Public API

DLB offers a public interface for C, C++ and Fortran. The DLB API can be divided into:

Basic set
The basic set contains the general purpose functions that are common to other modules. The different functions are explained in detail in section DLB Basic API.
LeWI: Lend When Idle
The LeWI API is oriented to be used by runtimes to manage the CPU sharing between other processes but can also be used on applications to use the LeWI algorithm. These functions are explained in detail in section LeWI API.
DROM: Dynamic Resource Ownership Manager
The DROM API manages the CPU ownership of each DLB running process. For a more detailed description see Dynamic Resource Ownership Manager (DROM). These functions are described in section Dynamic Resource Manager Interface.
TALP: Tracking Application Live Performance
The TALP API is used to obtain measured metrics from other processes as well as to define custom monitoring regions, see Tracking Application Live Performance (TALP). These functions are described in section TALP Interface.

This is a specific API for MPI. We offer an MPI interface that will be called by Extrae if we are tracing the application or internally in the MPI intercept API. All the calls of this API are of the form shown below, and thus not documented.

  • DLB_<mpi_call_name>_enter(…)
  • DLB_<mpi_call_name>_leave(…)


These functions make the basic API to be used independently from which DLB mode is enabled.

int DLB_Init(int ncpus, const_dlb_cpu_set_t mask, const char *dlb_args)

Initialize DLB library and all its internal data structures. Must be called once and only once by each process in the DLB system.

int DLB_Finalize(void)

Finalize DLB library and clean up all its data structures. Must be called by each process before exiting the system.

int DLB_Enable(void)

Enable DLB and all its features in case it was previously disabled, otherwise it has no effect. It can be used in conjunction with DLB_Disable to delimit sections of the code where DLB calls will not have effect.

int DLB_Disable(void)

Disable DLB actions for the calling process. This call resets the original resources for the process and returns any external CPU it may be using at that time. While DLB is disabled there will not be any resource sharing for this process.

int DLB_SetMaxParallelism(int max)

Set the maximum number of resources to be used by the calling process. Useful to delimit sections of the code that the developer knows that only a maximum number of CPUs can benefit the execution. If a process reaches its maximum number of resources used at any time, subsequent calls to borrow CPUs will be ignored until some of them are returned.

int DLB_CallbackSet(dlb_callbacks_t which, dlb_callback_t callback, void *arg)
int DLB_CallbackGet(dlb_callbacks_t which, dlb_callback_t *callback, void **arg)

Setter and Getter for DLB callbacks. See section Callbacks.

int DLB_PollDROM(int *ncpus, dlb_cpu_set_t mask)
int DLB_PollDROM_Update(void)

Poll DROM module to check if the process needs to adapt to a new mask or number of CPUs.

int DLB_SetVariable(const char *variable, const char *value)
int DLB_GetVariable(const char *variable, char *value)

Set or get a DLB internal variable. These variables are the same ones specified in DLB_ARGS, although not all of them can be modified at runtime. If the variable is readonly the setter function will return an error.

int DLB_PrintVariables(int print_extra)
int DLB_PrintShmem(int num_columns, dlb_printshmem_flags_t print_flags)

Print to stdout the information about the DLB internal variables and the status of the shared memories.

const char* DLB_Strerror(int errnum)

Obtain a string that describes the error code passed in the argument.


These functions are used to manage the CPU sharing between processes. Generally, each action may have up to four different variants depending if the action is:

  1. for all possible CPUs (no suffix)
  2. for a specified CPU (Cpu suffix)
  3. for a determined number of CPUs (Cpus suffix)
  4. for a specified CPU mask (CpuMask suffix)
int DLB_Lend(void)
int DLB_LendCpu(int cpuid)
int DLB_LendCpus(int ncpus)
int DLB_LendCpuMask(const_dlb_cpu_set_t mask)

Lend CPUs of the process to the system. A lent CPU may be assigned to other process that demands more resources. If the CPU was originally owned by the process it may be reclaimed.

