# US 2012/0254564 A1 · Managing High Speed Memory

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2012/0254564 A1
- Document ID: US20120254564A1
- Status: Published application
- Kind: A1
- Application number: 13/431,479
- Filed: 2012-03-27
- Published: 2012-10-04
- Earliest claimed: 2011-03-31
- Assignees: International Business Machines Corporation
- Inventors: Xue D. Gao; Chao Guang Li; Yang Liu; Yi Yang
- Web page: https://patentgrep.com/patent/US20120254564A1

## CPC

- G06F 12/00: Accessing, addressing or allocating within memory systems or architectures (digital input from, or digital output to record carriers, e.g. to disk storage units, G06F3/06)
- G06F 11/1451: {by selection of backup contents}
- G06F 12/121: using replacement algorithms
- G06F 12/0866: for peripheral storage systems, e.g. disk cache
- G06F 2212/217: Hybrid disk, e.g. using both magnetic and solid state storage devices
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A mechanism is provided for managing a high speed memory. An index entry indicates a storage unit in the high speed memory. A corresponding non-free index is set for a different type of low speed memory. The indicated storage unit in the high speed memory is assigned to a corresponding low speed memory by including the index entry in the non-free index. The storage unit in the high speed memory is recovered by demoting the index entry from the non-free index. The mechanism acquires a margin performance loss corresponding to a respective non-free index in response to receipt of a demotion request. The margin performance loss represents a change in a processor read operation time caused by performing a demotion operation in a corresponding non-free index. The mechanism compares the margin performance losses of the respective non-free indexes and selecting a non-free index whose margin performance loss satisfies a demotion condition as a demotion index. The mechanism selects an index entry from the demotion index to perform the demotion operation.

## Claims

### Claim 1

A method of managing a high speed memory, the method comprising:
determining, in response to receipt of a demotion request, margin performance losses corresponding to respective non-free indexes, wherein corresponding non-free indexes are set for different types of low speed memory, wherein each non-free index comprises index entries indicating a storage unit in a high speed memory assigned to a corresponding low speed memory, wherein a store unit in the high speed memory is recovered by demoting an index entry from a given non-free index, wherein a given margin performance loss represents a change in a processor read operation time caused by performing a demotion operation in a corresponding non-free index;
comparing the margin performance losses of the respective non-free indexes and selecting as a demotion index a non-free index whose margin performance loss satisfies a demotion condition; and
selecting an index entry from the demotion index to perform the demotion operation.

### Claim 2 (depends on 1)

The method according to claim 1, wherein determining a margin performance loss corresponding to a respective non-free index comprises:
determining a margin hit rate loss corresponding to the respective non-free index;
determining a synthesized response time corresponding to the respective non-free index; and
multiplying the margin hit rate loss by the synthesized response time to determine the margin performance loss.

### Claim 3 (depends on 2)

The method according to claim 2, wherein if a storage unit in the high speed memory corresponding to an index entry is accessed, the index entry is moved to the top end of the non-free index; and
wherein determining a margin performance loss corresponding to a respective non-free index comprises:
using the hit rate of the storage unit in the high speed memory corresponding to the index entry at the bottom end of the respective non-free index as the margin hit rate loss.

### Claim 4 (depends on 3)

The method according to claim 3, further comprising:
recording how many times the storage units in the high speed memory corresponding to a plurality of index entries located at the bottom of the respective non-free index are accessed.

### Claim 5 (depends on 2)

The method according to claim 2, wherein the synthesized response time corresponding to the respective non-free index is a weighted average of predicted values of the average response time of the low speed memories corresponding to the respective non-free index according to predicted numbers of accesses.

### Claim 6 (depends on 5)

The method according to claim 5, wherein the predicted value of the average response time of a low speed memory in a next sampling period is predicted by calculating the predicted value and the actual value of the average response time of the low speed memory in a previous sampling period.

