SQLVM: Performance Isolation in Multi-Tenant Relational Database-as-a-Service

Établi : February 14, 2013

Multi-tenancy and resource sharing are essential to make a Database-as-a-Service (DaaS). However, resource sharing usually results in the performance of one tenant’s workload to be affected by other co-located tenants. In the SQLVM project, our approach to performance isolation in a DaaS is to isolate the key resources, such as CPU, I/O and memory, needed by the tenants’ workload. Mechanisms designed in the SQLVM project are now in production in Azure SQL Database Service Tiers

In today’s relational Database-as-a-Service providers such as Microsoft Azure SQL Database, resources of a single database server are shared among workloads from multiple tenants. This multi-tenancy enables cost reduction for the cloud service provider which it can pass on as savings to the tenants. However, resource sharing can adversely affect a tenant’s performance due to resource demands of other tenants’ workloads. Service providers today do not provide any assurances to a tenant in terms of isolating its performance from other co-located tenants.

SQLVM is a mechanism for performance isolation via a promise of reservation of key database server resources, such as CPU, I/O, and memory for each tenant. One key challenge is supporting this abstraction within the database server process without statically allocating resources to tenants while scaling to large numbers of tenants. Another challenge is to devise low-overhead metering mechanisms to objectively establish accountability if a resource reservation is not met. SQLVM can be viewed as a building block upon which higher-level performance assurances (e.g., at the workload level) can be designed.

Business Impact

Technology developed in the SQLVM project formed the basis of CPU and I/O performance-isolation capabilities in multi-tenant Microsoft Azure SQL Database (opens in new tab). The new Service Tiers and Performance Levels (opens in new tab), made generally available in September 2014, relies on these capabilities to enforce improved performance levels. New I/O Resource Governor functionality (opens in new tab) based on this technology has shipped with SQL Server 2014.

Personne

Portrait de Arnd Christian König

Arnd Christian König

Principal Researcher

Portrait de Manoj Syamala

Manoj Syamala

Principal Software Development Engineer

Portrait de Surajit Chaudhuri

Surajit Chaudhuri

Technical Fellow, Data Platforms and Analytics

Portrait de Vivek Narasayya

Vivek Narasayya

Partner Research Manager