A cache management method, apparatus and system

Abstract

The present invention relates to the technical field of cache management, and in particular to a cache management method, device, and system. The present invention screens cache data by introducing pheromone concentration, combines the pheromone threshold T to distinguish hot data (i.e., hotspot data) and cold data (i.e., non-hotspot data), updates pheromones according to data access frequencies, increases the pheromone concentration difference between hot data and cold data, thereby preferentially caching frequently accessed data items, improving the hit rate and optimizing the use of limited cache space, and enhancing cache performance and resource utilization. The present invention introduces the concept of pheromones to dynamically evaluate and manage cache data, realizes a more efficient and flexible caching strategy, and overcomes the defects such as poor dynamic adaptation ability of the existing cache system.

Description

Technical Field

[0001] The present invention relates to the technical field of cache management, and in particular to a cache management method, device and system. Background Art

[0002] With the development of information technology and the popularization of Internet applications, cache technology has played an important role in improving system performance and reducing the pressure on the backend database. The cache system stores frequently used or hot data to reduce the access frequency to the backend data source, thereby improving the response speed and user experience. However, with the continuous change of data volume and access patterns, how to effectively manage cache resources, especially to maximize the cache hit rate within limited cache space, has become a major challenge.

[0003] Traditional cache management strategies, such as algorithms like LRU (Least Recently Used) and LFU (Least Frequently Used), although solve the cache eviction problem to a certain extent, these strategies seem too static in the face of dynamically changing access patterns. For example, in scenarios where hot data changes frequently, traditional strategies may not be able to adjust cache entries in time, resulting in a decrease in the cache hit rate.

[0004] Specific cases show that during promotional activities on large e-commerce platforms, the cache hit rate of the traditional LRU strategy may drop from 85% during normal periods to below 60%, seriously affecting user experience and system performance. Another study shows that on social media platforms, the rapid change of hot topics reduces the effectiveness of the LFU-based cache strategy by 30% within 2 hours.

[0005] After analysis, the existing cache systems have the following problems:

[0006] 1. Performance bottleneck: With the growth of data volume, the number of requests that the cache system needs to process increases sharply, resulting in storage and retrieval efficiency becoming a key issue;

[0007] 2. Lack of intelligent data access pattern recognition: The existing systems fail to fully analyze temporal and spatial locality, and it is difficult to distinguish between hot data and cold data, affecting the accuracy of cache decisions;

[0008] 3. Resource utilization: It is currently impossible to accurately predict hot data, resulting in unnecessary cache occupancy and reducing the overall resource utilization efficiency;

[0009] 4. Dynamic adaptation ability: Facing continuously changing access patterns and user behaviors, existing cache strategies are difficult to respond quickly, and their long-term efficiency is affected. Summary of the Invention

[0010] To overcome the deficiencies such as poor dynamic adaptability of the cache system in the above-mentioned prior art, the present invention proposes a cache management method that can guide subsequent searches through pheromone concentration, find better solutions in an uncertain and dynamically changing environment, improve the dynamic adaptability of the cache system, and optimize resource occupancy.

[0011] A cache management method proposed by the present invention assigns an initial value of pheromone concentration to newly accessed data in the cache area. When the data in the cache area is accessed again, the pheromone concentration of the accessed data is updated; the cache area is used to store data with a larger pheromone concentration.

[0012] When the data is accessed again, its pheromone concentration update formula is as follows:

[0013]

[0014] Among them, P new (a k ) represents the updated pheromone concentration of data a k , P(a k ) represents the pheromone concentration of data a k before update, α is the attenuation factor, Δt k is the time elapsed since data a k was last accessed; Q is a set constant; T is a set pheromone threshold, and β is a set weight coefficient and β>1.

[0015] Preferably, the pheromone of the cache area data is updated every interval of the time length window W, or the pheromone of the cache area data is updated every time N access requests are processed. The update rule is: if the data is accessed in the most recent time length window W, the pheromone is updated according to the pheromone concentration update formula; if the data is not accessed in the most recent time length window W, the pheromone concentration of the data is set to zero.