### Claim 7 (depends on 6)

The method according to claim 6, wherein the predicted value of the average response time of the low speed memory is calculated according to the following formula:
PredictedRT(n+1)=(1−α)×PredictedRT(n)+α×ActualRT(n)
where PredictedRT(n+1) represents the predicted value of the average response time of the sampling period n+1, and PredictedRT(n) represents the predicted value of the average response time of the sampling period n, ActualRT(n) represents a measured actual value of the average response time in the sampling period n, and α is a number greater than 0 and less than 1.

### Claim 8 (depends on 5)

The method according to claim 5, wherein the predicted number of accesses of a low speed memory in a next sampling period is predicted by calculating the predicted number of accesses and the actual number of accesses of the low speed memory in a previous sampling period.

### Claim 9 (depends on 8)

The method according to claim 8, wherein the predicted number of accesses of the low speed memory is calculated according to the following formula:
PredictedIOPS(n+1)=(1−α)×PredictedIOPS(n)+α×ActualIOPS(n)
where PredictedIOPS(n+I) represents the predicted number of accesses of the sampling period n+1, and PredictedIOPS(n) represents the predicted number of accesses of the sampling period n, ActualIOPS(n) represents a measured actual number of accesses in the sampling period n, and α is a number greater than 0 and less than 1.

### Claim 10 (depends on 9)

The method according to claim 9, wherein α equals a ratio of the sampling period to a workload changing period.

### Claim 11

An apparatus for managing a high speed memory, the apparatus comprising:
a processor; and
a computer readable storage medium coupled to the processor, wherein the computer readable storage medium stores instructions which, when executed by the processor, cause the processor to:
determine, in response to receipt of a demotion request, margin performance losses corresponding to respective non-free indexes, wherein corresponding non-free indexes are set for different types of low speed memory, wherein each non-free index comprises index entries indicating a storage unit in a high speed memory assigned to a corresponding low speed memory, wherein a storage unit in the high speed memory is recovered by demoting an index entry from a given non-free index, wherein a given margin performance loss represents a change in a processor read operation time caused by performing a demotion operation in a corresponding non-free index;
compare the margin performance losses of the respective non-free indexes and select as a demotion index a non-free index whose margin performance loss satisfies a demotion condition; and
select an index entry from the demotion index to perform the demotion operation.

### Claim 12 (depends on 11)

The apparatus according to claim 11, wherein determining margin performance losses comprises:
determining a margin hit rate loss corresponding to the respective non-free index;
determining a synthesized response time corresponding to the respective non-free index; and
multiplying the margin hit rate loss by the synthesized response time to determine the margin performance loss.

### Claim 13 (depends on 12)

The apparatus according to claim 12, wherein if a storage unit in the high speed memory corresponding to an index entry is accessed, the index entry is moved to the top end of the non-free index; and
wherein determining a margin performance loss corresponding to a respective non-free index comprises:
using the hit rate of the storage unit in the high speed memory corresponding to the index entry at the bottom end of the respective non-free index as the margin hit rate loss.

### Claim 14 (depends on 13)

The apparatus according to claim 13, wherein the instructions further cause the processor to:
record how many times the storage units in the high speed memory corresponding to a plurality of index entries located at the bottom of the respective non-free index are accessed.

### Claim 15 (depends on 12)

The apparatus according to claim 12, wherein the synthesized response time corresponding to the respective non-free index is a weighted average of predicted values of the average response time of the low speed memories corresponding to the respective non-free index according to predicted numbers of accesses.

### Claim 16 (depends on 15)

The apparatus according to claim 15, wherein the predicted value of the average response time of a low speed memory in a next sampling period is predicted by calculating the predicted value and the actual value of the average response time of the low speed memory in a previous sampling period.

