Wiki DIFA

Questa pagina è in sola lettura. Puoi visualizzare il sorgente, ma non puoi modificarlo. Contatta l'amministratore se pensi che ci sia un errore.
====== General informations ======

The hardware structure of the DIFA-OPH computing cluster is summarised in the table below, listing all the compute nodes currently available with their individual hostnames, their specific resources (number of cores and available RAM), and their access policies.

In particular, bldNN nodes are part of the **BladeRunner island**, mtxNN nodes are part of the **Matrix island**, and gpuNN are part of the **GPU island**. 

**The nodes associated with the OPH project are open to all users** while nodes associated with other individual projects may be subject to **access restrictions**. Such restrictions are indicated in the last column of the table.

  * **Shared** nodes, the access is available for all users
  * **Teaching** nodes, reserved for teaching purposes: can be accessed only by students of specific DIFA courses for their laboratory activities
  *** Reserved nodes**, the access is restricted to users explicitly authorised by the corresponding project PI till the given **exp**iration date.
<WRAP width=100%>
^ Nodes      ^ vCPUs/RAM   ^ GPUs   ^ Project (PI)          ^ Access type (exp)  ^
| bld[01-02] | 24 / 64G    | -      | OPH                   | Shared             |
| bld[03-04] | 32 / 64G    | -      | OPH                   | Shared             |
| bld05      | 32 / 128G   | -      | OPH                   | Shared             |
| bld[15-16] | 16 / 24G    | -      | OPH                   | Teaching           |
| bld[17-18] | 32 / 64G    | -      | OPH                   | Shared             |
| mtx[00-15] | 56 / 256G   | -      | OPH                   | Shared             |
| mtx[16-19] | 112 / 512G  | -      | ERC-Astero (Miglio)   | <del>Reserved (2024-09)</del> Shared |
| mtx20      | 112 / 1T    | -      | OPH (Di Sabatino)     | Shared             |
| mtx[21-22] | 192 / 1T    | -      | SLIDE (Righi)         | Reserved (2025-10) |
| mtx[23-25] | 112 / 512G  | -      | OPH (Marinacci)       | Shared             |
| mtx26      | 112 / 512G  | -      | CAN (Bellini)         | Reserved (2026-04) |
| mtx27      | 112 / 512G  | -      | FFHiggsTop (Peraro)   | Reserved (2026-04) |
| mtx[28-29] | 112 / 1.5T  | -      | :::                   | :::                | 
| mtx30      | 64 / 1T     | -      | Trigger (Di Sabatino) | Reserved (2027-02) |  
| mtx[31-32] | 64 / 1T     | -      | EcoGal (Testi)        | Reserved (2027-08) |  
| mtx[33-34] | 192 / 1.5T  | -      | :::                   | :::                |
| mtx[35-36] | 192 / 512G  | -      | RED-CARDINAL (Belli)  | Reserved (2028-04) |  
| mtx[37-40] | 192 / 512G  | -      | ELSA (Talia)          | Reserved (2026-12) |
| gpu00      | 64 / 1T     | 2xA100 | VEO (Remondini)       | Reserved (2026-04) |
| gpu[01-02] | 112 / 1T    | 4xH100 | EcoGal (Testi)        | Reserved (2027-08) |  
| gpu03      | 112 / 1T    | 4xH100 | ELSA (Talia)          | Reserved (2026-12) |
</WRAP>
====== Computing Resources ======

Resources are nodes, CPUs, GPUs((Only on GPU nodes)), RAM and time. You'll have to select the resources you need for the job. Do not overstimate too much or you'll be "billed" too much. But don't understimate or your job won't be able to complete. **When a job completes, you receive a mail with seff output: this should help a lot to optimize future requests.**

Nodes are grouped by partitions. **DO NOT** specify neither nodenames nor partitions unless directed to do so by tech staff.

===== Selecting nodes =====

To select a (set of) node(s) suitable for your job, **use constraints**. These include:
  * **blade**: older nodes, usually for smaller/sequential jobs, quite heterogeneus
  * **matrix**: newer nodes, for bigger parallel jobs; **allocated by "half node" units!**
  * **ib**: require IB-equipped nodes (**all** nodes in matrix are IB-equipped => no need to specify)
  * **filetransfer**: ask for a node with fast access to outside network to quickly tranfer big files
  * **intel**: require an Intel CPU
  * **amd**: require an AMD CPU
  * **avx**: require that the CPU supports AVX instructions
  * **dev**: require that the node can be used to compile (deprecated: all nodes host build tools)
  * **dida**: require nodes used for lessons (obsolete)
  * **gpu**: require a GPU-equipped node

===== Reserved nodes =====

Some nodes are reserved for specific projects (see table above). To be able to use 'em you have to be explicitly allowed by project manager (= added to the project group via DSA interface). Once you're in the allowed group (check with ''id'') you can submit jobs specifying ''--reservation=prj-...''.

^ Project     ^ Manager    ^ AD group (DSA)                 ^ OPH group (''id'')  ^ Reservation to use ^
| CAN         | Bellini    | Str04109.13664-OPH-CAN         | OPH-res-CAN         | prj-can            |
| ECOGAL      | Testi      | Str04109.13664-OPH-ECOGAL      | OPH-res-ECOGAL      | prj-ecogal         |
| ELSA        | Talia      | Str04109.13664-OPH-ELSA        | OPH-res-ELSA        | prj-elsa           |
| FFHiggsTop  | Peraro     | Str04109.13664-OPH-FFHiggsTop  | OPH-res-FFHiggsTop  | prj-ffhiggstop     |
| RedCardinal | Belli      | Str04109.13664-OPH-RedCardinal | OPH-res-RedCardinal | prj-redcardinal    |
| SLIDE       | Righi      | Str04109.13664-OPH-SLIDE       | OPH-res-SLIDE       | prj-slide          |
| Trigger     | DiSabatino | Str04109.13664-OPH-Trigger     | OPH-res-Trigger     | prj-trigger        |
| VEO         | Remondini  | Str04109.13664-OPH-VEO         | OPH-res-VEO         | prj-veo            |

===== QualityOfService =====

What other clusters call "queue", Slurm calls QoS.

By default all jobs are queued as "--qos=normal".

Each QoS offers different features:

^ QoS    ^ Max runtime ^ Priority ^ Note ^
| normal |  24h        | standard | Default |
| debug  |  15'        | high     | Max 2 nodes, 1 job per user, **not billed** |
| long   |  72h        | low      | |

====== Storage Resources ======

Storage is detailed [[oph:cluster:storage|on its own page]].

It's important to select the correct storage for the use you're going to do.