Apparatus and method for detecting poor component life zones in a datacenter

A method for determining zones within a datacenter that exhibit poor component life, the zones each including a plurality of components, each component configured to provide error indications associated with aging of the component, and each component having a predefined error threshold and a predefined warranty duration, the method comprising:

• receiving, by a component life manager, the error indications for each component;
• fitting the received error indications to a curve for each component;
• determining an end of life estimate for each component based upon the respective curve and the error threshold;
• determining an end of life date for each component based upon the respective end of life estimate and a date that the first error indication is received for each component;
• determining a loss of life for each component based on the respective end of life estimate and warranty duration;
• determining a component life score for each component based upon the respective loss of life and warranty duration;
• determining, for each zone, a zone life score based upon an average of the component life scores of the components of the zone;
• determining a datacenter life score based upon an average of the component life scores;
• determining that the datacenter life score for a particular zone is correlated to an operating parameter of the particular zone;
• providing an indication for the particular zone when the associated zone life score is below the datacenter life score by more than a threshold;
• and
• changing the operating parameter of the particular zone when the associated zone life score is below the datacenter life score by more than the threshold.

A method for determining zones within a datacenter that exhibit poor graphics processing unit (GPU) life, the zones each including a plurality of GPUs, each GPU configured to provide a retired page count, and each GPU having a predefined error threshold and a predefined warranty duration, the method comprising:

• receiving, by a GPU life manager, the retired page count for each GPU;
• fitting the received retired page counts to a curve for each GPU, wherein the curve for each GPU is of the form:
• # CE=AeBt−C
• where # CE represents a number of retired page counts received at a given time t, and where A, B, and C are terms that are derived via regression analysis;
• determining an end of life estimate for each GPU based upon the respective curve and the error threshold;
• determining an end of life date for each GPU based upon the respective end of life estimate and a date that the first retired page count is received for each GPU;
• determining a loss of life for each GPU based on the respective end of life estimate and warranty duration;
• determining a GPU life score for each GPU based upon the respective loss of life and warranty duration;
• determining, for each zone, a zone life score based upon an average of the GPU life scores of the GPUs of the zone;
• determining a datacenter life score based upon an average of the GPU life scores;
• determining that the GPU life score for a particular zone is correlated to an operating parameter of the particular zone;
• providing an indication for a particular zone when the associated zone life score is below the datacenter life score by more than a threshold;
• and
• changing the operating parameter of the particular zone when the associated zone life score is below the datacenter life score by more than the threshold.