious data that are too big for /home, consider a secondary copy on the storage server .

We impose disk quota.

Running GUI programs

In order to run GUI programs on these servers, you need to login with ssh -Y, else they will not have sufficient access to your display facilities and will appear with a grey window, without warning.

To run GUI programs from Windows systems, you need to install and setup an X Window server, such as Exceed.

Smaller resource requirements

If you want to work remotely but you need to perform simple tasks only, such as reading your mail, please use our remote Linux desktops.

Bigger resource requirements

In case you need more resources than your desktop provides, or just want to keep your desktop free for other activities, consider the systems below.

These systems can only be accessed from tue.nl systems, that is, computers on campus or connected using TU/e's VPN service.

ngrid01.win.tue.nl .. ngrid32.win.tue.nl
32 servers with each: 8 processor cores and 16 threads, 2 Ghz, 12 GB memory, 194 GB local disk, Fedora 14 / 64-bit
Provided by LaQuSo, managed by BCF. The 32 servers that can be used as separate systems or as a cluster. It is a cluster only in the sense that the subsystems have relative fast network connections. There is currently no special cluster software installed but you can ask for such software if needed. The live statistics page can help to find resources for your job.
mammoth.win.tue.nl
56 x 2 Ghz, 935 GB memory (aggregated), 2.8 TB local disk, Fedora 12 / 64-bit
Provided by LaQuSo, managed by BCF. The system is composed of 7 servers with each 144 GB that are aggregated by the vSMP software to appear as one system . All of the memory appears as a contiguous block of memory to the OS and to programs. The performance depends on your memory access profile, which is a complex matter. In general, vSMP gets the most out of the hardware. A few typical cases follow below:
  • Accessing all of the memory, fully randomly: latency of infiniband
  • Accessing all of the memory, sequentially: bandwith of inifinband
  • Clustered access, within a server's memory size: normal server performance
You can use numactl or taskset to have your process run on specific cores. The available nodes (physical cpus) and their free memory are reported by numactl -H.
elephant.win.tue.nl

4 x 2.5 Ghz, 64 GB memory, 2.7 TB local disk, Fedora 12 / 64-bit Managed by BCF. There is no more hardware service any more for this system so we may need to close it at any moment. The CPU board is broken which is why only 4 of 8 cores and 64 of 128 GB of memory are available.

Even more resource requirements

If the previous systems are too busy, or do not fulfill your needs adequately, you can ask us to set up a new (desktop) PC for private usage. Such requests will be granted without formalities but only for a limited period. PCs of this nature are usually set up in a day or two.

If the previous proposals do not work for you, request another system through your IT Committee representative.


Contact us | Webmaster

Questions? Mail to helpdesk.win@tue.nl