Allocations and compute scheduling: Difference between revisions

With the 2025 [[infrastructure renewal]] it will become possible to schedule a fraction of a GPU using [[multi-instance GPU]] technology.  Different jobs, potentially belonging to different users, can run on the same GPU at the same time.  Following [https://docs.nvidia.com/datacenter/tesla/mig-user-guide/#terminology NVidia's terminology], a fraction of a GPU allocated to a single job is called a "GPU instance", also sometimes called a "MIG instance".   

Note: a GPU instance of profile <b>1g</b> is worth 1/7 of a A100 or H100 GPU. The case of <b>3g</b> takes into consideration the extra amount of memory per <b>g</b>.

* [[File:Two_core_equivalents.png|frame|Figure 2 - Two core equivalents.]] Research groups using more cores than memory (above the 1 core/4GB memory ratio), will be charged by cores.  For example, a research group requesting two cores and 2GB per core for a total of 4 GB of memory. The request requires 2 core equivalents worth of cores but only one bundle for memory.  This job request will be counted as 2 core equivalents when priority is calculated. See Figure 2. <br clear=all>

=Viewing usage of compute resources time series at portal.alliancecan.ca/slurm=

[[File:Slurm portal land edit.png|thumb|alt=usage portal landing view|usage portal landing view]]

The portal.alliancecan.ca/slurm site provides an intuitive interface for exploring job usage time series. The page contains a time series histogram figure window that displays usage metrics derived from job records on the Alliance national clusters. The initial metric displayed in the histogram is CPU-equivalent days for the logged-in user on the Cedar cluster. If the logged-in user does not have usage on the Cedar cluster the figure window will not have a histogram but rather contain the text “No Data or usage too small to have a meaningful plot”.The data summaries presented in the figure window can be modified by the edit and selection panels along the left margin of the page. The editing panels are "Select system and dates", "Parameters" and "SLURM account".

==Displaying a specified account==

The "Select user’s account" pull down menu of the "SLURM account" panel allows the user to select the specific project account to be displayed in the figure window. If the “Select user’s account” is left empty the histogram figure window will display all of the usage for the logged in user across on the specified cluster over the selected time period. The "Select user’s account" pull down menu is populated by a list of all the accounts that have job records by the logged-in user on the selected cluster over the selected time interval. Once that a single project account is selected the usage histogram figure is updated and the summary panel titled "Allocation Information" is populated with relevant details of the project account. The height of each bar in the histogram figure corresponds to the metric for that day (e.g. CPU-equivalend days) across all users in the account on the system, where the top 8 users are displayed in a unique colour stacked on top of the summed metric for all other users in gray. The figure can be navigated using native Plotly tools (e.g. zoom, pan, etc) whose icons appear at the top-right when the mouse hovers over the figure window. The figure's right side legend can also be used to manipulate the figure display such that single clicking an item will toggle the item's presence in the figure, and double clicking the item will toggle isolating the figure to the single item.  

==Displaying the allocation target and queued resources

[[File:Slurm portal account usage queued.png|thumb|alt=Allocation target and queued resources displayed on the usage figure|Allocation target and queued resources displayed on the usage figure]]

Beyond the usage values displayed in the histogram two additional values can be displayed for "Allocation target" and "Queued jobs". Once that a single account has been selected for display the "Allocation target" is enabled by default but this can be changed by setting the “Display allocation target by default” to “No” in the “Parameters” panel. The "Queued jobs" metric, which presents a sum of all resources in pending jobs at each time point, can be enabled at any time by single clicking on the item in the figure legend to the right of the usage histogram axis.  

=Viewing group usage summaries of compute resources at ccdb.ca= <!--T:35-->

For resource allocation projects that have GPU usage the table is broken down into usage on various GPU models and measured in RGUs.