CN-121999852-A - Storage performance self-adaptive evaluation method, equipment and medium for real scene mobile equipment

CN121999852ACN 121999852 ACN121999852 ACN 121999852ACN-121999852-A

Abstract

The invention discloses a self-adaptive evaluation method, equipment and medium for storage performance of a real-scene mobile device, and relates to the technical field of storage performance evaluation, comprising the steps of connecting the mobile device through an ADB and configuring ftrace subsystems to enable block layer I/O event tracking; the method comprises the steps of carrying out multiple I/O event collection when the real user operation of a target application is executed, pulling and analyzing trace files after collection to extract multiple performance indexes, carrying out quality diagnosis on the trace files and outputting grades, obtaining historical baseline data matched with current test scene information from a baseline database when the grades are qualified, carrying out matching based on scene identification and wild card patterns, calculating single index scores by adopting a piecewise linear scoring algorithm based on the extracted indexes and the matched baselines, weighting and fusing to obtain comprehensive health scores, and determining performance grades. The invention provides a mobile device storage performance evaluation scheme which can realize automation and multidimensional and support cross-version performance comparison analysis based on a real application scene.

Inventors

ZHOU JIAYIN
ZHANG SONGYUAN
WEI GUIFANG
CAO TING

Assignees

东莞忆联信息系统有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A method for adaptively evaluating storage performance of a real-scene mobile device, comprising: Connecting mobile equipment through ADB, configuring Linux kernel ftrace subsystem of the mobile equipment to enable block layer I/O event tracking, collecting N times of I/O events when the real user operation of the target application is executed on the mobile equipment, wherein N is more than or equal to 3; performing quality diagnosis on the trace file, detecting a plurality of quality indexes including file size, event number, event pairing rate, abnormal delay and timestamp monotonicity, and outputting a quality rating based on a detection result; If the quality rating meets the preset condition, acquiring historical baseline data matched with the current test scene information from a pre-stored baseline database; Based on the extracted multiple performance indexes and the history baseline data obtained by matching, adopting a preset piecewise linear scoring algorithm to respectively calculate single scores of each performance index, carrying out weighted fusion on the multiple single scores to obtain a comprehensive health degree score, determining a performance grade based on the comprehensive health degree score, and outputting an evaluation result.
2. The method of claim 1, wherein the piecewise linear scoring algorithm comprises a Lower beer policy and Higher Better policy; For the delay indexes, a Lower beer strategy is adopted for grading, namely, the ratio r of the current index value to the historical baseline average value is calculated, when r is less than or equal to 1, the single item grading is full-scale, when r is less than or equal to 1 and r_max, the single item grading is linearly reduced from full-scale to a first Lower limit grading, and when r is more than r_max, the single item grading is the first Lower limit grading, wherein r_max is calculated according to the historical baseline average value and standard deviation; And scoring the throughput index by adopting a Higher Better strategy, namely calculating the ratio r of the current index value to the historical baseline average, wherein when r is more than or equal to 1, the single score is full score, when r_min is less than or equal to r < 1, the single score is linearly increased from a second lower limit to full score, and when r is less than r_min, the single score is the second lower limit score, wherein r_min is calculated according to the historical baseline average and standard deviation.
3. The method according to claim 1, wherein the obtaining historical baseline data matching the current test scenario information specifically comprises: accurately matching the scene identification in the current test scene information with the scene identification in the baseline record; Respectively matching the equipment information, the hardware information, the firmware version information and the application version information in the current test scene information with the corresponding wild card pattern in the baseline record; and when the scene identification is matched accurately and the equipment information, the hardware information, the firmware version information and the application version information are matched successfully, judging that the matching is successful.
4. The method of claim 1, wherein configuring the Linux kernel ftrace subsystem to execute with a three-level rights adaptation policy comprises: firstly, attempting to execute a configuration command under ADB Root authority; If the execution fails, attempting to execute the configuration command under the Shell authority; if the authority fails due to insufficient authority, the configuration command is attempted to be executed by the SU command.
5. The method of claim 1, wherein the plurality of performance metrics include a sequential access ratio, wherein the sequential access ratio is calculated by sequentially processing the parsed I/O requests, calculating an absolute value of a difference between a start logical block address of a current I/O request and an end logical block address of a previous I/O request, and determining that the current I/O request is sequential access if the absolute value is less than or equal to a preset sector threshold.
6. The method of claim 1, wherein the determining a performance level based on the integrated health score comprises: When the comprehensive health degree score is larger than or equal to a preset first threshold value, judging that the performance grade is normal; When the comprehensive health degree score is smaller than a preset first threshold value and larger than or equal to a preset second threshold value, judging that the performance grade is early warning; And when the comprehensive health degree score is smaller than a preset second threshold value, judging that the performance grade is seriously degraded.
7. The method of claim 1, wherein the performance metrics include average latency, P95 latency, P99 latency, average IOPS, sequential access proportion, maximum queue depth, block size distribution feature, and address access distribution feature.
8. The method of claim 1, wherein the quality indicator comprises a trace file size, a block_rq_complete event number, a pairing rate of complete events and complete events, whether there is an abnormal delay exceeding a duration threshold, and whether a timestamp strictly monotonically increases.
9. A computer device, characterized in that it comprises a memory on which a computer program is stored and a processor which, when executing the computer program, implements the method according to any of claims 1-8.
10. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method according to any of claims 1-8.

