Power savings via dynamic page type selection

• US 10,430,347 B2
• Filed: 02/25/2013
• Issued: 10/01/2019
• Est. Priority Date: 09/25/2012
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: King, Justin K.
• Primary Examiner(s): Queler, Adam M
• Assistant Examiner(s): Olson, Alex G
• Application Number:
  US13/775,768
• Publication Number:
  US 20140089631A1
• Time in Patent Office: 2,409 Days
• Field of Search: 711207
• CPC Class Codes: G06F 12/08 in hierarchically structured memory systems, e.g. virtual memory systems
• CPC Class Codes: G06F 12/1009 using page tables, e.g. page table structures
• CPC Class Codes: G06F 12/1027 using associative or pseudo-associative address translation means, e.g. translation look-aside buffer [TLB]
• CPC Class Codes: G06F 12/1081 for peripheral access to main memory, e.g. direct memory access [DMA]
• CPC Class Codes: G06F 2212/205 Hybrid memory, e.g. using both volatile and non-volatile memory
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

An operating system monitors a performance metric of a direct memory access (DMA) engine on an I/O adapter to update a translation table used during DMA operations. The translation table is used during a DMA operation to map a virtual address provided by the I/O adapter to a physical address of a data page in the memory modules. If the DMA engine is being underutilized, the operating system updates the translation table such that a virtual address maps to physical address corresponding to a memory location in a more energy efficient memory module. However, if the DMA engine is over-utilized, the operating system may update the translation table such that the data used in the DMA engine is stored in memory modules that provide quicker access times—e.g., the operating system may map virtual addresses to physical addresses in DRAM rather than phase change memory.

Claims

1. A method of optimizing a computing system, comprising:

receiving a plurality of performance metrics of a data access engine, the data access engine configured to assist in performing at least one memory access operation in one of a first memory module and a second memory module in the computing system, wherein the first memory module is a first type of memory device, the second memory module is a second type of memory device, and the first and second types of memory are different types of memory devices having different performance attributes;

determining, based on a first performance metric of the plurality of performance metrics, that the data access engine is underutilized according to a predefined threshold, and in response increasing the energy efficiency of the data access engine by selecting a first entry from a plurality of entries of the address translation table to reconfigure from association with the second type of memory device to association with the first type of memory device, wherein the first type of memory device is more energy efficient than the second type of memory device and wherein the selection is based on a predicted number of times a data page associated with the first entry is read from or written to, and then reconfiguring the address translation table such that the first entry in the address translation table maps a first virtual address to a first physical address corresponding to the first memory module;

determining based on a second performance metric of the plurality of performance metrics that the data access engine is over-utilized, and in response reconfiguring an address translation table such that the first entry in the address translation table re-maps the first virtual address from the first physical address corresponding to the first memory module to a second physical address corresponding to the second memory module in order to effect a change in utilization of the data access engine; and

determining, based on a third performance metric of the plurality of performance metrics, that the data access engine is over-utilized according to the predefined threshold, and in response reconfiguring the address translation table such that a second entry in the address translation table maps a second virtual address to a physical address corresponding to the second memory module.

2. The method of claim 1, wherein the first memory module consumes less energy to perform the memory access operation than the second memory module and the second memory module requires less time to perform the memory access operation than the first memory module.

3. The method of claim 1, wherein, before reconfiguring the first entry in the address translation table, the first entry mapped the first virtual address to a first previous memory module that consumes more energy when performing the memory access operation than the first memory module.

4. The method of claim 3, wherein the first previous memory module is the second memory module.

5. The method of claim 1, further comprising:

performing a direct memory address operation using the data access engine to transfer a data page to one of the first and second memory modules, wherein one of the first and second entries identifies a memory location in one of the first and second memory modules associated with the data page.

6. The method of claim 5, wherein the direct memory address operation transfers the data page between an I/O adapter and one of the first and second memory modules without the data page being processed by a central processor unit.

7. The method of claim 1, further comprising:

determining that the data access engine is underutilized and in response, before reconfiguring the first entry of the address translation table, predicting whether mapping the first entry to the first memory module results in the data access engine being over-utilized; and

determining that the data access engine is over-utilized and in response, before reconfiguring the second entry of the address translation table, predicting whether mapping the second entry to the second memory module results in the data access engine being underutilized.