### Claim 17 (depends on 16)

The apparatus according to claim 16, wherein the predicted value of the average response time of the low speed memory is calculated according to the following formula:
PredictedRT(n+1)=(1−α)×PredictedRT(n)+α×ActualRT(n)
where PredictedRT(n+1) represents the predicted value of the average response time of the sampling period n+1, and PredictedRT(n) represents the predicted value of the average response time of the sampling period n, ActualRT(n) represents a measured actual value of the average response time in the sampling period n, and α is a number greater than 0 and less than 1.

### Claim 18 (depends on 15)

The apparatus according to claim 15, wherein the predicted number of accesses of a low speed memory in a next sampling period is predicted by calculating the predicted number of accesses and the actual number of accesses of the low speed memory in a previous sampling period.

### Claim 19 (depends on 18)

The apparatus according to claim 18, wherein the predicted number of accesses of the low speed memory is calculated according to the following formula:
PredictedIOPS(n+1)=(1−α)×PredictedIOPS(n)+α×ActualIOPS(n)
where PredictedIOPS(n+1) represents the predicted number of accesses of the sampling period n+1, and PredictedIOPS(n) represents the predicted number of accesses of the sampling period n, ActualIOPS(n) represents a measured actual number of accesses in the sampling period n, and α is a number greater than 0 and, less than 1.

### Claim 20 (depends on 19)

The apparatus according to claim 19, wherein α equals a ratio of the sampling period to a workload changing period.

### Claim 21

A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:
determine, in response to receipt of a demotion request, margin performance losses corresponding to respective non-free indexes, wherein corresponding non-free indexes are set for different types of low speed memory, wherein each non-free index comprises index entries indicating a storage unit in a high speed memory assigned to a corresponding low speed memory, wherein a storage unit in the high speed memory is recovered by demoting an index entry from a given non-free index, wherein a given margin performance loss represents a change in a processor read operation time caused by performing a demotion operation in a corresponding non-free index;
compare the margin performance losses of the respective non-free indexes and select as a demotion index a non-free index whose margin performance loss satisfies a demotion condition; and
select an index entry from the demotion index to perform the demotion operation.

### Claim 22 (depends on 21)

The computer program product according to claim 21, wherein determining margin performance losses comprises:
determining a margin hit rate loss corresponding to the respective non-free index;
determining a synthesized response time corresponding to the respective non-free index; and
multiplying the margin hit rate loss by the synthesized response time to determine the margin performance loss.

### Claim 23 (depends on 22)

The computer program product according to claim 22, wherein the synthesized response time corresponding to the respective non-free index is a weighted average of predicted values of the average response time of the low speed memories corresponding to the respective non-free index according to predicted numbers of accesses.

### Claim 24 (depends on 23)

The computer program product according to claim 23, wherein the predicted value of the average response time of a low speed memory in a next sampling period is predicted by calculating the predicted value and the actual value of the average response time of the low speed memory in a previous sampling period, wherein the predicted value of the average response time of the low speed memory is calculated according to the following formula:
PredictedRT(n+1)=(1−α)×PredictedRT(n)+α×ActualRT(n)
where PredictedRT(n+1) represents the predicted value of the average response time of the sampling period n+1, and PredictedRT(n) represents the predicted value of the average response time of the sampling period n, ActualRT(n) represents a measured actual value of the average response time in the sampling period n, and α is a number greater than 0 and less than 1.

### Claim 25 (depends on 23)

The computer program product according to claim 23, wherein the predicted number of accesses of a low speed memory in a next sampling period is predicted by calculating the predicted number of accesses and the actual number of accesses of the low speed memory in a previous sampling period, wherein the predicted number of accesses of the low speed memory is calculated according to the following formula:
PredictedIOPS(n+1)=(1−α)×PredictedIOPS(n)+α×ActualIOPS(n)
where PredictedIOPS(n+1) represents the predicted number of accesses of the sampling period n+1, and PredictedIOPS(n) represents the predicted number of accesses of the sampling period n, ActualIOPS(n) represents a measured actual number of accesses in the sampling period n, and α is a number greater than 0 and less than 1.
