Because NeSI's resources are limited, we manage access to our resources through allocations. Typically, an allocation is a grant of a certain amount of a resource, or of a rate at which a resource can be consumed, during a defined period of time. Different types of resource have different allocation criteria.
An allocation will come from one of our allocation classes. We will decide what class of allocation is most suitable for you and your research programme, however you're welcome to review our article on allocation classes to find out what class you're likely eligible for.
HPC Compute & Analytics allocations
The form of NeSI allocation you may be most familiar with is an allocation of computing power. We currently offer three sorts of HPC Compute allocations, of which your project needs at least two (online storage plus one kind of compute allocation) in order to be valid and active.
HPC Compute allocations are expressed in terms of a number of units, to be consumed or reserved between a set start date and time and a set end date and time. For allocations of computing power, we use Fair Share to balance work between different projects. NeSI allocations and the relative "prices" of resources used by those allocations should not be taken as any indicator of the real NZD costs of purchasing or running the associated infrastructure and services.
Allocations on Mahuika are measured in Mahuika compute units. A job uses one Mahuika compute unit if it runs for one hour on one physical Mahuika CPU core (two logical CPUs), using 3 GB of RAM and no GPU devices. This means a single Mahuika compute unit is equivalent to what we previously called a "Fair Share adjusted core-hour" on Mahuika's standard compute nodes.
The price of hardware in terms of compute units is shown in the following table.
|Hardware type||Fair Share Price|
|CPU||0.35 compute units per logical-CPU-hour|
|Memory (RAM)||0.10 compute units per GB-hour|
|P100 GPU device||7.0 compute units per device-hour|
|A100 GPU device||18.0 compute units per device-hour|
|A100-1g.5gb GPU device||3.0 compute units per device-hour|
The total compute unit cost of a job is the sum of these three components. Once the job has finished running, this composite price is what affects your project's Fair Share score. However, whether your institution will be charged based on the composite price or based on your job's CPU core hour consumption alone, or on some other basis, will depend on your contractual arrangements with the NeSI host.
Note that the minimum number of logical cores a job can take on Mahuika is two (see Hyperthreading for details). Therefore:
- the lowest possible price for a CPU-only job is 0.70 compute units per hour, plus memory (RAM).
- the lowest possible price for a CPU + P100 GPU job is 7.70 compute units per hour, plus memory (RAM).
- the lowest possible price for a CPU + A100 GPU job is 18.70 compute units per hour, plus memory (RAM).
In reality, every job must request at least some RAM.
The compute capacity of the Māui supercomputer is allocated by node-hours. Though some Māui nodes have more RAM than others, we do not currently distinguish between low-memory and high-memory nodes for allocation, billing or Fair Share purposes.
Each allocation on Māui includes an entitlement to use the Māui ancillary nodes equally with other NeSI projects having Māui allocations at that time.
One Māui node hour is roughly equivalent to 40 Mahuika compute units.
Online storage allocations
An online storage allocation, unlike compute allocations, is more like a lease than a rate of consumption. It is an amount of disk space and, concurrently, a number of inodes (directory entries, i.e. files etc.) that have been made available for your project team to use on our online high-performance file system. An online storage allocation is typically granted to your persistent project directory.
We do not yet have a ratio of online storage disk space or inodes to Mahuika compute units.
Data Storage & Transfer allocations
Nearline storage allocations
A nearline storage allocation, like online storage allocations but unlike compute allocations, is more like a lease than a rate of consumption. It is an amount of space and, concurrently, a number of inodes (directory entries, i.e. files etc.) that have been made available for your project team to use on our nearline apparatus.
We do not yet have a ratio of nearline storage tape space or inodes to Mahuika compute units.
Consultancy & Training allocations
A consultancy allocation is for a number of scientific programmer hours between two dates, or is sometimes expressed as a fraction of an FTE between the same two dates. This reflects the commitment of NeSI scientific programming expertise to your project.
We do not yet have a ratio of consultancy hours to Mahuika compute units.
NeSI training courses are not subject to a systematic allocation mechanism. To enquire about the possibility of NeSI running a course at your institution please contact our training team.