Self-Hosted GCP
Supported GCP Regions
This document covers the Google Cloud Platform (GCP) regions and zones validated and supported by Glean for tenant deployment
Overall, there are two types of GCP regions for hosting Glean: Preferred and Non-Preferred regions.
Glean recommends that our customers leverage a Preferred Region for hosting their Glean tenant unless a specific country is required for data sovereignty purposes.
If you opt to use a non-preferred region:
The default region for tenants hosted by Glean is
us-central-1, which is located in North America. Please advise Glean if you would like your tenant deployed in an different region.Considerations for Region Selection
There are two main considerations that drive support for Glean in different regions:- GPU / TPU availability
- LLM model availability in Vertex AI
LLM model availability is only a consideration if you wish to use Vertex AI for access to LLM models. If you are using Glean’s LLM key or your own Azure OpenAI, OpenAI, or Anthropic account key, you can disregard this requirement.
Preferred Regions
Preferred regions are GCP regions that support the use of compatible TPUs:Tensor Processing Units (TPUs) are Google’s custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads. TPUs are more efficient and cost-effective than using GPUs for Machine Learning (M/L) workflows.
- More information: Introduction to Cloud TPU (cloud.google.com).
| Region | Location | GCP Region | GCP Zone |
|---|---|---|---|
| AMER | USA (Iowa) | us-central1 | us-central1-a |
| APAC | Taiwan (Changhua County) | asia-east1 | asia-east1-c |
| EMEA | Netherlands (Eemshaven) | europe-west4 | europe-west4-a |
Non-Preferred Regions
Non-Preferred regions are GCP regions that do not support TPUs. In these regions Glean must use GPUs for M/L workflows, and these attract a higher usage cost due to a reduction in efficiency when compared to TPUs.| Location | GCP Region | GCP Zone |
|---|---|---|
| Brazil (Sao Paulo) | southamerica-northeast1 | southamerica-northeast1-a,c |
| Canada (Montreal) | northamerica-northeast1 | northamerica-northeast1-c |
Where possible, you should elect to use the first GCP zone referenced as some zones contain higher memory GPUs that are better suited for M/L workflows than others.
- For Glean hosted tenants: Your subscription cost will be higher to offset the higher infrastructure costs of these regions.
- For Cloud-prem tenants: You should anticipate higher monthly infrastructure costs for the Glean component of your GCP Billing Account.
LLM Model Availability in Vertex AI
When selecting an LLM for use with your Glean environment, if you wish for the LLM to be accessed via Vertex AI, you should consider the availability of your desired model within Vertex AI for each GCP region.- More information: Generative AI on Vertex AI locations (cloud.google.com)