int DLB_Reclaim(void)
int DLB_ReclaimCpu(int cpuid)
int DLB_ReclaimCpus(int ncpus)
int DLB_ReclaimCpuMask(const_dlb_cpu_set_t mask)

Reclaim CPUs that were previously lent. It is mandatory that the CPUs belong to the calling process.

int DLB_AcquireCpu(int cpuid)
int DLB_AcquireCpus(int ncpus)
int DLB_AcquireCpuMask(const_dlb_cpu_set_t mask)

Acquire CPUs from the system. If the CPU belongs to the process the call is equivalent to a reclaim action. Otherwise the process attempts to acquire a specific CPU in case it is available or enqueue a request if it’s not.

int DLB_Borrow(void)
int DLB_BorrowCpu(int cpuid)
int DLB_BorrowCpus(int ncpus)
int DLB_BorrowCpuMask(const_dlb_cpu_set_t mask)

Borrow CPUs from the system only if they are idle. No other action is done if the CPU is not available.

int DLB_Return(void)
int DLB_ReturnCpu(int cpuid)
int DLB_ReturnCpuMask(const_dlb_cpu_set_t mask)

Return CPUs to the system commonly triggered by a reclaim action from other process but stating that the current process still demands the usage of these CPUs. This action will enqueue a request for when the resources are available again. If the caller does not want to keep the resource after receiving a reclaim, the correct action is lend.

Dynamic Resource Manager Interface

The next set of functions can be used when the user has enabled the Dynamic Resource Ownership Manager (DROM) Module (see Dynamic Resource Ownership Manager (DROM)). With this interface the user can set or retrieve the process mask of each DLB process.

int DLB_DROM_Attach(void)

Attach process to DLB as third party

int DLB_DROM_Detach(void)

Detach process from DLB

int DLB_DROM_GetNumCpus(int *ncpus)

Get the total number of available CPUs in the node

void DLB_DROM_GetPidList(int *pidlist, int *nelems, int max_len)

Get the PID’s attached to this module

int DLB_DROM_GetProcessMask(int pid, dlb_cpu_set_t mask, dlb_drom_flags_t flags)

Get the process mask of the given PID

int DLB_DROM_SetProcessMask(int pid, const dlb_cpu_set_t mask, dlb_drom_flags_t flags)

Set the process mask of the given PID

TALP Interface

The TALP interface is divided in two sets of services. The first set provides the functionality to obtain TALP data from an external process. This process needs first to attach to DLB and later it can obtain some data from the other DLB running processes.

int DLB_TALP_Attach(void)

Attach process to DLB as third party

int DLB_TALP_Detach(void)

Detach process from DLB

int DLB_TALP_GetNumCpus(int *ncpus)

Get the total number of available CPUs in the node

void DLB_TALP_GetPidList(int *pidlist, int *nelems, int max_len)

Get the PID’s attached to this module

int DLB_TALP_GetTimes(int pid, double *mpi_time, double *useful_time)

Get the CPU time on MPI and useful computation for the given process

The second set of services are designed to be called from witihn the DLB running proceses. With these funcions, the process can obtain live metrics from TALP, as well as to define new custom Monitoring Regions to delimit a specific part of the code.

const dlb_monitor_t* DLB_MonitoringRegionGetMPIRegion(void)

Get the pointer of the implicit MPI Monitorig Region

dlb_monitor_t* DLB_MonitoringRegionRegister(const char *name)

Register a new Monitoring Region

int DLB_MonitoringRegionReset(dlb_monitor_t *handle)

Reset monitoring region

int DLB_MonitoringRegionStart(dlb_monitor_t *handle)

Start (or unpause) monitoring region

int DLB_MonitoringRegionStop(dlb_monitor_t *handle)

Stop (or pause) monitoring region

int DLB_MonitoringRegionReport(const dlb_monitor_t *handle)

Print a Report by stdout of the monitoring region