Using Paid-For Resources§
Paid resources may be in the form of priority access (Gold), dedicated nodes or both.
Users with access to Gold have tools for monitoring budgets and usage.
Users with access to dedicated nodes have a tool for monitoring current use of their nodes.
Why we recommend priority access§
We recommend priority access via Gold rather than dedicated nodes in most circumstances because:
- The amount of Gold you get is equivalent to the core hours if you ran your node at 100% utilisation throughout its 3-year lifespan.
- If you have a dedicated node and aren't currently running anything on it, it is idle and you cannot recover that time (and it still uses power and incurs hosting costs).
- If you aren't currently using your Gold, you have 3 months (or your chosen allocation period length) to recover that time and use it.
Priority access via Gold§
If you have priority access, this is managed by a resource called Gold. Gold is divided into allocations, which have start and end dates and a given amount of Gold associated with them. Any unused Gold expires after the allocation period ends.
On Myriad, one Gold = one core hour. If you run a job that asks for a wallclock time of 3hrs and 10 cores, then the job costs 30 Gold.
Gold is reserved when you submit a job, so the entire 30 will be reserved
and taken away from the available balance when you run
qsub. After the
job ends, how long it ran for is checked, and any unused Gold is put back
for anyone to use.
For example, you asked for 3 hours and 10 cores, but your job finished in 2 hours. When you submit the job, it will reserve 30 Gold and your budget will go down by 30. When it ends, the final charge is only 20 Gold, so 10 Gold gets put back in your available budget.
How it works§
On Myriad, standard jobs that are submitted all start out with a priority of 2.xxxxx once the scheduler has seen them. (Priority 0.00000 means the scheduler hasn't cycled round to looking at this job and assigning it a priority yet). Gold jobs begin with priority 3.xxxxx and so will be higher in the queue than any non-Gold jobs.
This does not mean they will be scheduled instantly - the resources they are asking for still need to become free, and there could be other Gold jobs also in the queue. But it greatly increases the likelihood that this will be the next job to run.
It is possible in rare circumstances on Myriad for a non-Gold job to reach priority 3. It is likely that Gold jobs will still have higher priority.
View your Gold budgets§
To see the Gold budgets available to you, run:
You will see something like this:
Project Machines Balance --------- -------- ----------- hpc.999 ANY 124560.00 hpc.998 ANY 0.00
The project column shows which budgets you have access to and the balance shows how much is left unused in the current allocation.
Jobscript additions for Gold jobs§
You choose whether you want a specific job to be a Gold job or a normal priority job. For a Gold job, add these to your jobscript:
#$ -P Gold #$ -A hpc.xx
You can also pass these in on the command line to the
qsub -P Gold -A hpc.xx myscript.sh
Viewing allocation dates§
You can look at all the start and end dates for your allocations:
glsalloc -p hpc.xx
Output will look like this:
Id Account Projects StartTime EndTime Amount Deposited Description --- ------- -------- ---------- ---------- --------- --------- -------------- 001 01 hpc.999 -infinity infinity 0.00 0.00 Auto-Generated 002 01 hpc.999 2021-12-01 2022-03-01 105124.00 205124.00 003 01 hpc.999 2022-03-01 2022-06-01 205124.00 205124.00 004 01 hpc.999 2022-06-01 2022-09-01 205124.00 205124.00
Allocations begin and end at approximately 00:05 on the date mentioned.
- 'Deposited' is the total amount this allocation had to begin with.
- 'Amount' is the amount it has left just now.
Monitoring Gold usage§
You can view some information about when your Gold was used, in which jobs, and by whom.
# show help gstatement --man # show statement between the given dates gstatement -p hpc.xx -s 2020-12-01 -e 2021-12-01 # give a summary between the given dates gstatement -p hpc.xx -s 2020-12-01 -e 2021-12-01 --summarize
For dedicated nodes, only members of your project are allowed to run jobs
on your node. Your project is usually set by default so you do not
need to specify it in your jobscript. You can check this by looking at
qstat -j $JOB_ID for an existing job ID, and looking at the
line near the bottom.
Jobscript additions for dedicated nodes§
If the project is not being set by default, for a job to be eligible to run on your nodes you will need to specify your project in your jobscript:
# Specify project #$ -P <project>
This will allow a job to run on your nodes, but it can also be scheduled on general-use nodes if some are available first. This should be the main way you run jobs.
If you need to, you can force jobs to run on your nodes only. This is suitable when you have arranged policies on your nodes that are different from the normal policies (eg. a longer maximum wallclock time), as it means your policies will be in effect instead of the general policies.
# Specify paid flag to force running on paid nodes only, with your policies #$ -l paid=1
Check what is running on your nodes§
We have a script named
whatsonmynode, that runs
qhost -j for all the
nodes belonging to your project, so you can see which nodes you have,
what is running on them and from which user.
module load userscripts whatsonmynode <project>