System and method for SLA-aware database consolidation using per-tenant memory size configuration

1. A method for consolidating workload on one or more computing devices, comprising:

generating tenant performance profiles for at least one of a buffer memory size and a buffer warmness;

assigning one or more tenants to a database management system (DBMS) instance;

configuring the one or more tenants to minimize a total server count needed to host the one or more tenants;

reducing the total server count by adjusting of buffer sizes while satisfying Service Level Agreement (SLA) requirements;

packing tenants as a two-dimensional vector packing problem (2-DVPP) where each tenant is represented by a 2-dimensional vector whose first dimension is a buffer size, and whose second dimension is a load; and

determining a minimum number of servers needed by determining a memory assignment function which assigns a buffer size to each tenant such that the optimum output of the tenant placement 2-DVPP is minimized; and

packing of tenants into servers, where buffer sizes m* of tenants are determined by a memory assignment function m

m

*

=

arg

⁢

⁢

min

m

⁢

{

Opt_

⁢

2

⁢

DV

⁢

⁢

P

⁡

(

m

)

}

2. The method as recited in claim 1 , further comprising adjusting of a private buffer size to achieve server count reduction.

3. The method as recited in claim 1 , further comprising buffer sharing among two or more tenants and adjusting of a shared buffer size to achieve server count reduction.

4. The method as recited in claim 1 , further comprising restricting the one or more tenants running on each disk to one tenant at a time.

5. The method as recited in claim 1 , further comprising employing multiple database management system (DBMS) instances on a same server to impose controlled boundaries between buffer pools of different tenants or groups of tenants.

6. A system for consolidating workload on one or more computing devices, comprising:

a profiler configured to generate tenant performance profiles, using a processor, for at least one of a buffer memory size and a buffer warmness;

one or more tenants assigned to a database management system (DBMS), the one or more tenants configured to minimize a total server count needed to host the one or more tenants, wherein tenants are packed as a two-dimensional vector packing problem (2-DVPP) where each tenant is represented by a 2-dimensional vector whose first dimension is a buffer size, and whose second dimension is a load;

code for determining a minimum number of servers needed by determining a memory assignment function which assigns a buffer size to each tenant such that the optimum output of the tenant placement 2-DVPP is minimized;

one or more buffer pools configured to reduce the total server count through optimization of buffer sizes while satisfying Service Level Agreement (SLA) requirements; and

packing of tenants into servers, where buffer sizes m* of tenants are determined by a memory assignment function m

m

*

=

arg

⁢

⁢

min

m

⁢

{

Opt_

⁢

2

⁢

DV

⁢

⁢

P

⁡

(

m

)

}

7. The system as recited in claim 6 , further comprising a buffer configured to adjust a private buffer size to achieve server count reduction.

8. The system as recited in claim 6 , further comprising a buffer configured to adjust a shared buffer size and to share the buffer among two or more tenants to achieve server count reduction.

9. The system as recited in claim 6 , further comprising one or more disks configured to restrict the one or more tenants running on each disk to one at a time.

10. The system as recited in claim 6 , further comprising a server configured to impose controlled boundaries between buffer pools of different tenants or groups of tenants by employing multiple database management system (DBMS) instances on the same server.