Load estimation and balancing in virtualized radio access networks
During a first transmission time interval (TTI) of a vRAN, data traffic between a radio unit (RU) of a cellular network and a first vRAN instance of the vRAN is monitored. The first vRAN instance executes on a first server of the vRAN and the first vRAN instance is configured to perform PHY layer processing and L2 processing of the data traffic. Based on the data traffic between the RU of the cellular network and the first vRAN instance during the first TTI, a workload at the first vRAN instance during a second TTI is estimated.
1. A method for estimating workloads in a virtualized radio access network (vRAN), comprising:
monitoring, during a first transmission time interval (TTI) of the vRAN, data traffic between a radio unit (RU) of a cellular network and a first vRAN instance of the vRAN, the first vRAN instance executing on a first server of the vRAN, the first vRAN instance configured to perform PHY layer processing and L2 processing of the data traffic, wherein the PHY layer is Layer 1 of the Open Systems Interconnection (OSI) model and the L2 is Layer 2 of the OSI model, and wherein the TTI is a smallest time duration over which data is scheduled and transmitted between a base station of the cellular network (gNodeB or eNodeB) and user equipment (UE);
based on the data traffic between the RU of the cellular network and the first vRAN instance during the first TTI, estimating a workload at the first vRAN instance during a second TTI, wherein the second TTI is a next immediate TTI following the first TTI;
determining that the first vRAN instance does not have computing capacity to process the workload during the second TTI based on determining that a time required to process the estimated workload exceeds the second TTI;
in response to determining that the first vRAN instance does not have computing capacity to process the workload during the second TTI, allocating the workload to a second vRAN instance executing on a second server of the vRAN; and
causing the workload to be processed by the second vRAN instance executing on the second server of the vRAN during the second TTI.
2. The method of claim 1 , wherein the monitoring, estimating, and determining are performed by a programmable switch of the vRAN.
3. The method of claim 1 , wherein the data traffic comprises messages sent via a functional application platform interface (FAPI), wherein the FAPI is a standard interface between the PHY layer and the L2.
4. The method of claim 3 , wherein the estimating is based on one or more of signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size.
5. The method of claim 4 , wherein the estimating is performed using a prediction model trained with the signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size, and measured processing time.
6. The method of claim 1 , wherein the monitoring, estimating, and determining are performed in a data plane of the vRAN.
7. The method of claim 5 , wherein the prediction model comprises a tree-based classification model.
8. A programmable switch configured to perform operations comprising:
monitoring, during a first transmission time interval (TTI) of a vRAN, data traffic between a radio unit (RU) of a cellular network and a first vRAN instance of the vRAN, the first vRAN instance executing on a first server of the vRAN, the first vRAN instance configured to perform PHY layer processing and L2 processing of the data traffic, wherein the PHY layer is Layer 1 of the Open Systems Interconnection (OSI) model and the L2 is Layer 2 of the OSI model, and wherein the TTI is a smallest time duration over which data is scheduled and transmitted between a base station of the cellular network (gNodeB or eNodeB) and user equipment (UE);
based on the data traffic between the RU of the cellular network and the first vRAN instance during the first TTI, estimating a workload at the first vRAN instance during a second TTI, wherein the second TTI is a next immediate TTI following the first TTI;
determining that the first vRAN instance does not have computing capacity to process the workload during the second TTI based on determining that a time required to process the estimated workload exceeds the second TTI;
in response to determining that the first vRAN instance does not have computing capacity to process the workload during the second TTI, allocating the workload to a second vRAN instance executing on a second server of the vRAN; and
causing the workload to be processed by the second vRAN instance executing on the second server of the vRAN during the second TTI.
9. The programmable switch of claim 8 , wherein the data traffic comprises messages sent via a functional application platform interface (FAPI), wherein the FAPI is a standard interface between the PHY layer and the L2.
10. The programmable switch of claim 9 , wherein the estimating is based on one or more of signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size.
11. The programmable switch of claim 10 , wherein the estimating is performed using a prediction model trained with the signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size, and measured processing time.
12. The programmable switch of claim 8 , wherein the monitoring, estimating, and determining are performed in a data plane of the programmable switch.
13. The programmable switch of claim 11 , wherein the prediction model comprises a tree-based classification model.
14. A system comprising:
a programmable switch; and
a plurality of servers configured to execute a plurality of virtualized radio access network (vRAN) instances of a vRAN;
the programmable switch configured to:
monitor, during a first transmission time interval (TTI) of the vRAN, data traffic between a radio unit (RU) of a cellular network and a first vRAN instance executing on a first server, the first vRAN instance configured to perform PHY layer processing and L2 processing of the data traffic, wherein the PHY layer is Layer 1 of the Open Systems Interconnection (OSI) model and the L2 is Layer 2 of the OSI model, and wherein the TTI is a smallest time duration over which data is scheduled and transmitted between a base station of the cellular network (gNodeB or eNodeB) and user equipment (UE);
based on the data traffic between the RU of the cellular network and the first vRAN instance during the first TTI, estimate a workload at the first vRAN instance during a second TTI, wherein the second TTI is a next immediate TTI following the first TTI;
determine that the first vRAN instance does not have computing capacity to process the workload during the second TTI based on determining that a time required to process the estimated workload exceeds the second TTI;
in response to determining that the first vRAN instance does not have computing capacity to process the workload during the second TTI, allocate the workload to a second vRAN instance executing on a second server; and
cause the workload to be processed by the second vRAN instance executing on the second server of the vRAN during the second TTI.
15. The system of claim 14 , wherein the data traffic comprises messages sent via a functional application platform interface (FAPI), wherein the FAPI is a standard interface between the PHY layer and the L2.
16. The system of claim 15 , wherein the estimating is based on one or more of signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size.
17. The system of claim 16 , wherein the estimating is performed using a prediction model trained with the signal-to-noise ratio (SNR), type of modulation, type of coding scheme, number of channel layers used, or transmit block size, and measured processing time.
18. The system of claim 17 , wherein the monitoring,
estimating, and determining are performed in a data plane of the programmable switch.
19. The system of claim 17 , wherein the prediction model comprises a tree-based classification model.