[0016] Preferably, every interval of the set time t R , the pheromone threshold T, the time window length W, the attenuation factor α, and the weight coefficient β are updated. The parameter update formula is expressed as follows:

[0017] R new =R old ×(1 + γ H ·(H - H target ) + γ A ·(A - A target ))

[0018] Among them, R new represents the updated parameter R, R old represents the parameter R before update, R∈{W,α,β,T};

[0019] Let \(H\) denote the current hit rate, \(H\) target be the set target hit rate; Let \(A\) denote the current average access time of the data stored in the buffer; \(A\) target be the target average access time of the buffer; \(\gamma\) H be the adjustment coefficient that controls the influence degree of the hit rate on \(R\), \(\gamma\) A the adjustment coefficient that controls the influence degree of the average access time on \(R\).

[0020] Preferably, the initial value of the pheromone concentration is inversely proportional to the data change frequency of the system, and its value range is \((0, 0.5]\); \(\gamma\) H and \(\gamma\) A have a value range of \([0.01, 0.1]\).

[0021] Preferably, when an access request occurs and the access request is not stored in the buffer, and there is no empty space in the buffer currently; then search for the dirty data in the buffer, replace the dirty data with the accessed data, and assign the initial value of the pheromone concentration to the accessed data; if there is no dirty data in the buffer, keep the current state of the buffer; dirty data refers to the data that needs to be eliminated first in the buffer;

[0022] When an access request occurs and the access request is not stored in the buffer, and there is an empty space in the buffer, then add the accessed data to the buffer, and assign the initial value of the pheromone concentration.

[0023] Preferably, dirty data refers to the data in the buffer with the minimum pheromone concentration and less than the initial value; if there are multiple data in the buffer with the same pheromone concentration and less than the minimum value of the initial value, then select the data with the longest waiting access time recently and the minimum pheromone concentration as the dirty data. If there are multiple such data, randomly select one as the dirty data; the recent waiting access time is the time from the most recent access to the current time.

[0024] Preferably, the method for screening dirty data is as follows: First, find the minimum value of the pheromone concentration in the buffer. If the minimum value of the pheromone concentration is greater than or equal to the initial value, it means there is no dirty data in the buffer; if the minimum value of the pheromone concentration is less than the initial value and the minimum value of the pheromone concentration is the only item, then take the data corresponding to the minimum value of the pheromone concentration as the dirty data; if the minimum value of the pheromone concentration is less than the initial value and the pheromone concentrations of multiple data in the buffer are all the minimum value of the pheromone concentration, then let the data corresponding to the minimum value of the pheromone concentration be the candidate dirty data, and then use the elimination algorithm to select one item from the candidate dirty data as the dirty data.

[0025] Preferably, the LRU algorithm is used as the elimination algorithm.

[0026] A cache management device proposed by the present invention includes a data access pattern analyzer, a dynamic parameter adjustment module, and a cache manager;

[0027] The data access pattern analyzer is used to collect and analyze the timestamps and address information of read and write operations, and calculate the cache hit rate; the dynamic parameter adjustment module is connected to the data access pattern analyzer and updates the key parameters according to the cache hit rate. The key parameters include the pheromone threshold T, the time window length W, the decay factor α, and the weight coefficient β; the cache manager is respectively connected to the data access pattern analyzer and the dynamic parameter adjustment module.

[0028] A cache management system proposed by the present invention includes a main storage area, a cache area, and a processor. The processor is used to execute a computer program to implement the cache management method described above.

[0029] The advantages of the present invention are as follows:

[0030] (1) By introducing the pheromone concentration to screen cache data and combining the pheromone threshold T to distinguish hot data (i.e., hotspot data) and cold data (i.e., non-hotspot data), the present invention updates the pheromone according to the data access frequency, increases the pheromone concentration difference between hot data and cold data, thereby preferentially caching the data items with high access frequency, improving the hit rate and optimizing the use of limited cache space, and enhancing the cache performance and resource utilization rate. The present invention introduces the pheromone concept to dynamically evaluate and manage cache data, realizing a more efficient and flexible cache strategy.

