CN103414747A - Adaptive streaming media caching method - Google Patents

Abstract

The invention discloses an adaptive streaming media caching method. Combined with the behavior analysis of user access, different caching methods are employed for streaming media of different heat degrees. For 20% high heat degree streaming media, the caching method is that the streaming media is all cached when a caching space is sufficient, and an effect of fully using the space is achieved. When the caching space is insufficient, a strategy of getting 50% of the caching space with low hit rate can realize the purpose of greatly saving space and reducing cost in the condition that a user experience degree is basically satisfied. For 80% low heat degree streaming media, a model is fit according to the play on demand behavior of a user, as an adaptive template, the caching space is initialized, then according to the concrete play on demand behavior of the user, a caching segment is dynamically added and deleted, thus in one hand, the caching space is greatly saved in the initialization, the storage cost is saved, on the other hand, the number of catch replacement is reduced as much as possible, and the effect of reducing transmission cost is achieved.

Description

A kind of adaptive stream media caching method

Technical field

The present invention relates to communication network value-added service technical field, more specifically, relate to a kind of adaptive stream media caching method.

Background technology

Along with the maturation of the universal and digital video technology of network, the stream media systems such as video on-demand system (VOD), long-distance education and interactive game also develop rapidly.Streaming Media is compared with the traditional static media, has the advantages that amount of content data is large, the time delay susceptibility is high and interactivity is strong.In the stream media system of existing content-based distributing network CDN, Streaming Media source server and streaming cache proxy server are to provide Key Platform and the nucleus equipment of stream service.The Streaming Media source server generally is in IP kernel heart net, and be used to depositing files in stream media, response user request also sends stream medium data to terminal.The streaming cache proxy server, be positioned at the edge of network, and near the user, so the effect of streaming cache proxy server just seems very important.

Because the propagation of files in stream media need to take a large amount of network bandwidths, in this process as how lower cost is pushed content to Edge Server from central server, and how different files in stream media is carried out to buffer memory and guarantee that lower user's broadcast time-delay and higher Buffer Utilization have received the concern of industry day by day.

For addressing the above problem, CDN provider generally can by the user often the content caching of access on the Edge Server close to from the user, by Edge Server, processes local user's request, buffer memory forwarding are from the data of source server.Yet the whole files in stream media of buffer memory needs huge disk space and network overhead.According to existing cache policy, even only the sub-fraction of each Streaming Media is existed on Edge Server, also can reach very high clicking rate.Cache policy commonly used has the prefix buffer memory at present, index buffer memory, segmentation buffer memory, interval buffer memory, minute segment prefix buffer memory etc.Wherein, prefix buffer memory and segmentation buffer memory can reduce bandwidth pressure, broadcast time-delay and the shake of backbone network, and the interval buffer memory can ask to improve hit rate for user subsequently.

Yet existing strategy has been ignored the characteristic of user's behavioural habits and Streaming Media self:

1) ignored user's click and watched this important information of behavior;

2) do not consider the otherness of different film sources, just prefix or certain part of cache flow media regularly, and actual last Streaming Media often has several climax fragments, these climax fragments but are not buffered on Edge Server.

Existing caching method comes front 20% Streaming Media by temperature and all is buffered on Edge Server, and the Streaming Media that in fact temperature is high is because most of spectators have seen, the user watches Shi Buhui to watch through and through again, tend to jump and broadcast and playback, therefore the clicking rate of whole Streaming Media might not be very high, and the distribution of clicking rate is also inhomogeneous.So adopt this cache policy, although hit rate can be very large, wasted too many spatial cache, Buffer Utilization is not high.

Summary of the invention

Main purpose of the present invention is to overcome the not high problem of flow medium buffer utilance in prior art, proposes the adaptive stream media caching method that a kind of Buffer Utilization is high.