8. The method of claim 1, further comprising, determining that the data access engine is underutilized and in response, before reconfiguring the first entry of the address translation table, selecting two or more entries of the address translation table to map to the first memory module, wherein a total number of the plurality of entries varies according to a calculated difference between the first performance metric and a predefined threshold.

9. The method of claim 1, wherein the data access engine is a direct memory access (DMA) engine.

10. The method of claim 1, wherein increasing the energy efficiency of the data access engine further comprises determining that the first type of memory device is more energy efficient than the second type of memory device.

11. The method of claim 1, further comprising identifying a first energy efficiency ranking for the first type of memory device and a second energy efficiency ranking for the second type of memory device.

12. A method of optimizing a computing system, comprising:

receiving a plurality of utilization rates of a data access engine, the data access engine configured to assist in performing at least one memory access operation in one of a first memory module and a second memory module in the computing system, wherein the first memory module is a first type of memory device, the second memory module is a second type of memory device, and the first and second types of memory are different types of memory devices having different performance attributes;

determining, based on a first utilization rate of the plurality of utilization rates, that the data access engine is underutilized according to a predefined threshold, and in response increasing the energy efficiency of the data access engine by selecting a first entry from a plurality of entries of the address translation table to reconfigure from association with the second type of memory device to association with the first type of memory device, wherein the first type of memory device is more energy efficient than the second type of memory device and wherein the selection is based on a predicted number of times a data page associated with the first entry is read from or written to, and then reconfiguring the address translation table such that the first entry in the address translation table maps a first virtual address to a first physical address corresponding to the first memory module;

based on a second utilization rate of the plurality of utilization rates, reconfiguring an address translation table such that a first entry in the table re-maps a first virtual address from a first physical address corresponding to the first memory module to a second physical address corresponding to the second memory module in order to effect a change in utilization of the data access engine; and

determining, based on a third utilization rate of the plurality of utilization rates, that the data access engine is over-utilized according to the predefined threshold, and in response reconfiguring the address translation table such that a second entry in the address translation table maps a second virtual address to a physical address corresponding to the second memory module.

13. The method of claim 12, wherein the first memory module consumes less energy to perform the memory access operation than the second memory module and the second memory module requires less time to perform the memory access operation than the first memory module.

14. The method of claim 12, wherein, before reconfiguring the first entry in the address translation table, the first entry mapped the first virtual address to a first previous memory module that consumes more energy when performing the memory access operation than the first memory module.

15. The method of claim 12, further comprising:

performing a direct memory address operation using the data access engine to transfer a data page to one of the first and second memory modules, wherein one of the first and second entries identifies a memory location in one of the first and second memory modules associated with the data page.

16. The method of claim 15, wherein the direct memory address operation transfers the data page between an I/O adapter and one of the first and second memory modules without the data page being processed by a central processor unit.

17. The method of claim 12, further comprising:

determining that the data access engine is underutilized and in response, before reconfiguring the first entry of the address translation table, predicting whether mapping the first entry to the first memory module results in the data access engine being over-utilized; and

determining that the data access engine is over-utilized and in response, before reconfiguring the second entry of the address translation table, predicting whether mapping the second entry to the second memory module results in the data access engine being underutilized.

18. A method of optimizing a computing system, comprising:

receiving a plurality of performance metrics of a data access engine, the data access engine configured to assist in performing at least one memory access operation in one of a first memory module and a second memory module in the computing system, wherein the first memory module is a first type of memory device, the second memory module is a second type of memory device, and the first and second types of memory are different types of memory devices having different performance attributes;

determining, based on a first performance metric of the plurality of performance metrics, that the data access engine is underutilized according to a predefined threshold, and in response increasing the energy efficiency of the data access engine (i) selecting a first entry from a plurality of entries of an address translation table to reconfigure from association with the second type of memory device to association with the first type of memory device, wherein the first type of memory device is more energy efficient than the second type of memory device and wherein the selection is based on a predicted number of times a data page associated with the first entry is read from or written to and (ii) reconfiguring the address translation table such that the first entry in the address translation table re-maps a first virtual address from a first physical address corresponding to the first memory module to a second physical address corresponding to the second memory module in order to effect a change in utilization of the data access engine; and

determining, based a second performance metric of the plurality of performance metrics, that the data access engine is over-utilized according to the predefined threshold, and in response reconfiguring the address translation table such that a second entry in the address translation table re-maps a second virtual address from a second physical address corresponding to the second memory module to a third physical address corresponding to the first memory module in order to effect a change in utilization of the data access engine.