US 12,079,101 B2
System and method for modeling and forecasting input/output (IO) performance using adaptable machine learning
Shaul Dar, Petach Tikva (IL); Avitan Gefen, Tel Aviv (IL); David Sydow, Merrimack, NH (US); and Anil Kumar Koluguri, Durham, NC (US)
Assigned to EMC IP Holding Company, LLC, Hopkinton, MA (US)
Filed by EMC IP Holding Company, LLC, Hopkinton, MA (US)
Filed on Apr. 22, 2022, as Appl. No. 17/727,046.
Prior Publication US 2023/0342280 A1, Oct. 26, 2023
Int. Cl. G06F 11/34 (2006.01); G06F 3/06 (2006.01); G06N 5/02 (2023.01); G06N 20/20 (2019.01)
CPC G06F 11/3485 (2013.01) [G06N 5/02 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method, executed by a processor on a computing device, comprising:
processing historical input/output (IO) performance data associated with one or more storage objects of a storage system;
training a plurality of IO machine learning modeling systems using the historical IO performance data;
determining modeling performance information for the plurality of IO machine learning modeling systems across the historical IO performance data;
determining a forecast score for each IO machine learning modeling system based upon, at least in part, the modeling performance information for the plurality of IO machine learning modeling systems;
selecting a subset of the plurality of IO machine learning modeling systems based upon, at least in part, the forecast score for each IO machine learning modeling system;
training at least one IO machine learning modeling system from the subset of the plurality of IO machine learning modeling systems using the historical IO performance data;
receiving new IO performance data for the one or more storage objects; and
determining whether to retrain the at least one trained IO machine learning modeling system based upon, at least in part, one or more of:
a comparison of the historical IO performance data to the new IO performance data for the one or more storage objects; and
a performance degradation of the at least one trained IO machine learning modeling system with the new IO performance data for the one or more storage objects.