To achieve these goals, technical scheme of the present invention is:

A kind of adaptive stream media caching method, carry out temperature and estimate being stored in Streaming Media in the Streaming Media source server, and according to estimation results by flow medium buffer to Edge Server, be specially:

The Streaming Media of inputting is carried out to temperature and estimate, Streaming Media is divided into to high temperature Streaming Media and low temperature Streaming Media;

The high and low temperature Streaming Media of difference buffer memory;

High temperature flow medium buffer is to Edge Server in the following ways: high temperature Streaming Media is arranged by the height of temperature, high temperature Streaming Media successively is cached to Edge Server from height to hanging down by temperature, until the spatial cache utilance reaches M, wherein M is 90%-96%;

When inadequate buffer space 100%-M, the total length of storing each Streaming Media is 40-60%;

Hang down the temperature flow medium buffer to Edge Server in the following ways:

Utilize the Lorentzian model, according to the behavior curve of user's program request Streaming Media;

Program request position while determining that by fitting function peak value appears in curve, obtain the threshold value of video-on-demand times according to the program request position;

Streaming Media interval when occurring that video-on-demand times is peak value is cached to Edge Server.

At first the present invention is divided into high temperature Streaming Media and low temperature Streaming Media according to the temperature difference of Streaming Media by it, and for the flow medium buffer of different temperatures.When high temperature Streaming Media is carried out to buffer memory, after high temperature Streaming Media is just sorted by temperature, at first by high temperature Streaming Media by temperature buffer memory successively from high to low, until the spatial cache utilance is while reaching certain value, again by the part of cache of remaining high temperature Streaming Media total length, by sacrificing the lower hit rate of the higher Streaming Media of temperature, exchange more memory spaces for, the hit rate of loss has very large probability to make up in remaining inferior backing.Wherein time backing refers to the backing of all preserving with respect to original whole media, because present backing all be kept at Edge Server, now by high temperature Streaming Media by temperature buffer memory successively from high to low, until the spatial cache utilance is while reaching certain value, again by the part of cache of remaining high temperature Streaming Media total length, and these not all sheets of storage just comprised inferior backing.

This high temperature flow medium buffer mode, in conjunction with high temperature Streaming Media hit rate and buffer memory space empty spare time degree; Both take full advantage of spatial cache, be both the playing request that has met most of user, greatly improved user experience.

Cache policy in the past is for the Streaming Media of the low temperature previous section at Edge Server on stored stream media, although reduced broadcasting start-up time, according to the hit rate computing formula, this strategy can not obviously improve hit rate.

The buffer memory of high temperature Streaming Media tended to 20% high temperature Streaming Media all is stored on caching server, Buffer Utilization is low in the past.

When low temperature Streaming Media is carried out to buffer memory, be based on the caching method of user's access habits.

Further, described Streaming Media to input carries out the mode that temperature estimates: the rule that adopts Zipf to distribute draws Pareto Law, namely 80% user only accesses 20% streaming medium content, 20% Streaming Media is defined as to high temperature Streaming Media, and 80% Streaming Media is defined as low temperature Streaming Media.

The user has very strong tendentiousness to the access of files in stream media, and this tendentiousness can be divided into video popularity between file and the video popularity of file inside.Under the former impact, the frequency difference that files in stream media is accessed, the focus files in stream media is accessed frequently; Under the latter's impact,, also there is the focus part be accessed frequently in the frequency difference that the different piece of files in stream media is accessed by the user.Have now and mainly the former was carried out to correlative study, Zipf distributes, and is widely used for describing this specific character:

$\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{P}_i=1,P_k=\frac{\mathrm{\λ}}{K^{1-\mathrm{\α}}},\mathrm{\λ}=\frac{1}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{1}{i^{1-\mathrm{\α}}}}$

Wherein, N is the sum of existing Streaming Media, and i is that N section Streaming Media is by the ordinal number of popularity from certain Streaming Media after the extremely low sequence of height, P _kThe probability that the Streaming Media that is i for ordinal number occurs, K is the rank of a Streaming Media frequency of occurrences, α is adjustable parameter.

Further, described high temperature flow medium buffer, to Edge Server, successively is cached to Edge Server from height to low by temperature by high temperature Streaming Media, until the spatial cache utilance reaches 95%; When spatial cache did not deposit 5%, the total length of stored stream media was 40-60%.

Further, when spatial cache did not deposit 5%, the total length of stored stream media was 50%.

Further, the behavior of statistic of user accessing Streaming Media again, when the threshold value of video-on-demand times higher than video-on-demand times appears in the Streaming Media interval, should draw in Edge Server from the Streaming Media source server in interval, otherwise delete.

General threshold value only has 1, and when in certain sheet, having a plurality of regional program requests to be greater than threshold value, all preserve in these zones, and delete in several zones that this sheet is preserved originally.