[0031] (2) In the present invention, only the pheromone of the cache area data is updated, and the pheromone concentration of the data is updated when it is accessed, ensuring the flexibility and real-time nature of the pheromone concentration update of the data, enhancing the pheromone advantage of the high-frequency access data, and guaranteeing the hit rate of the cache area. The cache area periodically updates the pheromone concentration of the stored data to ensure that the cache area is always in the best configuration, avoiding the problem of long-term performance degradation caused by static characteristics in traditional strategies; it is beneficial to timely discover low-frequency data and reduce the pheromone concentration of low-frequency data, so as to be eliminated in a timely manner, avoid unnecessary cache occupation, improve the utilization efficiency of the cache space, reduce the storage cost, and enhance the resource utilization rate.

[0032] (3) In the present invention, the key parameters such as the pheromone threshold T, the time window length W, the decay factor α, and the weight coefficient β are updated in real time in combination with the hit rate, realizing the interaction between the key parameters and the feedback, being able to better adapt to the changing workload, and enhancing the performance of the method in a dynamic environment.

[0033] (4) The present invention flexibly adjusts the cache content according to the changes in the access frequency and pheromone concentration value, ensuring the freshness of data and the effective allocation of resources; regularly recalculates the pheromone concentration and the hot data set to ensure that the system is always in the best operating state. The combination of the two enhances the dynamic adaptability in cache management.

[0034] (5) The present invention combines the pheromone concentration and the elimination algorithm to determine the dirty data, realizes the intelligent elimination of low-frequency data, and reduces the waste of cache resources by invalid data. Through the replacement of dirty data with newly accessed data, it provides intelligent decision support, makes the cache management more scientific and reasonable, reduces the need for manual intervention, and lowers the operation and maintenance complexity. Brief Description of the Drawings

[0035] Figure 1 It is a flowchart of a cache management method;

[0036] Figure 2 It is a comparison of the hit rates of multiple cache strategies in the embodiment;

[0037] Figure 3 It is a schematic diagram of a cache management device;

[0038] Figure 4 It is a timing diagram of cache management actions. Detailed Embodiment

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] A cache management method proposed by the present invention first assigns a pheromone concentration to the data and updates the pheromone concentration according to the data access frequency. The cache area is used to store the data with a larger pheromone concentration; specifically, since the data can be modified, the data access frequency is statistically calculated based on the address access record.

[0041] The pheromone concentration of the newly accessed data is set to the initial value. When the data is accessed again, the pheromone concentration of the accessed data is updated; the pheromone concentration update formula is as follows:

[0042]

[0043] Among them, P new (a k ) represents the updated pheromone concentration of data a k , and P(a k ) represents data ak The pheromone concentration before update, α is the attenuation factor, and Δt k is the time elapsed since data a k was last accessed; Q is a set constant representing the amount of pheromone increased each time of access; T is the set pheromone threshold, and the access data with pheromone concentration greater than or equal to the threshold T is classified as hot data with high access frequency, while the access data with pheromone concentration less than the threshold T is classified as cold data with low access frequency; β is a weight coefficient greater than 1.

[0044] The data in the buffer updates the pheromone concentration periodically. The time window length for the periodic update in the buffer is W, that is, every interval of time W, the data in the buffer updates the pheromone concentration. When performing the periodic update, for the data stored in the buffer, first determine whether the data has been accessed in the most recent time window length W; if the data has been accessed, update the pheromone concentration P according to the above pheromone concentration update formula (1) nwe (a k ); if the data has not been accessed in the most recent time window length W, then set P new (a k ) = 0.

[0045] It should be noted that when the buffer is updated periodically, Δt k is the time from the most recent access of data a k to the current time; when the data is accessed and updated, Δt k is the interval time from the last access of data a k to the current access.

