LOTUS overview
This article gives an overview of the LOTUS batch processing cluster which is part of JASMIN. It covers:
- What LOTUS is and what it can be used for
- Where LOTUS can be accessed from
What is LOTUS?
LOTUS is not, in itself, a High Performance Computing (HPC) facility, but provides the batch and parallel processing component of the JASMIN data-intensive scientific analysis environment. It is a group of physical machines, running the Platform LSF workload manager, enabling efficient scheduling of larger data analysis tasks across nodes in the cluster as a single unit. Each node in the cluster is connected by 10Gbit/s Ethernet to JASMIN's high-performance 40Gbit/s core network. Although not its primary function, LOTUS also facilitates MPI-based parallel processing.
JASMIN provides both interactive and batch computing environments, recognising that scientists often need to develop and test workflows interactively before running those workflows efficiently at scale. Nodes within LOTUS run the same stack of software and can access the same high-performance storage as the JASMIN Scientific Analysis servers, ensuring a consistent working environment for all phases of users' workflows.
LOTUS is ideally suited to workflows which need to process or compare entire datasets, stored either in Group Workspaces or in the CEDA archives. The latter are directly accessible read-only so can be processed in-place without the need to copy files. Intermediate working files (within batch jobs) such be stored temporarily in /work/scratch
which is shared across the cluster, while persistent outputs can be written efficiently to Group Workspaces and shared with collaborators for the duration of a project.
See Access to Storage for details about which file systems are appropriate to use and how to access them.
LOTUS currently has around 4000 cores, but is heavily used and implements a fair-share scheduling system between users. It is not intended as a substitute for dedicated HPC facilities, rather as a complementary environment in which model outputs can be analyzed and compared with observational data. Users with large-scale compute-heavy requirements (in particular those requiring large-scale parallel processing) should look to access other parts of the national HPC infrastructure such as ARCHER or MONSooN.
See LOTUS Hardware for details of the current LOTUS environment.
In order to maintain a safe and reliable working environment for all within LOTUS and more widely within JASMIN, users are expected to follow best practice outlined in this documentation.
Access to LOTUS
LOTUS is accessed from the JASMIN scientific analysis servers:
jasmin-sci[1-5].ceda.ac.uk
cems-sci[1-2].cems.rl.ac.uk
From these servers, it is possible to estimate and allocate resources for jobs, to submit, monitor and control jobs, via commands which interact with the workload manager, Platform LSF.
Please note that if you have only recently created your login account and had this approved, there can sometimes be a delay (typically up to a day, but in rare cases can be longer) before the necessary configuration is created for you on LOTUS. You will not be able to submit jobs to LOTUS queues until this has been completed. Typically you would see an error message such as this, in this case after an unsuccessful attempt to submit to the short-serial
queue:
short-serial: User cannot use the queue. Job not submitted.
If this occurs, please try again in 24 hours before contacting the help desk.