Compared with prior art, combine the behavioural analysis of user's access, Streaming Media for different temperatures has adopted different caching methods, has effectively improved spatial cache utilance and user and has accessed quality, has greatly alleviated transmission pressure and the cost of backbone network to fringe node.

High temperature Streaming Media for 20%, caching method of the present invention is whole buffer memorys when spatial cache is sufficient, reach the effect that takes full advantage of space.And when inadequate buffer space, the hit rate that the use that the present invention takes is lower exchanges the strategy of 50% spatial cache for can realize greatly saving space, the purpose reduced costs in the situation that substantially meet user experience.

Low temperature Streaming Media for 80%, the present invention first simulates a model according to user's program request behavior, as adaptive template to the spatial cache initialization, then according to user's concrete program request behavior, dynamically increase deletion buffer memory fragment, in initialization, greatly saved so on the one hand spatial cache, reduce carrying cost, reduce as much as possible on the other hand the number of times that buffer memory is replaced, reach the effect that reduces transmission cost.

The accompanying drawing explanation

Fig. 1 is that to choose length be that the film of 5400 seconds carries out after the isometric section at 10 seconds intervals cutting into slices the broadcasting time statistical chart.

Fig. 2 is the function curve after to the statistics match according to the Lorentzian model.

Fig. 3 is the adaptive cache schematic diagram of 80% low temperature flow medium buffer.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described further, but embodiments of the present invention are not limited to this.

Adopt the Zipf regularity of distribution to assess Streaming Media, the rule from Zipf distributes, can draw Pareto Law, and namely 80% user only accesses 20% streaming medium content.20% Streaming Media can be defined as to the Streaming Media of high temperature thus, 80% Streaming Media is defined as the Streaming Media of low temperature.The present invention will propose different cache policies according to the temperature difference of Streaming Media.

In conjunction with high temperature Streaming Media hit rate and buffer memory space empty spare time degree, to the cache way of 20% high temperature Streaming Media:

I. when spatial cache is arranged, according to the Streaming Media temperature, 20% high temperature Streaming Media is cached to Edge Server successively until the spatial cache utilance reaches 95%;

Ii. when inadequate buffer space 5%, 50% of stored stream media total length, and then sacrifice the lower hit rate of the higher Streaming Media of temperature and exchange more memory spaces for, the hit rate of loss has very large probability to make up in remaining inferior backing.Wherein time backing refers to the backing of all preserving with respect to original whole media, because present backing all be kept at Edge Server, now by high temperature Streaming Media by temperature buffer memory successively from high to low, until the spatial cache utilance is while reaching certain value, again by the part of cache of remaining high temperature Streaming Media total length, and these not all sheets of storage just comprised inferior backing.

This high temperature flow medium buffer strategy, both taken full advantage of spatial cache, is both the playing request that has met most of user, greatly improves user experience.

Cache policy in the past is for the Streaming Media of the low temperature previous section at Edge Server on stored stream media, although reduced broadcasting start-up time, according to the hit rate computing formula, this strategy can not obviously improve hit rate.

Counting user program request behavior, the film of choosing length and be 5400 seconds carries out the isometric section at 10 seconds intervals, and the situation of then user being accessed to film is added up.Statistics is as shown in Figure 1: a film has two peak access region usually.First zone, at the initial period of film, generally appears at about 200 seconds, and second zone, in the film interstage, is also the climax part of most of films, generally appears between 2100 seconds and 2500 seconds.

According to the statistics of Fig. 1, abscissa for the section sequence number, ordinate is the broadcasting time of each section, utilizes the Lorentzian model, carries out piecewise fitting according to peak value, matched curve as shown in Figure 2:

The Lorentzian model:

$y={\mathrm{<figure-callout\; id="0"\; label="y"\; filenames="HDA00003497298200012.png"\; state="\{\{state\}\}">y</figure-callout>}}_0+\frac{2A}{\mathrm{\&pi;}}\&CenterDot;\frac{w}{4{(x-x_0)}^2+w^2}---\left(2\right)$

Fitting function:

$y={\mathrm{<figure-callout\; id="0"\; label="y"\; filenames="HDA00003497298200012.png"\; state="\{\{state\}\}">y</figure-callout>}}_0+\frac{{2A}_1}{\mathrm{\&pi;}}\&CenterDot;\frac{w_1}{4{(x-x_{c1})}^2+{w_1}^2}+\frac{{2A}_2}{\mathrm{\&pi;}}\&CenterDot;\frac{w_2}{4{(x-x_{c2})}^2+{w_2}^2}---\left(3\right)$

