15.7. Reduce memory usage

A great number of modern systems have tight constraints on available memory, making the reduction of memory usage to a minimum critical. Reducing memory consumption of a Fast DDS application can be achieved through various approaches, mainly through architectural restructuring of the application, but also by limiting the resources the middleware utilizes, and by avoiding static allocations.

15.7.1. Limiting Resources

The ResourceLimitsQosPolicy controls the resources that the service can use in order to meet the requirements imposed. It limits the amount of allocated memory per DataWriter or DataReader, as per the following parameters:

max_samples: Configures the maximum number of samples that the DataWriter or DataReader can manage across all the instances associated with it, i.e. it represents the maximum samples that the middleware can store for a DataReader or DataWriter.
max_instances: Configures the maximum number of instances that the DataWriter or DataReader can manage.
max_samples_per_instance: Controls the maximum number of samples within an instance that the DataWriter or DataReader can manage.
allocated_samples: States the number of samples that will be allocated on initialization.

All these parameters may be lowered as much as needed to reduce memory consumption, limit the resources to the application’s needs. Below is an example of a configuration for the minimum resource limits possible.

Warning

The value of max_samples must be higher or equal to the value of max_samples_per_instance.
The value established for the HistoryQosPolicy depth must be lower or equal to the value stated for max_samples_per_instance.

C++

ResourceLimitsQosPolicy resource_limits;

// The ResourceLimitsQosPolicy is constructed with max_samples = 5000 by default
// Change max_samples to the minimum
resource_limits.max_samples = 1;

// The ResourceLimitsQosPolicy is constructed with max_instances = 10 by default
// Change max_instances to the minimum
resource_limits.max_instances = 1;

// The ResourceLimitsQosPolicy is constructed with max_samples_per_instance = 400 by default
// Change max_samples_per_instance to the minimum
resource_limits.max_samples_per_instance = 1;

// The ResourceLimitsQosPolicy is constructed with allocated_samples = 100 by default
// No allocated samples
resource_limits.allocated_samples = 0;

XML

<?xml version="1.0" encoding="UTF-8" ?>
<profiles xmlns="http://www.eprosima.com">
    <data_writer profile_name="data_writer_min_samples">
        <topic>
            <historyQos>
                <kind>KEEP_LAST</kind>
                <depth>1</depth>
            </historyQos>
            <resourceLimitsQos>
                <max_samples>1</max_samples>
                <max_instances>1</max_instances>
                <max_samples_per_instance>1</max_samples_per_instance>
                <allocated_samples>0</allocated_samples>
            </resourceLimitsQos>
        </topic>
    </data_writer>

    <data_reader profile_name="data_reader_min_samples">
        <topic>
            <historyQos>
                <kind>KEEP_LAST</kind>
                <depth>1</depth>
            </historyQos>
            <resourceLimitsQos>
                <max_samples>1</max_samples>
                <max_instances>1</max_instances>
                <max_samples_per_instance>1</max_samples_per_instance>
                <allocated_samples>0</allocated_samples>
            </resourceLimitsQos>
        </topic>
    </data_reader>
</profiles>

15.7.2. Set Dynamic Allocation

By default MemoryManagementPolicy is set to PREALLOCATED_WITH_REALLOC_MEMORY_MODE, meaning that the amount of memory required by the configured ResourceLimitsQosPolicy will be allocated at initialization. If some more memory has to be allocated at run time, it is reallocated.

Using the dynamic settings of the RTPSEndpointQos will prevent unnecessary allocations. Lowest footprint is achieved with DYNAMIC_RESERVE_MEMORY_MODE at the cost of higher allocation counts, in this mode memory is allocated when needed and freed as soon as it stops being used. For higher determinism at a small memory cost the DYNAMIC_REUSABLE_MEMORY_MODE option is available, this option is similar but once more memory is allocated it is not freed and is reused for future messages.

C++

RTPSEndpointQos endpoint;
endpoint.history_memory_policy = eprosima::fastdds::rtps::DYNAMIC_REUSABLE_MEMORY_MODE;

XML

<?xml version="1.0" encoding="UTF-8" ?>
<profiles xmlns="http://www.eprosima.com">
    <data_writer profile_name="data_writer_low_memory">
        <!-- ...  -->
        <historyMemoryPolicy>DYNAMIC_REUSABLE</historyMemoryPolicy>
    </data_writer>

    <data_reader profile_name="data_reader_low_memory">
        <!-- ...  -->
        <historyMemoryPolicy>DYNAMIC_REUSABLE</historyMemoryPolicy>
    </data_reader>
</profiles>