[0046] The pheromone threshold T, the time window length W, the attenuation factor α, and the weight coefficient β are all updated in real time, and the update time interval is t R ; and the initial values of T, W, α, and β are all set values, and the update formulas for the four are as follows:

[0047] R new = R old ×(1 + γ H ·(H - H target ) + γ A ·(A - A target ))

[0048] Among them, R new represents the updated parameter R, and R old represents the parameter R before update, and R ∈ {W, α, β, T};

[0049] H represents the current hit rate, that is, the probability that the accessed data exists in the buffer; H targetis the set target hit rate; A represents the current average access time of the data stored in the buffer, i.e., the average value of Δt of all data in the buffer k ; A target is the target average access time of the buffer; γ H and γ A are adjustment coefficients used to control the influence degree of the hit rate and the average access time on R.

[0050] The adjustment coefficients {γ H , γ A} corresponding to T, W, α, and β are not necessarily the same. γ H is positively correlated with the target hit rate, and γ A is negatively correlated with the access time; that is, in order to increase the hit rate, γ H can be appropriately increased;

[0051] if it is for reducing the access time, then γ A can be appropriately increased.

[0052] When in use, γ A and γ H can be set to relatively small positive values, for example, specifically in the range [0.01, 0.1], and then fine-tuned according to the actual operation situation. This allows the system to gradually adapt to performance changes while maintaining a certain stability.

[0053] Referring to Figure 1 , in the initial state of this cache management method, the buffer is empty; when an access request occurs, the accessed data is processed according to the following steps.

[0054] S1. Determine whether the accessed data is stored in the buffer; if yes, directly return the result and update the pheromone concentration of the accessed data according to the above formula (1); if no, execute step S2;

[0055] S2. Determine whether there is a vacancy in the buffer;

[0056] If yes, store the accessed data in the buffer and set the pheromone concentration of the accessed data to the initialization value;

[0057] If no, execute step S3;

[0058] S3. Determine whether there is data in the buffer whose pheromone concentration is less than the initialization value;

[0059] If no, directly return the result;

[0060] If yes, execute step S4;

[0061] S4. Replace the dirty data in the buffer with the accessed data, and set the pheromone concentration of the accessed data to the initial value; that is, remove the dirty data stored in the buffer, and then store the accessed data in the buffer.

[0062] The method for screening the dirty data in the buffer is as follows: First, determine whether there is data in the buffer with a pheromone concentration less than the initial value; if not, it is determined that there is no dirty data in the buffer; if so, the data with the minimum pheromone concentration is preferentially used as the dirty data; if there are multiple minimum values with the same pheromone concentration, the data with the longest waiting access time and the minimum pheromone concentration is selected as the dirty data. The recent waiting access time is the time from the most recent access to the current time. If there are multiple such data, one will be randomly selected as the dirty data.

[0063] Specifically, when there are multiple minimum values with the same pheromone concentration, an elimination algorithm can be used to determine the dirty data among the data corresponding to the minimum value. The elimination algorithm can adopt LRU (Least Recently Used, the data that has been least recently used is preferentially eliminated), FIFO (First In First Out, the data that entered the cache earliest is preferentially eliminated), LFU (Least Frequently Used, the data with the lowest usage frequency is preferentially eliminated), and RR (Random Replacement, randomly select one for elimination), etc.

[0064] The initial value of the pheromone concentration of the data is a set value. Obviously, the higher the initial value, the slower the update speed of the buffer, and the lower the initial value, the faster the update speed of the buffer.

[0065] In this way, in the case of limited resources, a slightly higher initial value can be set to ensure that enough data items have the opportunity to be cached in the initial stage, and to avoid too low a hit rate caused by frequent cache replacement; when the cache capacity is large, the initial value can be appropriately reduced to allow more data items to enter the cache for competition, and gradually screen out the real hot data.

[0066] In addition, the initial value can also be set according to the environmental characteristics; for a high-traffic and rapidly changing environment, a lower initial value can be tried, such as 100 or lower, to accelerate the system's response to new hot data; for a low-traffic and relatively stable environment, a higher initial value can be considered, such as 200 or 300, to maintain a relatively high cache hit rate in the initial stage.

[0067] The following verifies the above cache management method in combination with specific embodiments.

[0068] In this embodiment, the total number of addresses participating in the calculation is 60000, among which there are 500 hot data, and the cache size is 1024, that is, there are 1024 data when the buffer is full.

