Method and system for controlling hardware performance for high frequency trading
A method for adjusting a hardware performance for high frequency trading is provided, which is performed by one or more processors, and includes receiving context information including at least one of a traffic of market data for one or more target items or a computational complexity of using a machine learning model performed based on the market data, determining whether or not to change a processing performance of the hardware based on the received context information, and changing a resource supplied to the hardware based on the determination result for changing the processing performance.
1 . A method for adjusting a performance of hardware for high frequency trading, the method being performed by one or more processors and comprising:
receiving, by the one or more processors, context information including at least one of a traffic of market data for one or more target items or a computational complexity of using a machine learning model performed based on the market data;
determining, by the one or more processors, a workload for an inference computation of the machine learning model with at least one target item based on time allowed until a task is completed and available resources of the hardware, wherein the determining the workload includes determining one or more resources of the hardware for processing the workload;
determining, by the one or more processors, whether or not to change a processing performance of the hardware based on the received context information and the determined workload;
changing, by the one or more processors, one or more resources to be supplied to the hardware based on the determination result for changing the processing performance;
supplying, by the one or more processors, the one or more changed resources to the hardware using supply lines formed between the one or more changed resources and the hardware; and
performing, by using the hardware to which the one or more changed resources are supplied, the machine learning model based on the market data.
2 . The method according to claim 1 , wherein the changing the resource includes changing at least one of a speed of a clock or a voltage supplied to the hardware.
3 . The method according to claim 2 , wherein the determining whether or not to change the processing performance includes determining a processing performance required for the hardware based on the received context, and
the changing the resource includes adjusting at least one of the speed of the clock or the voltage supplied to the hardware according to the determined processing performance of the hardware.
4 . The method according to claim 3 , wherein the hardware includes a plurality of dedicated accelerators for computation of the machine learning model, and
the adjusting at least one of the speed of the clock or the voltage includes:
identifying one or more dedicated accelerators from among the plurality of dedicated accelerators based on the determined processing performance; and
adjusting at least one of the speed of the clock or the voltage supplied to the identified one or more dedicated accelerators.
5 . The method according to claim 3 , wherein the hardware includes at least one of a dedicated accelerator, a memory, or a processor, at least one of the dedicated accelerator, the memory, or the processor is divided into a plurality of processing regions associated with a plurality of layers included in the machine learning model, and
the adjusting at least one of the speed of the clock or the voltage includes:
identifying one or more processing regions from among a plurality of processing regions based on the determined processing performance; and
adjusting at least one of the speed of the clock or the voltage supplied to the identified one or more processing regions.
6 . The method according to claim 3 , wherein the adjusting at least one of the speed of the clock or the voltage includes controlling to activate one or more voltage sources of a plurality of voltage sources in order to supply a target voltage corresponding to the determined processing performance to the hardware.
7 . The method according to claim 1 , further comprising,
after the changing the resource:
determining whether or not the changed resource is equal to or greater than a predetermined threshold;
if the changed resource is equal to or greater than the predetermined threshold, measuring a time at which the changed resource is supplied; and
if the measured time reaches a threshold time, controlling to supply resources lower than the threshold to the hardware.