10 ms
Uplink
1 ms
Uplink

packet

packet

loss 75%

loss 99%

Downlink

Downlink

packet

packet

loss 100%

loss 100%

From the above table, it can be seen that in the strong network, there are no abnormalities in uplink transmission and downlink transmission of the frame synchronization game. The downlink transmission comprises transmitting the data frames by the game server to the respective game terminals and the uplink transmission comprises transmitting the data frames by the respective game terminals to the game server. In the weak network, the uplink packet loss rate increases, and the downlink packet loss rate also increases. The downlink packet loss rate is greater than the uplink packet loss rate. It can be seen that in a weak network environment, there is a clear competitive relationship between the uplink and the downlink (e.g., the higher the uplink packet loss rate, the lower the downlink packet loss rate, and vice versa).

In a frame synchronization game, a server may cycle at a fixed frequency (such as 66 ms), receive uplink packets from clients within a period, pack the uplink packets, and then transmit the packed uplink packets to the respective clients. Therefore, based on an architecture of the frame synchronization game, there may be not only a network delay between the server and the clients, but also a fixed system delay. In the weak network environment, the competitive relationship exists between the uplink and the downlink. If the uplink transmission is not controlled to meet the requirements, an increase in the network delay of the downlink transmission may further increase the system delay.

Based on this, if there is a network congestion, in order to improve the user experience in the weak network situation and avoid the increase in the system delay caused by competition between the uplink and the downlink, a report time interval of the data frames may be adjusted on the basis of the network congestion trend. The network congestion trend may be in direct proportion to the report time interval.

The report time interval may be a time interval between two transmissions of the uplink data frames by the game terminals to the game server. The report time interval may be related to a report frequency. A smaller report frequency may indicate a longer report time interval. A larger report frequency may indicate a shorter report time interval.

Specifically, when the network congestion is in an uptrend, the report time interval may also increase. If the network congestion is in an uptrend, the report time interval of the game terminals may increase to avoid preempting of bandwidth resources of downlink packets caused by multiple transmissions of uplink packets.

For example, when it is predicted that the downlink network congestion is in the uptrend, the report time interval of the uplink data frames may be increased (e.g., the current report time interval may be prolonged). The report time interval can be gradually increased. For example, the report time interval may be increased by 6 ms at each time, and the congestion condition of the downlink data frames may be continued to be monitored to predict a downlink network congestion trend. If the downlink network congestion continues to be in the uptrend, the report time interval may be prolonged by 6 ms again.

Correspondingly, when the network congestion is in a downtrend, the report time interval may be also shortened. When the network congestion is in the downtrend, the report time interval of the uplink data frames of the game terminals may be shortened. For example, when it is predicted that the downlink network congestion is in the downtrend, the current report time interval may be shortened (e.g., the number of times of uplink packet transmissions may be increased), which effectively utilizes a change of network improvement.

Step 206: Transmit the uplink data frames to the game server according to the report time interval.

Specifically, each game terminal may actively adjust the uplink report time interval on the basis of the predicted network congestion trend and change its transmitting frequency of transmitting the uplink data frames in advance. The game terminal may wait to transmit uplink data frames if a time interval does not reach the report time interval. When the time interval satisfies the report time interval, the game terminal may report the uplink data frames to the game server.

The method for processing the data in the frame synchronization game described herein may comprise: predicting a downlink network congestion trend according to the congestion condition of the data frames that have been received (e.g., predicting the downlink network congestion trend in advance, and then adjusting the report time interval of the game terminals according to the predicted downlink network congestion trend). The network congestion trend may be in direct proportion to the report time interval. If the network congestion degree is in the uptrend, the report time interval may be increased. If the network congestion degree is in the downtrend, the report time interval may be shortened, thereby avoiding the situation of preempting of a bandwidth of the downlink packets caused by multiple transmissions of the uplink packets in case of the network congestion, reducing a delay of a frame synchronization link and improving the game operation experience.

In an example, before the operation of predicting a downlink network congestion trend according to a congestion condition of received downlink data frames, the method may further comprise: detecting network quality; and predicting, based on a determination, according to a network quality detection result, that the terminal is in a weak network environment, a downlink network congestion trend according to a congestion condition of received downlink data frames.