Through curve, finally can obtain fitting function:

$y=62.76707+\frac{74017.2727}{\mathrm{\&pi;}}\&CenterDot;\frac{105.46636}{4{(x+6.29677)}^2+11123.15309}+$

(4)

$\frac{38853.4685}{\mathrm{\&pi;}}\&CenterDot;\frac{82.85475}{4{(x-235.90001)}^2+6864.909598}$

Wherein in above-mentioned (2), A is the integral area on baseline under curve in (3), (4), w is halfwidth.Each sign represents anything

By the matched curve function, can be obtained, the section section region that Attraction Degree is high be about film front 8% and 38% to 48% between, fitting function can also be obtained the peak width at 2 peaks thus, and the initial value at peak, utilize this initial value to obtain the threshold value of a broadcasting time, this threshold value is used for adaptive cache afterwards.The peak width of obtaining two peaks herein is be used to asking threshold value, but threshold value is not peak value, and refers to a clicks value of the definite storage area relevant to memory space

According to above-mentioned analysis, draw a kind of cache policy based on user's access habits, for low temperature Streaming Media, the edge server storage film front 8% and 38% to 48% between zone.

All user's access situation is added up every day afterwards, if number of clicks draws resource to local cache higher than threshold value from upstream server, lower than deleting, as shown in Figure 3, thereby realize adaptive cache.

Adopt the present invention can realize following purpose:

1) reduce the backbone network bandwidth consumption;

2) reduce broadcast time-delay, avoid the user to ask buffered media to postpone excessive;

3) significantly improve Buffer Utilization and rate and byte hit;

4) take full advantage of the memory space of Edge Server;

5) improve user satisfaction (QoS).

Above-described embodiments of the present invention, do not form the restriction to protection range of the present invention.Any modification of having done within spiritual principles of the present invention, be equal to and replace and improvement etc., within all should being included in claim protection range of the present invention.

Claims (5)

1. an adaptive stream media caching method, is characterized in that, to being stored in Streaming Media in the Streaming Media source server, carrying out temperature and estimate, and according to estimation results, Streaming Media is cached to respectively to Edge Server, is specially:

The Streaming Media of inputting is carried out to temperature and estimate, Streaming Media is divided into to high temperature Streaming Media and low temperature Streaming Media;

The high and low temperature Streaming Media of difference buffer memory;

High temperature flow medium buffer is to Edge Server in the following ways: high temperature Streaming Media is arranged by the height of temperature, high temperature Streaming Media successively is cached to Edge Server from height to hanging down by temperature, until the spatial cache utilance reaches M, wherein M is 90%-96%;

When inadequate buffer space 100%-M, the total length of storing each Streaming Media is 40-60%;

Low temperature film is cached to Edge Server in the following ways:

Utilize the Lorentzian model, according to the behavior curve of user's program request Streaming Media number of times;

Program request position while determining that by fitting function peak value appears in curve, obtain the threshold value of video-on-demand times according to the program request position;

Streaming Media interval when occurring that video-on-demand times is peak value is cached to Edge Server.

2. adaptive stream media caching method according to claim 1, it is characterized in that, described Streaming Media to input carries out the mode that temperature estimates: the rule that adopts Zipf to distribute draws Pareto Law, namely 80% user only accesses 20% streaming medium content, 20% Streaming Media is defined as to high temperature Streaming Media, 80% Streaming Media is defined as low temperature Streaming Media.

3. adaptive stream media caching method according to claim 1 and 2, it is characterized in that, described high temperature flow medium buffer, to Edge Server, successively is cached to Edge Server from height to low by temperature by high temperature Streaming Media, until the spatial cache utilance reaches 95%; When spatial cache did not deposit 5%, the total length of stored stream media was 40-60%.

4. adaptive stream media caching method according to claim 3, is characterized in that, when spatial cache did not deposit 5%, the total length of stored stream media was 50%.

5. adaptive stream media caching method according to claim 1 and 2, it is characterized in that, the behavior of statistic of user accessing Streaming Media again, when the threshold value of video-on-demand times higher than video-on-demand times appears in Streaming Media, should from the Streaming Media source server, draw in Edge Server in interval, otherwise delete.

Images