GCP Service Quotas
Google Cloud imposes default quotas on services for all new projects and new accounts, and in specific regions. It is very important to check your current quotas & limits prior to starting your CloudyCluster environment. Low quotas are one of the more common issues clients encounter with CloudyCluster.
Quotas are visible in the console, and they may need to be increased. If your CloudyCluster configuration exceeds the limits, your cluster will not spin up and will roll back. See GCP Service Quotas KB for more information or to request a limit increase.
Check/Adjust Project Quotas
Some projects have resource limits in place, which are called quotas. These quotas must be high enough to support the cluster’s workloads. In some cases, the user must contact Google to adjust the quotas. other times, the user will be able to adjust them. Make sure to adjust the quotas for the Compute Engine API.
For more information on quotas, see the documentation.
Below is a table outlining the minimum quotas for each standard cluster size. This allows for a base cluster to be set up.
Size Persistent Disk SSD CPUs RAM VM Instances Test 300GB 4 16GB 4 Small 2330GB 10 40GB 5 Medium 4400GB 80 320GB 5 Large 10400GB 112 450GB 7 XL 20610GB 176 710GB 11
You will need to increase your quotas to accommodate your actual workloads as well. To run jobs, your quotas will have to be increased.
Fortunately, it is fairly simple to calculate and adjust quotas. Read our blog post on this process to learn more.
Note: Not all machine types and processors are available in all regions. We recommend checking with Google if you are uncertain which machines have the resources you plan on using for your project. Machine Types CloudyCluster has been test validated on N2, C2 and E2 machine types.
Prior to starting your CloudyCluster setup, you should know the basic requirements for your HPC tasks. The size of your job and your desired time to results will determine the types and quantities of resources needed to run your job. Example: (n)nodes = x (n)tasks per node = y (n)CPUs or GPUs
If you don’t know the resources required, you can start with one of the CloudyCluster Quick Start configurations and try it out to get a feel for what you need to complete your job within your desired time frame. You may also contact CloudyCluster Services Team for guidance. support [@] cloudycluster.com