The method of this application may be implemented based on determining that the network quality is a weak environment. The network quality can be determined according to a network type, a packet loss rate, a delay, jitter, and/or the like.

The network type may be taken as an example for determining the network quality. If the terminal is in a 3G network, an HSPA network, or an HSPA+ network, it may be determined that the terminal is in a weak network environment.

The packet loss rate may be taken as an example for determining the network quality. If a calculated packet loss rate is greater than a preset value, it may be determined that the terminal is in a weak network environment.

The delay may be taken as an example for determining the network quality. If a detected delay is greater than a delay threshold, it may be determined that the terminal is in a weak network environment.

The jitter may be taken as an example for determining the network quality. If a detected jitter frequency is greater than a threshold, it may be determined that the terminal is in a weak network environment.

In this example, when it is detected that the terminal is in the weak network environment, the downlink network congestion trend can be predicted in advance. The report time interval of the game terminals can be adjusted according to the predicted downlink network congestion trend to avoid the situation of preempting of a bandwidth of the downlink packets caused by multiple transmissions of the uplink packets in the weak network. For example, the transmitting frequency may be changed in advance, so that a delay of an overall link can be reduced by the terminal in the weak network environment, and the operation experience of a user can be enhanced.

FIG. 3 shows a method of predicting a downlink network congestion trend according to a congestion condition of received downlink data frames. The method comprises the following steps:

Step 302: Predict a congestion value of a next downlink data frame of each of the downlink data frames according to the congestion condition of the received downlink data frames.

A congestion condition of the downlink data frames can be determined according to a time interval of receiving the downlink data frames by the game terminal. In the process of transmitting the downlink data frames, the downlink data frames may be transmitted at a fixed time interval. Correspondingly, the game terminal may receive the downlink data frames at a fixed receiving frequency. If there is no network congestion, a receiving time interval between two frames of downlink data frames of the game terminal may be fixed, which is consistent with a transmitting time interval of the downlink data frames. In the process of receiving the downlink data frames by the game terminal, if a congestion occurs, a time interval of receiving two frames of downlink data frames by the game terminal may become longer. Based on this, the congestion condition of the downlink data frames can be embodied (e.g., determined) according to the congestion values of the downlink data frames.

Specifically, in the frame synchronization game, in the process of transmitting the data frames by the game server, data frames and data packets may be transmitted at a fixed frequency. In the process of receiving the data frame and the data packets by the game terminals, if a network congestion occurs, a receiving interval will change. If a downlink network is not congested, the transmitting time interval of the downlink data frames may be consistent with the receiving time interval of the downlink data frames. For example, when the difference value between the transmitting time interval of the downlink data frames and the receiving time interval of the downlink data frames is equal to 0, it may indicate that the downlink network is not congested. If the receiving time interval of the downlink data frames is longer than the transmitting time interval (e.g., when the difference value between the receiving time interval of the downlink data frames and the transmitting time interval of the downlink data frames is greater than 0), it may indicate that the downlink network is congested.

A delay variable of the network congestion may conform to a normal distribution. Specifically, the network congestion value of each frame may be used as a delay variable. The delay variable d(i)=w(i) may be considered as a sampling point in a random process W. Furthermore, it may be a result of a combination of factors such as a link carrying capacity, a current network transmission state, and a current transmitting rate. If a network is in an over-use state, an expected collection value w(i) may increase. If the network is in an under-use state, the expected collection value w(i) may decrease. It can be seen that the random process W may conform to the normal distribution.

Meanwhile, if a game terminal receives the data frames at a slower speed because a network congestion occurs in one stage, the game terminal receives the data frames at a higher speed in a next stage when the network condition is improved. Therefore, in the long run, a cumulative congestion value tends to be 0.

Due to the above two characteristics, it can be inferred that in frame synchronization, the cumulative congestion values of a certain segment of the network may increase continuously according to a certain slope, and a slope trend of each segment is inconsistent. Therefore, the congestion value of the next downlink data frame can be predicted according to the cumulative congestion value of the current data frame.

Step 304: Predict the downlink network congestion trend according to a predicted congestion value sequence of the downlink data frames.

