ortools.math_opt.python.solver_resources
Configures solver resources.
1# Copyright 2010-2024 Google LLC 2# Licensed under the Apache License, Version 2.0 (the "License"); 3# you may not use this file except in compliance with the License. 4# You may obtain a copy of the License at 5# 6# http://www.apache.org/licenses/LICENSE-2.0 7# 8# Unless required by applicable law or agreed to in writing, software 9# distributed under the License is distributed on an "AS IS" BASIS, 10# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 11# See the License for the specific language governing permissions and 12# limitations under the License. 13 14"""Configures solver resources.""" 15 16import dataclasses 17from typing import Optional 18 19from ortools.math_opt import rpc_pb2 20 21 22@dataclasses.dataclass 23class SolverResources: 24 """The hints on the resources a remote solve is expected to use. 25 26 These parameters are hints and may be ignored by the remote server (in 27 particular in case of solve in a local subprocess, for example). 28 29 When using remote_solve() and remote_compute_infeasible_subsystem(), these 30 hints are mostly optional as some defaults will be computed based on the other 31 parameters. 32 33 When using remote_streaming_solve() these hints are used to dimension the 34 resources available during the execution of every action; thus it is 35 recommended to set them. 36 37 MOE:begin_intracomment_strip 38 39 The go/uoss server will use these parameters to do a bin-packing of all 40 requests. Parameter cpu is a soft-limit, the solve may still be able to use 41 more CPUs. The ram parameter is an hard-limit, an out-of-memory error will 42 occur if the solve attempts to use more memory. 43 44 MOE:end_intracomment_strip 45 46 Attributes: 47 cpu: The number of solver threads that are expected to actually execute in 48 parallel. Must be finite and >0.0. For example a value of 3.0 means that 49 if the solver has 5 threads that can execute we expect at least 3 of these 50 threads to be scheduled in parallel for any given time slice of the 51 operating system scheduler. A fractional value indicates that we don't 52 expect the operating system to constantly schedule the solver's work. For 53 example with 0.5 we would expect the solver's threads to be scheduled half 54 the time. This parameter is usually used in conjunction with 55 SolveParameters.threads. For some solvers like Gurobi it makes sense to 56 use SolverResources.cpu = SolveParameters.threads. For other solvers like 57 CP-SAT, it may makes sense to use a value lower than the number of threads 58 as not all threads may be ready to be scheduled at the same time. It is 59 better to consult each solver documentation to set this parameter. Note 60 that if the SolveParameters.threads is not set then this parameter should 61 also be left unset. 62 ram: The limit of RAM for the solve in bytes. Must be finite and >=1.0 (even 63 though it should in practice be much larger). 64 """ 65 66 cpu: Optional[float] = None 67 ram: Optional[float] = None 68 69 def to_proto(self) -> rpc_pb2.SolverResourcesProto: 70 return rpc_pb2.SolverResourcesProto(cpu=self.cpu, ram=self.ram)
@dataclasses.dataclass
class
SolverResources:
23@dataclasses.dataclass 24class SolverResources: 25 """The hints on the resources a remote solve is expected to use. 26 27 These parameters are hints and may be ignored by the remote server (in 28 particular in case of solve in a local subprocess, for example). 29 30 When using remote_solve() and remote_compute_infeasible_subsystem(), these 31 hints are mostly optional as some defaults will be computed based on the other 32 parameters. 33 34 When using remote_streaming_solve() these hints are used to dimension the 35 resources available during the execution of every action; thus it is 36 recommended to set them. 37 38 MOE:begin_intracomment_strip 39 40 The go/uoss server will use these parameters to do a bin-packing of all 41 requests. Parameter cpu is a soft-limit, the solve may still be able to use 42 more CPUs. The ram parameter is an hard-limit, an out-of-memory error will 43 occur if the solve attempts to use more memory. 44 45 MOE:end_intracomment_strip 46 47 Attributes: 48 cpu: The number of solver threads that are expected to actually execute in 49 parallel. Must be finite and >0.0. For example a value of 3.0 means that 50 if the solver has 5 threads that can execute we expect at least 3 of these 51 threads to be scheduled in parallel for any given time slice of the 52 operating system scheduler. A fractional value indicates that we don't 53 expect the operating system to constantly schedule the solver's work. For 54 example with 0.5 we would expect the solver's threads to be scheduled half 55 the time. This parameter is usually used in conjunction with 56 SolveParameters.threads. For some solvers like Gurobi it makes sense to 57 use SolverResources.cpu = SolveParameters.threads. For other solvers like 58 CP-SAT, it may makes sense to use a value lower than the number of threads 59 as not all threads may be ready to be scheduled at the same time. It is 60 better to consult each solver documentation to set this parameter. Note 61 that if the SolveParameters.threads is not set then this parameter should 62 also be left unset. 63 ram: The limit of RAM for the solve in bytes. Must be finite and >=1.0 (even 64 though it should in practice be much larger). 65 """ 66 67 cpu: Optional[float] = None 68 ram: Optional[float] = None 69 70 def to_proto(self) -> rpc_pb2.SolverResourcesProto: 71 return rpc_pb2.SolverResourcesProto(cpu=self.cpu, ram=self.ram)
The hints on the resources a remote solve is expected to use.
These parameters are hints and may be ignored by the remote server (in particular in case of solve in a local subprocess, for example).
When using remote_solve() and remote_compute_infeasible_subsystem(), these hints are mostly optional as some defaults will be computed based on the other parameters.
When using remote_streaming_solve() these hints are used to dimension the resources available during the execution of every action; thus it is recommended to set them.
MOE:begin_intracomment_strip
The go/uoss server will use these parameters to do a bin-packing of all requests. Parameter cpu is a soft-limit, the solve may still be able to use more CPUs. The ram parameter is an hard-limit, an out-of-memory error will occur if the solve attempts to use more memory.
MOE:end_intracomment_strip
Attributes:
- cpu: The number of solver threads that are expected to actually execute in parallel. Must be finite and >0.0. For example a value of 3.0 means that if the solver has 5 threads that can execute we expect at least 3 of these threads to be scheduled in parallel for any given time slice of the operating system scheduler. A fractional value indicates that we don't expect the operating system to constantly schedule the solver's work. For example with 0.5 we would expect the solver's threads to be scheduled half the time. This parameter is usually used in conjunction with SolveParameters.threads. For some solvers like Gurobi it makes sense to use SolverResources.cpu = SolveParameters.threads. For other solvers like CP-SAT, it may makes sense to use a value lower than the number of threads as not all threads may be ready to be scheduled at the same time. It is better to consult each solver documentation to set this parameter. Note that if the SolveParameters.threads is not set then this parameter should also be left unset.
- ram: The limit of RAM for the solve in bytes. Must be finite and >=1.0 (even though it should in practice be much larger).