Best flow for notifications

Question

HI, I have a custom service with customs characteristics and I need the maximum speed posible.

I have one service for write without response and another service only with notifications.

I have the min_conn_interval and the max_conn_interval at 7.5.

The NRF_SDH_BLE_GATT_MAX_MTU_SIZE is in 247.

The NRF_SDH_BLE_GAP_DATA_LENGTH is in 251.

On advertising_init() I have setted the init.config.ble_adv_primary_phy = BLE_GAP_PHY_1MBPS.

Also I have the DLE enabled.

And I have two questions :

I'm using a boolean as semaphore, It turn off each time that I update the characteristic, and after that I turn on again with the event BLE_GATTS_EVT_HVN_TX_COMPLETE, but doing that the data trans its so so slow. But, if I don't do this I lost a lot of data packages

I use the next function to update the notify characteristic:

uint32_t update_acc_read_value(ble_cus_t *p_cus, uint8_t *custom_value) {
    if (p_cus == NULL) {
        return NRF_ERROR_NULL;
    }
    uint32_t
    err_code = NRF_SUCCESS;
    // Send value if connected and notifying.
    if ((p_cus->conn_handle != BLE_CONN_HANDLE_INVALID)) {

        uint16_t
        length = 1 + strlen(custom_value);
        isNotificationReaded = false; // SEMAPHORE = FALSE
        ble_gatts_hvx_params_t hvx_params;
        memset(&hvx_params, 0, sizeof(hvx_params));
        hvx_params.handle = p_cus->custom_value_acc.value_handle;
        hvx_params.type = BLE_GATT_HVX_NOTIFICATION;
        hvx_params.p_len = &length;
        hvx_params.p_data = custom_value;

        err_code = sd_ble_gatts_hvx(p_cus->conn_handle, &hvx_params);
    } else {
        err_code = NRF_ERROR_INVALID_STATE;
    }
    return err_code;
}

With the next function the data transfer is 'faster' but I get repeated data.

I get in 50 seconds 5780 packages * 246 bytes = 1421880 bytes = 11375050 bits,

11375050 / 50 = 22750 bits/second

void test_mode() {
    NRF_LOG_INFO("TEST")
    nrf_delay_ms(1000);
    isNotificationReaded = true;

    char buff[247];
    for (int i = 0; i < 246; i++) {
        buff[i] = 'A';
    }
    for (int i = 0; i < 4000;) {
        update_acc_read_value(&m_cus, buff);
        if (isNotificationReaded) {
            i++;
        }
    } 
}

But if I use the function correctly with my semaphore flow:

I get in 63 seconds 4000 packages * 246 bytes = 984000 bytes = 7872000 bits,

7872000 / 63 = 124952 bits/second

void test_mode() {
    NRF_LOG_INFO("TEST")
    nrf_delay_ms(1000);
    isNotificationReaded = true;

    char buff[247];
    for (int i = 0; i < 246; i++) {
        buff[i] = 'A';
    }
    for (int i = 0; i < 4000;) {
        if (isNotificationReaded) {
        update_acc_read_value(&m_cus, buff);
            i++;
        }
    } 
}

So, My questions are :

- Its correct my semaphore flow ? If is not correct, how can I send fasters notifications but without data loss ?

- How can I improve the speed even more ? Because I wasted a few days researching about that and I only find that, time_interval min as posible and packages larger as posible, DLE enabled and not much more.

Im developing with raspberry pi as ble_server but I have tested on iPhone and android and the results are even worst.

Thanks a lot !