This is an old revision of the document!
Maris is a small computational cluster at the Lorentz Institute, financed by external research grants. Access is primarily for those research groups who have purchased the machines, but there may well be computing time available for others. If you would like to use the Maris cluster, please get in touch with the local contact persons at the Lorentz Institute to see what resources can be made available for your needs. You can then request access by sending an email to Carlo Beenakker.
The cluster is optimised for multithreading applications and embarrassingly parallel problems.
Once you have been granted access to the cluster, you must login on to its head node to access it. Whitin the IL network, Maris' head node is reachable at novamaris.lorentz.leidenuniv.nl
or trough its aliases maris
and mariscluster
. For connections from outside the IL network, an ssh tunnel into the IL ssh server (styx.lorentz.leidenuniv.nl
) is needed.
Note that ssh.lorentz.leidenuniv.nl is an alias of styx.
Direct ssh access to any computation node is disabled. To open an interactive session to one of the nodes use for example
srun <whatever srun options you'd like> --pty bash
Any calculation must be submitted to slurm for execution. If not, the associated process will be terminated without any notice.
A summary of the current cluster load/usage is available on here (only within the IL network).
Maris' head node and nodes run the GNU/Linux OS Fedora 29. The number of nodes available for calculations is subject to change depending on maintenance works that need to be carried out on a regular basis. You are advised to run sinfo
prior to any job submission. The nodes are organized in groups according to their specs and configuration
CPU node(s) | Server Type | CPU(s) | Clock Speed | Cores | Threads | RAM | CPU Family |
---|---|---|---|---|---|---|---|
maris0[04 - 22] | ASUS RS161-E5/PA2 | 2 x AMD 2350 | 2GHz | 8 | 16 GB | AMD | |
maris0[23 - 46] | Supermicro H8DMT | 2 x AMD 2350 | 2GHz | 8 | 32 GB | ||
maris0[47 - 59] | Dell PowerEdge R815 | 4 x AMD 6174 | 2.2GHz | 48 | 48 | 128 GB | |
maris060 | Dell PowerEdge R815 | 4 x AMD 6276 | 3.2GHz | 64 | 64 | 512 GB | |
maris061 | Dell PowerEdge R815 | 4 x AMD 6174 | 3.2GHz | 48 | 48 | 128 GB | |
maris062 | Dell PowerEdge R815 | 4 x AMD 6174 | 3.2GHz | 48 | 48 | 256 GB | |
maris0[63,64] | Dell PowerEdge R815 | 4 x AMD 6276 | 3.2GHz | 64 | 64 | 512 GB | |
maris0[65,67,68] | Dell PowerEdge R815 | 4 x AMD 6376 | 3.2GHz | 64 | 64 | 256 GB | |
maris0[69-73] | Dell PowerEdge R815 | 4 x AMD 6376 | 3.2GHz | 64 | 64 | 512 GB | |
maris066 | Dell PowerEdge R815 | 4 x AMD 6376 | 3.2GHz | 64 | 64 | 192 GB | |
maris074 | Dell PowerEdge R815 | 4 x AMD 6276 | 3.2GHz | 64 | 64 | 192 GB | |
maris076 | Dell PowerEdge R830 | 4 x Intel Xeon E5-4640 v4 | 2.1GHz | 96 | 96 | 512 GB | INTEL |
maris077 | Dell PowerEdge R830 | 4 x Intel Xeon E5-4640 v4 | 2.1GHz | 96 | 96 | 512 GB | |
InfiniBand Subcluster | |||||||
CPU node(s) | Server Type | CPU(s), IB(s) | Clock Speed | Cores | Threads | RAM | |
maris078 | Dell PowerEdge R840 | 4 x Intel Xeon Gold 6126 | 2.6 GHz | 96 | 96 | 512 GB | |
1 x Mellanox EDR | |||||||
GPU Subcluster | |||||||
GPU node(s) | Server Type | CPU(s), GPU(s) | Clock Speed | Cores | Threads | RAM | |
maris075 | Dell PowerEdge R730 | 2 x Intel(R) Xeon(R) CPU E5-2680 v4 | 2.4GHz | 56 | 56 | 256 GB | |
2 x Tesla P100 16GB |
The resources allocatable on each node might differ from those in the table above because of the resources necessary to run a node's OS.