Description

Storage performance self-adaptive evaluation method, equipment and medium for real scene mobile equipment Technical Field The invention relates to the technical field of storage performance evaluation, in particular to a storage performance self-adaptive evaluation method, device and medium for a real scene mobile device. Background In the field of mobile device storage performance evaluation, the prior art mainly comprises two types of schemes. One type is to use a run-off tool to generate a single indicator of throughput by installing a dedicated application on the device and running a preset standardized load. The method is simple and convenient to operate, but the load model is preset in a laboratory, and has obvious differences with the mixed input and output modes generated in the running process of the real application program, so that the evaluation result cannot accurately reflect the user experience, and the performance degradation problem in cross-application or version iteration is difficult to identify. Another type of scheme relies on manual operations of engineers, captures storage access events under a real application scenario by using a kernel tracking mechanism, and then manually parses event files and counts performance data. Although the method can acquire a real input and output sequence, the whole flow is complex in steps and long in time consumption, the analysis result is seriously dependent on personal experience, a unified index system and a standard data management mechanism are lacked, and the method cannot be suitable for batch automatic regression testing scenes. Therefore, in the prior art, on the premise of ensuring the authenticity of a scene, the storage performance evaluation of the mobile equipment with high efficiency, automation and unified quantization standard is difficult to realize. Disclosure of Invention The embodiment of the invention provides a self-adaptive evaluation method, equipment and medium for storage performance of a real-scene mobile device, and aims to solve the technical problem of providing a mobile device storage performance evaluation scheme which can realize automation and multidimensional and support cross-version performance comparison analysis based on a real application scene. In a first aspect, an embodiment of the present invention provides a method for adaptively evaluating storage performance of a real-scene mobile device, including: Connecting mobile equipment through ADB, configuring Linux kernel ftrace subsystem of the mobile equipment to enable block layer I/O event tracking, collecting N times of I/O events when the real user operation of the target application is executed on the mobile equipment, wherein N is more than or equal to 3; performing quality diagnosis on the trace file, detecting a plurality of quality indexes including file size, event number, event pairing rate, abnormal delay and timestamp monotonicity, and outputting a quality rating based on a detection result; If the quality rating meets the preset condition, acquiring historical baseline data matched with the current test scene information from a pre-stored baseline database; Based on the extracted multiple performance indexes and the history baseline data obtained by matching, adopting a preset piecewise linear scoring algorithm to respectively calculate single scores of each performance index, carrying out weighted fusion on the multiple single scores to obtain a comprehensive health degree score, determining a performance grade based on the comprehensive health degree score, and outputting an evaluation result. Optionally, the piecewise linear scoring algorithm includes a Lower beer policy and Higher Better policy; For the delay indexes, a Lower beer strategy is adopted for grading, namely, the ratio r of the current index value to the historical baseline average value is calculated, when r is less than or equal to 1, the single item grading is full-scale, when r is less than or equal to 1 and r_max, the single item grading is linearly reduced from full-scale to a first Lower limit grading, and when r is more than r_max, the single item grading is the first Lower limit grading, wherein r_max is calculated according to the historical baseline average value and standard deviation; And scoring the throughput index by adopting a Higher Better strategy, namely calculating the ratio r of the current index value to the historical baseline average, wherein when r is more than or equal to 1, the single score is full score, when r_min is less than or equal to r < 1, the single score is linearly increased from a second lower limit to full score, and when r is less than r_min, the single score is the second lower limit score, wherein r_min is calculated according to the historical baseline average and standard deviation. Optionally, the acquiring the historical baseline data matched with the current test scene information specifically includes: accurately matching the scen