[0069] In this embodiment, the initial value of the time window length W is set to 500, the initial value of the pheromone threshold T is set to 70, the initial value of the attenuation factor α is set to 0.02, the initial value of the weight coefficient β is set to 1.08, and the initial value of the initial concentration of pheromone is set to 100; the adjustment coefficient γ H and γ A are both set to 0.5.

[0070] In this embodiment, the number of operations is defined to include reads and writes; as the number of operations increases, the cache has more opportunities to learn the access pattern and optimize its content, and generally has a higher hit rate.

[0071] In this embodiment, the cache management method provided by the present invention is compared with traditional LRU and LFU, and the comparison results are as Figure 2 shown.

[0072] It can be seen that as the number of operations increases, the hit rates of the three cache policies increase with the increase of the number of operations until a stable value is reached. This is because more operations provide more opportunities for the cache to learn the data access pattern, so as to store frequently accessed data more effectively.

[0073] The cache management method proposed by the present invention shows a higher hit rate than LRU and LFU, especially in the case of large-scale operations. This indicates that the present invention has better adaptability in dealing with complex access patterns.

[0074] For convenience, the method of the present invention is abbreviated as ACO; this embodiment further analyzes the three cache policies.

[0075] 1. Small-scale operations (between 500 and 5000 times):

[0076] The hit rates of all three cache policies are very low, about 9% or so, because the number of operations is not enough to form an obvious access pattern, making it impossible for the cache to effectively predict which data will be accessed again.

[0077] 2. Medium-scale operations (between 5,000 and 50,000 times):

[0078] The hit rates of LRU and ACO are significantly higher than that of LFU, and ACO performs slightly better than LRU. This shows that in this type of scenario, ACO can better capture the locality of data access and the characteristics of hot data.

[0079] 3. Large-scale operations (between 500,000 and 5,000,000 times):

[0080] ACO continues to maintain its advantages, especially in the case of very large operation sets (more than 5,000,000 times). The hit rate of ACO is 40.2%, significantly higher than 35.85% of LRU and 3.72% of LFU.

[0081] It is worth noting that the performance of LFU drops sharply in large-scale operations and almost loses its effectiveness. This may be because LFU relies too much on frequency counting. Under a huge number of operations, the frequency counting may become less accurate or lose timeliness, resulting in its inability to well reflect the recent access patterns.

[0082] It can be seen from this implementation that under all test conditions, the ACO cache shows the highest hit rate, especially in medium to large-scale operations. This proves that the present invention can better adapt to complex access patterns and can more effectively identify and retain hot data in the case of long-term operation and a large number of operations.

[0083] The present invention also proposes a cache management device, including a data access pattern analyzer, a dynamic parameter adjustment module, and a cache manager;

[0084] The data access pattern analyzer is used to collect and analyze the timestamps and address information of read and write operations, and calculate the cache hit rate;

[0085] The dynamic parameter adjustment module is connected to the data access pattern analyzer and updates the key parameters according to the cache hit rate. The key parameters include the pheromone threshold T, the time window length W, the attenuation factor α, and the weight coefficient β;

[0086] The cache manager is respectively connected to the data access pattern analyzer and the dynamic parameter adjustment module; the cache manager is used to update the pheromone of the cache data and optimize the cache data, that is, to implement steps S1-S4 in the above cache management method.

[0087] When the cache management device works, its timing actions are as Figure 4 shown.

[0088] Of course, for those skilled in the art, the present invention is not limited to the details of the above exemplary embodiments, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0089] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0090] The technologies, shapes, and structures not detailedly described in the present invention are all well-known technologies.

Claims

1. A cache management method, characterized in that, Initialize the pheromone concentration value for the newly accessed data in the buffer. When the data in the buffer is accessed again, update the pheromone concentration of the accessed data; the buffer is used to store the data with a larger pheromone concentration. When the data is accessed again, its pheromone concentration update formula is as follows: Among them, P new (a k ) represents the data a k The updated pheromone concentration, P(a k ) represents the data a k The pheromone concentration before update, α is the attenuation factor, Δt k is the time elapsed since the data a k was last accessed; Q is a set constant; T is a set pheromone threshold, and β is a set weight coefficient and β is greater than 1.