The nodes have been configured in a similar way to your workstation to maximize productivity although there are substantial differences worth noting. One of these is the home directory. In order to keep the network traffic at a low rate, the /home directory on the Maris cluster is different from the one of your Lorentz GNU/Linux Workstation. In order to access the latter from the head node you can use
cd /lorentz/your_username
In a similar fashion, you can access all data[1,..,n] on each Lorentz Institute workstation using for instance
/net/workstation_name/data1
Also novamaris
and all maris nodes mount at boot two storage devices one available under the /clusterdata
mount point and the other under /marisdata
. The former is an old storage device which will be retired upon failure, the latter is a newly installed (as of 17/05/2016) storage device which you are encouraged to use.
NOTE: Some devices get mounted upon access by autofs. Do not get surprised if the command df
does not show the desired output.
Maris home directories are mounted under /home
. This is a dedicated 5.5TB storage system which keeps hourly, daily and weekly snapshots of the home disk. Should you need to access one of the snapshots in the event of a lost file, then access the directory .snapshot
and recover the file you need, for instance
$ ls /home/.snapshot daily.2016-05-16_0010 hourly.2016-05-17_0405 hourly.2016-05-17_0605 hourly.2016-05-17_0805 weekly.2016-05-08_0015 daily.2016-05-17_0010 hourly.2016-05-17_0505 hourly.2016-05-17_0705 hourly.2016-05-17_0905 weekly.2016-05-15_0015 $ cd /home/.snapshot/hourly.2016-05-17_0805/<username>
10 GB quotas are enforced on the home disk. Please use /marisdata to temporarily store large datafiles.
Maris has currently two storage disks mounted respectively under /clusterdata and /marisdata.
The plan is to gradually replace /clusterdata with its newer counterpart /marisdata. As of 17/05/2016 you are encouraged to use ONLY /marisdata to store your data because /clusterdata is no longer maintained and all data on it will get permanently lost in case of hardware failure. In fact, you are encouraged to move all data you deem important from /clusterdata to /marisdata as soon as possible. /marisdata
has 37TB available.
Please remember that /marisdata is supposed to act as a temporary storage device during your calculations. You are encouraged not to use it as an archive disk.
maris078 has an EDR InfiniBand (IB) connector to and SSD storage server to achieve high I/O rates during calculations. The SSD storage server disks are mounted under /IBSSD
and configured as a RAID0 (no redundancy for data!). The protocol used is iSER.
df -h /IBSSD/ Filesystem Size Used Avail Use% Mounted on /dev/sdc 5.1T 89M 4.8T 1% /IBSSD
We try to minimize the differences between your Lorentz workstation and Maris so you can be as productive as possible. A variety of compilers and libraries are available for you to use in a similar fashion to your workstation. Explore which modules are available by typing module avail
. For extra information take also a look at this page and read the manual pages: man module
.
All calculations must be submitted and executed through slurm
. A typical pseudo-session would look like
If this is the first time you use slurm, please have a look at this overview of maris' slurm configuration and this short guide to slurm before submitting your calculations for execution.
NOTE that calculations executed outside slurm's control will be terminated without notice.
Maris has a large list of pre-installed python modules available for you to use. Nonetheless, it is possible to install new modules or hack existing ones as long as they are (re)installed in a location for which you have writing credentials. You can find an extensive guide with examples here.
Further readings: virtualenv and conda.
Help with cluster issues can be requested through the helpdesk. It is strongly advised you discuss your problem also with other Maris users because they might have helpful tips. Here follows a list of Lorentz Institute members that could be consulted for tips
* Thomas O'Brien (Office 259, Telephone: 5534) * IT Support (HL 40[7-9], Telephone:8484)