The predicted congestion value sequence of the downlink data frames may be the congestion values of a plurality of consecutive downlink data frames. For example, the predicted congestion value sequence of the downlink data frames may comprise 10 predicted congestion values of the downlink data frames. Curve fitting can be performed according to a segment of predicted congestion values of the downlink data frames to predict the downlink network congestion trend. The network congestion trend may represent a trend of increase or decrease in the degree of the downlink network congestion within a future time period, a time period, and/or a future time point.

In this example, by using the characteristics of the congestion values of the downlink data frames, the congestion value of each next downlink data frame can be reasonably and accurately predicted, so that the downlink network congestion trend can be predicted in advance according to the predicted congestion value sequence of the downlink data frames.

In another example, the step of predicting a congestion value of a next downlink data frame of each of the downlink data frames according to the congestion condition of the received downlink data frames, as shown in FIG. 4, may comprise:

Step 402: Determine a congestion value of each downlink data frame according to a difference value between a receiving time interval of two consecutive downlink data frames and a fixed transmitting interval of the downlink data frames.

In the frame synchronization game, in the process of transmitting the data frames by the game server, as shown in FIG. 5, data frames and data packets may be transmitted at a fixed frequency. At a fixed transmitting frequency, a fixed transmitting time interval of the downlink data frames is T(i)−T(i−1). In the process of receiving the data frames and the data packets by the game terminals, if network congestion occurs, the receiving interval may change, and the receiving time interval of the downlink data frames is t(i)−t(i−1). If a downlink network is not congested, the transmitting time interval of the downlink data frames may be consistent with the receiving time interval of the downlink data frames. For example, when the difference value between the transmitting time interval of the downlink data frames and the receiving time interval of the downlink data frames is equal to 0, it may indicate that the downlink network is not congested. If the receiving time interval of the downlink data frames is longer than the transmitting time interval (e.g., when the difference value between the receiving time interval of the downlink data frames and the transmitting time interval of the downlink data frames is greater than 0), it may indicate that the downlink network is congested.

The difference value between the receiving time interval of the downlink data frames and the transmitting time interval of the downlink data frames may be the congestion value of the downlink data frame. The congestion value of each downlink data frame may be:

d(i)=t(i)−t(i−1)−(T(i)−T(i−1))

Step 404: Predict the congestion value of the next downlink data frame according to the congestion value of the current downlink data frame and a current cumulative congestion value of the current downlink data frame.

The cumulative congestion value may be cumulation of the congestion values of a plurality of data frames. The current cumulative congestion value may be cumulation of the congestion values from the first data frame to the current data frame. A calculation formula of the current cumulative congestion value is: acc_d(i)+=d(i), where acc_d(i) represents the current cumulative congestion value, and d(i) represents the congestion value of an i^thdownlink data frame.

A delay variable of the network congestion may conform to a normal distribution. Specifically, the network congestion value of each frame is used as a delay variable. The delay variable d(i)=w(i) is considered as a sampling point in a random process W. Furthermore, it is a result of a combination of factors such as a link carrying capacity, a current network transmission state, and a current transmitting rate. If a network is in an over-use state, an expected collection value w(i) may increase. If the network is in an under-use state, the expected collection value w(i) may decrease. It can be seen that the random process W may conform to the normal distribution.

If a game terminal receives the data frames at a slower speed because a network congestion occurs in one stage, the game terminal receives the data frames at a higher speed in a next stage when the network condition is improved. Therefore, in the long run, a cumulative congestion value tends to be 0.

In this example, the congestion values of the downlink data frames can be accurately predicted on the basis of the distribution characteristic of the network congestion and the change characteristic of the cumulative congestion values by calculating the congestion values of the downlink data frames.

In another example, the step of predicting the congestion value of the next downlink data frame according to the congestion value of the current downlink data frame and a current cumulative congestion value of the current downlink data frame may comprise: performing exponential smoothing processing on the congestion value of the current downlink data frame and the cumulative congestion value of the current downlink data frame, and predicting the congestion value of the next downlink data frame, a weight of the congestion value of the current data frame being greater than a weight of the cumulative congestion value of the current downlink data frame.

In the frame synchronization game, the congestion values of the downlink data frames of a period of time may conform to the characteristics of the normal distribution. Based on this, it can be inferred that in frame synchronization, the cumulative congestion values of a certain segment of the network increase continuously according to a certain slope, and a slope trend of each segment may be inconsistent.

