index.rst 2.07 KB
Newer Older
Michael Krause's avatar
Michael Krause committed
1
2
3
4
5
6
7
8
9
10
11
Introduction
============

This manual describes what the TARDIS is and provides some basic examples and
useful code snippets for various software packages that can be used on LIP's
cluster environment. We are trying to update the documentation every time we
successfully test a new use case. It is meant to be a guideline and if you need
help at any point we will be glad to assist you. We appreciate all of your
feedback so please contact us: `grid-admin@mpib-berlin.mpg.de
<grid-admin@mpib-berlin.mpg.de>`_

Michael Krause's avatar
next    
Michael Krause committed
12
13
Specs
-----
Michael Krause's avatar
Michael Krause committed
14

Michael Krause's avatar
next    
Michael Krause committed
15
16
Tardis
   **T**\ardis, **A** **R**\apid **D**\istributed **I**\nformation **S**\ystem
Michael Krause's avatar
Michael Krause committed
17

Michael Krause's avatar
next    
Michael Krause committed
18
19
20
21
.. image:: ../img/tardis_real.png
   :width: 60%

Some technical facts:
22
    + **832** Intel® Xeon® CPU E5-2670 CPU cores(no HT) inside 48 Dell m6x0 blade servers
Michael Krause's avatar
Michael Krause committed
23
    + **R**:sub:`max` = 9.9 TFlops, **R**:sub:`peak` = 14.8 TFlops
Michael Krause's avatar
Michael Krause committed
24
25
26
    + **10.6TB** total amount of memory
    + **747 TB** of attached BeeGFS storage
    + **10GB/s** fully-connected Ethernet
Michael Krause's avatar
next    
Michael Krause committed
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53


Workflows
---------

**Sequential**

The simplest and very typical processing workflow consist of 3 steps.

+ data download from file servers
+ sequential data processing
+ result upload to file servers

.. image:: ../img/WF-seq.svg
   :width: 80%

Obviously this is not very efficient and with data processing on the Tardis we
would want to do something more efficient. Also this approach requires desktop
machines in a shared working environment to be always on. This procudes
excessive noise and heat. It is also not very nice considering other colleagues
might want to use your machine as well. But most importantly it is **slow**.

**Parallel**

With the Tardis you can login from your laptop or workstation with SSH (see:
:doc:`login`) to a single head node called ``tardis``. On that node users can
prepare and test their code and analysis and then submit it to a queue (see:
54
:doc:`../pbs/torque`). Jobs will then **eventually** be submitted to one of the
Michael Krause's avatar
next    
Michael Krause committed
55
56
57
58
59
computing nodes to get a guaranteed set of processing resources. Afterwards
users can collect the results and copy them back to the file servers.

.. image:: ../img/WF-par.svg
   :width: 80%