● It has a simpler architecture within Snowflake. Simplicity matters because object
proliferation makes managing myriad objects increasingly difficult over time. With MTT,
adding tenants does not cause the number of objects to grow, but adding tenants to OPT
and APT can result in hundreds or thousands of objects being created within Snowflake.
From a cost standpoint, MTT is usually more cost-efficient because multiple customers utilize
shared compute and other resources more efficiently.
But MTT has a somewhat rigid requirement: To use MTT, an app's data model has to have the same
general shape across all tenants. Application builders can achieve slight variances using custom
columns that only apply to certain types of tenants, but this approach introduces sparsity into the
data.
Object per tenant (OPT)
OPT is a great fit if each tenant has a different data model. Unlike MTT, the tenant data shape can
be unique for each tenant. OPT does not scale as easily as MTT, however. OPT typically scales well
from tens to hundreds of tenants, but starts to become unwieldy when it includes thousands of
tenant databases.
Security can factor into the decision to use an OPT design pattern. Some customers prefer the
OPT model because they don't want to manage an entitlement table, secure views, or row-level
security with strong processes behind it. They are, however, comfortable using RBAC to control
who has specific access to a database.
Some apps that use the OPT model give customers their own dedicated compute resources to
satisfy contractual, security, or regulatory requirements.
Account per tenant (APT)
APT isolates tenants at the account level. Typically, customers have a strong security reason for
choosing this approach. For example, organizations bound by strict regulatory mandates may
choose this option if:
● They need to implement a dedicated connection string per tenant
● They require security measures such as Bring Your Own Tool (BYOT)
● They want to use per-tenant IP restrictions at the account level
APT requires the customer to also implement OPT, which can support a huge variety of tenant
data shapes. In addition, APT introduces more scaling limitations—tenant counts in the tens to low
hundreds are typical, however, customers with higher tenant numbers exist. APT can become
unwieldy when managing thousands of tenant accounts.
Page 3