Actual measured changes in the congestion values of the downlink data frames of a segment are shown in FIG. 6, which are relatively scattered and irregular. In this example, the congestion value of the next downlink data frame may be predicted using an exponential smoothing method on the basis of the characteristic that the overall rule of the congestion values of the downlink data frames conforms to linear prediction.

Specifically, the exponential smoothing method may be actually a special weighted moving average method. This method has the characteristics below: Firstly, the exponential smoothing method further enhances the effect of recently observed values during an observation period on predicted values. Weights assigned to observed values at different time may be unequal, thereby increasing the weights of the recently observed values, so that the predicted values to quickly reflect the actual changes. The weights may be reduced in geometric progression. A first term of this progression may be a smoothing constant a, and a common ratio is (1−a). Secondly, the exponential smoothing method has scalability for the weights assigned to the observed values, and different values of a can be used to change a change rate of the weights. If a is a small value, the weight may change quickly, and a recent change trend of the observed values can be reflected more quickly in exponential moving average values. Therefore, by using the exponential smoothing method, different values of a can be selected to adjust the uniformity of the observed values of the time series (e.g., the smoothness of trend changes).

Single exponential smoothing may be as follows:

s_d(i)=a*s_d(i−1)+(1−a)*acc_d(i),

where s_d is the predicted cumulative congestion value of the downlink data frame, which reflects queuing delay information of the data packets passing through a network link and can reflect the deflation of a network queue. A predicted value is a weighted sum of previous observed values, where a and 1−a are weight coefficients, and different weights are assigned to different data. Specifically, the weight of the congestion value of the current downlink data frame may be greater than the weight of the current cumulative congestion value. The congestion value of the next frame may be predicted using the exponential smoothing method.

A time constant τ can be used to perform reverse inference to determine a reasonable value of a. Therefore, the weight of the predicted cumulative congestion value of the previous downlink data frame and the weight of the cumulative congestion value of the current data frame may be determined.

Specifically, according to an experiment, the time constant of the whole formula is 110 ms; the stabilization time is 440 ms; a signal value at time t is 1; 0.63 is contributed after 110 ms, and 0.99 is contributed after 440 ms. It can be considered as no contribution subsequently. Therefore, if the value of a is determined to be 0.6, the formula of the single exponential smoothing is:

s_d(i)=0.6*s_d(i−1)+(1−0.6)*acc_d(i)

After the changes in the congestion values of the downlink data frames in FIG. 6 using the above formula is smoothed, as shown in FIG. 7, it can be seen from FIG. 7 that after the exponential smoothing, there are obvious peaks and valleys in the changes in the congestion values of the downlink data frames.

In this example, the congestion value of the next downlink data frame may be predicted using the exponential smoothing method on the basis of the characteristic that the overall rule of the congestion values of the downlink data frames conforms to linear prediction, so that the prediction accuracy of the congestion values of the downlink data frames can be improved.

In another example, the step of predicting the downlink network congestion trend according to a predicted congestion value sequence of the downlink data frames may comprise: calculating a slope regression coefficient according to the predicted congestion value sequence of the downlink data frames; predicting, based on a determination that the slope regression coefficient is greater than 0, that the downlink network congestion is in an uptrend; and predicting, based on a determination that the slope regression coefficient is less than 0, that the downlink network congestion is in a downtrend.

Specifically, the predicted congestion value sequence of the downlink data frames refers to the congestion values of a plurality of consecutive downlink data frames. As shown in FIG. 7, after the single exponential smoothing is used, there are significant peaks and valleys in the changes of the predicted congestion values of the downlink data frames. The network congestion values of the downlink data frames of a certain segment may show a linear change, and the slope trend of each segment may be inconsistent. Based on this, the downlink network congestion trend can be predicted according to a predicted congestion value sequence of the downlink data frames of each segment. The predicted congestion value sequence of the downlink data frames that includes the congestion values of 10 smoothed downlink data frames is taken as an example. As shown in FIG. 7, average values of horizontal and vertical coordinates are calculated according to the horizontal and vertical coordinates of the 10 data frames. The horizontal coordinate represents the number of frames, and the vertical coordinate represents the congestion value.