2. The cache management method according to claim 1, wherein Update the pheromone of the buffer data every time interval length window W, or update the pheromone of the buffer data every time N access requests are processed. The update rule is: if the data is accessed in the most recent time interval length window W, update the pheromone according to the pheromone concentration update formula; if the data is not accessed in the most recent time interval length window W, set the pheromone concentration of the data to zero.

3. The cache management method according to claim 2, wherein Every set time interval t R , the pheromone threshold T, time window length W, attenuation factor α, and weight coefficient β are updated; the parameter update formula is as follows: R new = R old ×(1 + γ H ·(H - H target ) + γ A ·(A - A target )) Among them, R new represents the updated parameter R, and R old represents the parameter R before update, where R ∈ {W, α, β, T}; H represents the current hit rate, H target is the set target hit rate; A represents the current average access time of the data stored in the buffer; A target is the target average access time of the buffer; γ H is the adjustment coefficient that controls the influence degree of the hit rate on R, γ A The adjustment coefficient that controls the influence degree of the average access time on R.

4. The cache management method according to claim 3, wherein The initial value of the pheromone concentration is inversely proportional to the data change frequency of the system, and its value range is (0, 0.5]; γ H and γ A are in the value range of [0.01, 0.1].

5. The cache management method according to any one of claims 1 to 4, characterized in that, When an access request occurs and the access request is not stored in the buffer, and there is no empty space in the buffer currently; then search for the dirty data in the buffer, replace the dirty data with the accessed data, and assign the initialization value of the pheromone concentration to the accessed data. If there is no dirty data in the buffer, keep the current state of the buffer; dirty data refers to the data that needs to be evicted from the buffer first. When an access request occurs and the access request is not stored in the buffer, and there is an empty space in the buffer, add the accessed data to the buffer and assign the initialization value of the pheromone concentration.

6. The cache management method according to claim 5, characterized in that, Dirty data refers to the data in the buffer with the smallest pheromone concentration and less than the initialization value; if there are multiple data in the buffer with the same pheromone concentration and less than the minimum value of the initialization value, select the data with the longest waiting access time recently and the smallest pheromone concentration as the dirty data. If there are multiple such data, randomly select one as the dirty data; the recent waiting access time is the time from the most recent access to the current time.

7. The cache management method according to claim 6, wherein The screening method for dirty data is: first, search for the minimum value of the pheromone concentration in the buffer. If the minimum value of the pheromone concentration is greater than or equal to the initialization value, it means that there is no dirty data in the buffer. If the minimum value of the pheromone concentration is less than the initialization value and the minimum value of the pheromone concentration is the only item, then use the data corresponding to the minimum value of the pheromone concentration as the dirty data. If the minimum value of the pheromone concentration is less than the initialization value and the pheromone concentrations of multiple data in the buffer are all the minimum value of the pheromone concentration, then use the data corresponding to the minimum value of the pheromone concentration as the candidate dirty data, and then use the elimination algorithm to select one as the dirty data from the candidate dirty data.

8. The cache management method according to claim 7, wherein The elimination algorithm uses the LRU algorithm.

9. A cache management device adopting the cache management method according to any one of claims 1-8, characterized in that, It includes a data access pattern analyzer, a dynamic parameter adjustment module, and a cache manager. The data access pattern analyzer is used to collect and analyze the timestamps and address information of read and write operations, and calculate the cache hit rate. The dynamic parameter adjustment module is connected to the data access pattern analyzer and updates the key parameters according to the cache hit rate. The key parameters include the pheromone threshold T, the time window length W, the decay factor α, and the weight coefficient β; the cache manager is connected to the data access pattern analyzer and the dynamic parameter adjustment module respectively.

10. A cache management system, characterized in that, It includes a main storage area, a buffer, and a processor. The processor is used to execute a computer program to implement the cache management method described in any one of claims 1-8.

Images