Method and system for routing service requests based on throughput of teams of servers
View Patent ↗A method and system are disclosed for creating portfolios of requests so as to reduce or minimize the variance in the workloads generated by those requests. The invention also takes into account the throughput of the servers that are servicing the requests. In a preferred embodiment, the method comprises the steps of establishing a set of system parameters; using said parameters to generate a model, said model including a defined optimization problem; and solving said optimization problem to output a set of mixing weights. The set of parameters and said mixing weights are used to generate a task; and the service requests are distributed to different service teams according to the generated task. Also, for example, the distributing may be done by distributing requests of type i to team j with weight proportional to w ij .
1. A method of organizing and routing a multitude of service requests of a multitude of different types, to a plurality of teams of servers, the method comprising the steps of:
establishing a set of system parameters;
using said parameters to generate a model, said model including a defined optimization problem with the goal of finding teams of servers that optimizes variance of total processing;
solving said optimization problem to output a set of mixing weights;
using said set of parameters and said mixing weights to generate a task; and
distributing the service requests to different service teams according to the generated task,
wherein the optimization problem is:
find values w ij (team i, request j) satisfying the condition:
min sum i=1,l (sum j=1 w ij sd j ) 2 ,
where:
w ij >=0,
w ij <=n ij ,
sum w ij =N j ,
and where: the requests are denoted as r 1 , . . . , r k ,
the teams of servers are denoted as t 1 , . . . , t l ,
each team i is able to provide throughput n ij for the requests r 1 , . . . , r k ,
a load is characterized by counts N 1 , . . . , N k for the requests r 1 , . . . , r k , and
sd j is the standard deviation in the distribution of the throughputs of the server teams for processing request j.
2. A method according to claim 1 , wherein the distribution step includes the step of distributing requests of type i to team j with weight proportional to wij.
3. A method according to claim 2 , wherein the routing of the service requests based on throughput of the teams of servers provides near zero variance in the workload.
4. A method according to claim 2 , wherein the routing of the service requests provides workload with minimal possible variance for a given IPC management team.
5. A method according to claim 2 , wherein the distributing step includes the step of building portfolios of the queues of the requests with minimal processing variance.
6. A system for organizing and routing a multitude of service requests of a multitude of different types, to a plurality of teams of servers, the system comprising:
a front-end component for establishing a set of system parameters;
a modeler for using said parameters to generate a model, said model including a defined optimization problem with the goal of finding teams of servers that optimizes variance of total processing;
an optimizer for solving said optimization problem to output a set of mixing weights;
a task generator for using said set of parameters and said mixing weights to generate a task; and
a workload balancer for distributing the service requests to different service teams according to the generated task;
wherein the optimization problem is:
find values w ij (team i, request j) satisfying the condition:
min sum i=1,l (sum j=1 w ij sd j ) 2 ,
where:
w ij >=0,
w ij <=n ij ,
sum w ij =N j ,
and where: the requests are denoted as r 1 , . . . , r k ,
the teams of servers are denoted as t 1 , . . . , t l ,
each team i is able to provide throughput n ij for the requests r 1 , . . . , r k ,
a load is characterized by counts N 1 , . . . , N k for the requests r 1 . . . , r k , and
sd j is the standard deviation in the distribution of the throughputs of the server teams for processing request j.
7. A system according to claim 6 , wherein the workload balancer distributes requests of type i to team j with weight proportional to w ij .