The least squares method may be used to calculate a slope regression coefficient according to the average values of the horizontal and vertical coordinates of the congestion values of the respective data frames. Specifically, the numerator of the slope regression coefficient is:

numerator+=(p.X−averagex)*(p.Y−averagey);

- the denominator of the slope regression coefficient is:
  
  denominator+=(p.X−averagex)*(p.X−averagex); and
- the slope regression coefficient is:
  
  trend=numerator/denominator.

When the slope regression coefficient is greater than 0, it may indicate that the network congestion in a link decreases, and it may be predicted that the network congestion is in an uptrend. When the slope regression coefficient is less than 0, it may indicate that the network congestion in a link decreases, and it may be predicted that the network congestion is in a downtrend. As shown in FIG. 8, the slope regression coefficient of the congestion values of the data frames within each time period can be finally calculated to predict the downlink network congestion trend.

In this example, the slope regression coefficient may be calculated using the predicted congestion value sequence of the downlink data frames, and then the network congestion trend may be characterized by the slope regression coefficient. This method may be in line with the change trend of the congestion values of the downlink data frames within a period of time, thus accurately predicting the network congestion trend.

In another example, the step of adjusting a report time interval of uplink data frames on the basis of the downlink network congestion trend may comprise: adjusting, on the basis of the downlink network congestion trend, a transmitting frequency reported by the uplink data frames, where the transmitting frequency is decreased based on a determination that the downlink network congestion is in an uptrend, and the transmitting frequency is increased based on a determination that the downlink network congestion is in a downtrend; and determining the report time interval of the data frames according to the adjusted transmitting frequency.

Specifically, the transmitting frequency may be in reverse proportion to the report time interval. A low transmitting frequency may indicate a long report time interval. A high transmitting frequency may indicate a short report time interval. When the downlink network congestion is in the uptrend, the transmitting frequency may decrease, and the report time interval may increase. When the downlink network congestion is in the down trend, the transmitting frequency increases, and the report time interval decreases. In one example, the report time interval is: min(T)=66 ms/transmitting frequency, where 66 ms is a fixed transmitting frequency of the game server.

A decrease adjustment amount of the transmitting frequency may be greater than an increase adjustment amount of the transmitting frequency. For example, when the slope regression coefficient is less than 0, the transmitting frequency may be increased. During adjustment, one more transmission may be added at each time till 10 times per frame.

When the slope regression coefficient is greater than 0, the transmitting frequency is decreased. For example, the transmitting frequency is directly reduced by half, and rounding down is performed until 1 time per frame.

The decrease adjustment amount of the transmitting frequency is greater than the increase adjustment amount of the transmitting frequency, which has the effect of slow increase and fast decrease, so that the adjustment is more sensitive.

In one example, in a method for processing the data in the frame synchronization game, as shown in FIG. 9, game terminals may report data frames to a game server, and the game server may transmit downlink data frames to the game terminals at a fixed interval. Each game terminal may detect the network quality. According to a network quality detection result that the terminal is in a weak network environment, a congestion value of each downlink data frame may be determined according to a difference value between a receiving time interval of two consecutive downlink data frames and a fixed transmitting interval of the downlink data frames. A congestion value of a next downlink data frame may be predicted according to the congestion value of the current downlink data frame and a current cumulative congestion value. A downlink network congestion trend may be predicted according to the predicted congestion value sequence of the downlink data frames. A report time interval of uplink data frames may be adjusted on the basis of the downlink network congestion trend, where the downlink network congestion trend may be in direct proportion to the report time interval. The uplink data frames may be transmitted to the game server according to the report time interval.

Specifically, exponential smoothing processing may be performed on the congestion value of the current downlink data frame and the cumulative congestion value of the current downlink data frame, and the congestion value of the next downlink data frame may be predicted, where a weight of the congestion value of the current data frame may be greater than a weight of the cumulative congestion value of the current downlink data frame.

Specifically, a slope regression coefficient may be calculated according to the predicted congestion value sequence of the downlink data frames. Based on a determination that the slope regression coefficient is greater than 0, the downlink network congestion may be predicted to be in an uptrend. Based on a determination that the slope regression coefficient is less than 0, the downlink network congestion may be predicted to be in a downtrend.