Summary¶

The Metrix portal is a website for Alliance users. It collects information on compute nodes and management servers to interactively generate data, allowing you to track your resource usage (CPUs, GPUs, memory, filesystems) in real time.

Cluster	URL
Rorqual	https://metrix.rorqual.alliancecan.ca
Narval	https://metrix.narval.alliancecan.ca
Nibi	https://portal.nibi.sharcnet.ca
tamIA	https://portail.tamia.ecpia.ca
Vulcan	https://metrix.vulcan.alliancecan.ca

Filesystem performance This section provides graphs for bandwidths and metadata operations, with viewing options for the last week, last day, and last hour.
Login nodes Under this tab, usage statistics for CPUs, memory, system load, and network are presented, with viewing options for the last week, last day, and last hour.
Scheduler This tab shows statistics for the cluster's allocated cores and GPUs, with viewing options for the last week, last day, and last hour.
Scientific software These graphs display the most frequently used software, including CPU cores and GPUs.
Data transfer nodes Bandwidth statistics for data transfer nodes are shown under this tab.

User portal¶

Under this tab, you find your quotas for the filesystems, followed by your 10 last jobs. You can select a job by its number to see the details. Also, by clicking on (More details), you are redirected to the Job statistics tab, where all your jobs are listed.

Job statistics¶

The first block shows your current usage (CPU cores, memory, and GPUs). These statistics represent the average usage by all currently running jobs. You can easily compare the resources allocated to you with those you actually use.

An additional graph shows the average usage over the last few days.

The system then displays your activity on the filesystems. This includes the number of disk write commands you have performed (input/output operations per second or IOPS) and the amount of data transferred to the servers over a given period (Bandwidth).

The next section shows all the jobs you have already started, which are currently running or pending. In the top left corner, you can filter jobs by their status (OOM, completed, running, etc.). In the top right corner, you can search by job ID or by job name. Finally, in the bottom right corner, there is an option to quickly navigate between pages by performing multiple jumps.

CPU jobs¶

At the top, you see the job name, its number, your username, and the status. Details of your submission script are displayed by clicking on Show submitted job script. If the job was launched in interactive mode, the submission script will not be available.

The working directory and the submission command can be seen by clicking on Show submit command.

The next section shows information on the scheduler. To display the information on your CPU account, click on your account number.

In the Ressources section, you can see the resources used by your job by comparing columns Allocated and Used for the listed parameters.

A graph illustrates the CPU cores you have requested, over time, allowing selection of specific cores.

Short Jobs

This graph is not available for very short jobs.

A corresponding graph displays the usage of your requested memory, over time.

The Processes and threads graph shows different parameters. For a multithread job, adding parameters Running threads and Sleeping threads should not exceed twice the number of cores requested. However, having some Sleeping threads is normal for certain types of programs (Java, Matlab, commercial software, or complex programs). There is also a parameter for the program applications that have been executed over time.

Charts show filesystem usage by the current job, distinct from overall node usage. These include the number of I/O operations per second (IOPS) and the data transfer rate between the job and the filesystem, over time, to highlight periods of high or low filesystem activity.

Node-wide resource statistics may be inaccurate if the node is shared. The displayed data includes the evolution of job bandwidth over time, relating to software and licenses, and the evolution of network bandwidth used by jobs via the Infiniband network, where periods of massive data transfer (e.g., Lustre filesystem I/O, MPI communication) can be observed.

Further statistics show the evolution of local disk IOPS and local disk bandwidth (data read/written per second) over time.

Information on local disk space usage is also available.

Power consumption data is provided.

CPU jobs (job arrays)¶

The page for a CPU job in an array is the same as that for a regular CPU job, except for the Other jobs in the array section. The table lists the other job numbers that are part of the job array, along with information about their status, name, start time, and finish time.

GPU jobs¶

At the top, you see the job name, its number, your username, and the status. Details of your submission script are displayed by clicking on Show submitted job script. If the job was launched in interactive mode, the submission script will not be available.

The working directory and the submission command are shown by clicking on Show submit command.

The next section shows information on the scheduler. To display the information on your GPU account, click on your account number.

In the Ressources section, you can see the resources used by your job by comparing columns Allocated and Used for the listed parameters.

A graph illustrates the CPU cores you have requested, over time, allowing selection of specific cores.

Short Jobs

This graph is not available for very short jobs.

A corresponding graph displays the usage of your requested memory for CPUs, over time.

The Processes and threads section provides various parameters.

Charts display filesystem usage by the current job, separate from overall node usage. These include the number of I/O operations per second (IOPS) and the data transfer rate between the job and the filesystem, over time, to highlight periods of high or low filesystem activity.

A dedicated section outlines your GPU usage metrics. This includes Streaming Multiprocessors (SM) which indicates the percentage of time a GPU executes a warp (ideally around 80%). SM occupancy, the ratio of assigned to maximum warps, is generally expected around 50%. The Tensor setting should be as high as possible, indicating optimal use of the GPU's tensor cores for matrix operations. For FP64, FP32, and FP16 floating-point operations, significant activity should be observed on only one, based on your code's specified precision.

Details are provided on GPU memory usage and GPU memory access cycles, showing the percentage of cycles the device's memory interface is active sending or receiving data.

A power graph illustrates the GPU's power consumption (in watts) over time.

GPU bandwidth on the PCIe (Peripheral Component Interconnect Express) bus is also presented.

Node-wide resource statistics may be inaccurate if the node is shared. The data includes the evolution of job bandwidth over time, relating to software and licenses, and the evolution of network bandwidth used by jobs via the Infiniband network, where periods of massive data transfer (e.g., Lustre filesystem I/O, MPI communication) can be observed.

Further statistics show the evolution of local disk IOPS and local disk bandwidth (data read/written per second) over time.

Information on local disk space usage is also available.

Power consumption data is provided.

Account statistics¶

The Account Statistics section shows your group's usage in two subsections: CPU and GPU.

CPU accounts¶

This section provides the total number of CPU cores requested by your group and their usage over the past few months, allowing you to track your priority status as it fluctuates with usage.

A summary of the most frequently used applications is available.

Detailed resource usage by each user in your group is displayed.

CPU cores wasted by each user over time are tracked.

Memory usage by each user in your group is presented.

Memory wasted by each user is also reported.

Filesystem activity is represented, including the number of disk write commands (input/output operations per second (IOPS)) and the amount of data transferred (Bandwidth) over a given period.

This lists the last jobs run by all members of the group.

GPU accounts¶

This section presents the total GPU requests for your group and their usage over the past few months, enabling you to track your priority as it changes with usage.

A summary of the most frequently used software is provided.

Detailed resource usage by each user in your group is displayed.

The quantity of GPUs wasted by each user is tracked.

CPU resources allocated and used by your GPU jobs are presented.

CPUs wasted by your GPU jobs are reported.

Memory usage by each user in your group is displayed.

Memory wasted by each user is also reported.

Filesystem activity is represented, including the number of disk write commands (input/output operations per second (IOPS)) and the amount of data transferred (Bandwidth) over a given period.

Here you see the last jobs that were run by your group.

Cloud statistics¶

The table Your Instances displays all the virtual machines associated with your account. The Flavor column refers to the virtual machine type. The UUID column is a unique identifier assigned to each virtual machine.

Each virtual machine has its own usage statistics (CPU cores, memory, disk bandwidth, IOPS, and network bandwidth) that can be shown for the last month, week, day, or hour.