CN102034263B - Shear-warp-based parallel volume rendering system - Google Patents

Abstract

The invention discloses a shear-warp-based parallel volume rendering system. The system is characterized in that: a distributed parallel visualization system is constructed on the basis of a high-speed shear-warp volume rendering algorithm aiming at the data visualization application requirement of a large-scale data field. Load balance between nodes and rendering speed are further improved by a task division method based on a scanning line and a load balance principle and an adaptive frame rate control strategy based on an army integrated air and missile defense (AIAMD) idea on the premise of guaranteeing quality of a visual image.

Description

A kind of parallel volume rendering system based on wrong shear deformation

Technical field

The present invention relates to a kind of parallel volume rendering service system, belong to the information visualization technical field based on wrong shear deformation.

Background technology

Direct volume drawing (abbreviation volume drawing) is a kind of important data fields method for visualizing.Different with iso-surface patch based on changeable shape, it directly with volume data according to certain mapping ruler composograph, have abundant expressive force.Therefore, have a wide range of applications in every field such as biology, medical science, Fluid Mechanics Computation, finite element analysis, Aero-Space, nuclear blast simulation, geologic prospecting, meteorologies.It is huge that volume drawing also has data volume, the characteristics that calculated load is strong but simultaneously.

Main volume rendering algorithm has light projecting algorithm (Ray-Casting), Splatting algorithm and wrong shear deformation (Shear-Warp) algorithm etc. at present.Wherein the Shear-Warp algorithm all is considered to fastest pure software volume rendering algorithm all the time.Data volume is big, data type is complicated, calculating strength is big because volume drawing has, and computing velocity becomes a subject matter of volume rendering algorithm slowly.Only rely on the raising of storage capacity of computing machine own and computing power, still be not enough to address this problem.Therefore various parallel accelerated methods to volume drawing have appearred.Volume rendering algorithm mainly is divided into the image space to be preface (like RayCasting) and to be two kinds of prefaces (like Splatting) with the object space, and corresponding, its parallel drawing algorithm also is divided into image parallel algorithm and object parallel algorithm etc.

Mainly there are following problems in above-mentioned parallel volume rendering system:

1) although wrong shear deformation (Shear-Warp) algorithm is the fastest direct volume drawing algorithm; But the parallel drawing system based on wrong shear deformation is less relatively; Need to take multinomial optimized Measures, with the performance of further raising parallel volume rendering system to the characteristics of wrong shear deformation algorithm.

2) with regard to the parallel volume rendering system that multinode is formed; Need provide a cover effective; Adaptive frame rate control and simultaneous techniques; Otherwise each node all adopts the strategy of drawing as possible and transmitting, and will cause a large amount of cognition systems that repeat to be calculated and the Network Transmission of identical topography, the performance of meaningless reduction parallel volume rendering system.

To above problem, the present invention proposes a kind of parallel volume rendering system based on wrong shear deformation, its core concept is: whole parallel volume rendering system is made up of through LAN is interconnected the PC crowd, and a host node and a plurality of from node is wherein arranged; Based on the volume drawing thought of wrong shear deformation, host node is responsible for the principle according to load balancing, and to respectively carrying out task scheduling from node, frame rate control and general are respectively from the synthetic final drawing result output of the local visualization result of node; The wrong shear deformation Visual calculation of the partial sweep line that the sweep trace task of being responsible for dividing according to host node from node is accomplished forms topography and also sends to host node.

With regard to whole parallel volume rendering system, can dynamically add and withdraw from from node, thereby parallel volume rendering of the present invention system is with good expansibility and dirigibility.In addition, to links such as task division method and the controls of self-adaptation frame per second, the present invention has proposed optimisation strategy and measure respectively, has further improved render speed.

Summary of the invention

The technical matters that the present invention will solve is: overcome the deficiency of prior art, based on wrong shear deformation (Shear-Warp) algorithm, provide an adaptation large-scale data field visual, have the parallel volume rendering system of higher render speed.

The technical scheme that the present invention adopts: based on the parallel volume rendering system of wrong shear deformation; Its characteristics are to adopt PC crowd system as shown in Figure 2; Host node with link to each other through LAN from node; Host node is as Control Node, be responsible for from the dynamic load of node with withdraw from management, Task Distribution and scheduling, frame synchronization control and synthesize from node topography; , be responsible for accomplishing local Visual calculation in the coordination of host node control down as computing node from node, formation partial sweep image also gathers to main frame, thereby constitutes a parallel volume drawing system.Native system is specifically realized parallel volume rendering through following steps, and is as shown in Figure 1:

(1) parallel volume rendering system initialization starts host node as server end, to monitor the connection request of slave as client, starts from node as client.

(2) host node is handled from the dynamic adding of node and is withdrawed from request.

(3) accept user interactions, host node receives user interactive, obtains viewpoint, modal position, and information such as direction of visual lines are calculated rendering parameter.

(4) based on the task division of scan line and load balancing principle; The rendering parameter that host node obtained based on (3) step and respectively (still do not have previous frame like certain from node and draw the time from the drafting time of node previous frame; Then will be somebody's turn to do from the previous frame drafting time of node and be initialized as 1/24 second); Host node is based on the visual required scan line quantity of carrying out of data fields; And respectively from drafting time of node previous frame; Again be respectively to distribute drafting task (promptly respectively from the scan line quantity and the sequence number of the required calculating of node next frame) from node, and be distributed to that each has connected from node;

(5) AIAMD time prediction parameter adaptive dynamically updates, host node upgrade with respectively from the relevant AIAMD time prediction model parameter of node.

(6) frame synchronization control, host node were calculated respectively from the zero hour that the node next frame is drawn, and were distributed to respectively from node, respectively controlled the drafting frame frequency of this node in view of the above from node.

(7) from parallel visual drafting of node and data transmission, respectively carry out the Visual calculation of partial sweep line according to the sweep trace task division and according to wrong shear deformation algorithm from node, form topography, and view data is passed to host node.

(8) final image generates, and host node is pressed the sweep trace preface and merged respectively from the node partial image data, synthetic final image output.

The present invention's beneficial effect compared with prior art is:

(1) relative prior art, support of the present invention is the dynamic adding of parallel computation node and withdraws from from node, thereby this parallel volume rendering system has better dirigibility and extensibility.

(2) according to the characteristics of wrong shear deformation (Shear-Warp) algorithm; The present invention is overlapping the task division optimisation strategy to having provided one in the parallelization process of this algorithm; Effectively guaranteed respectively from internodal load balancing, thereby improved the render speed of parallel volume rendering system.

(3) respectively repeat wrong shear deformation (Shear-Warp) Visual calculation of identical content and the re-transmitted of identical topography for avoiding from node; Meaningless reduction system performance; The present invention is based on the thought of AIAMD in the frame rate control strategy; And expand and improve, proposed one and overlapped the frame synchronization control strategy that is applicable to the parallel volume rendering system, guaranteed respectively to draw the consistance of frame frequency from node topography.

Description of drawings

Fig. 1 is the parallel volume rendering flow process diagram based on wrong shear deformation;

Fig. 2 forms structural drawing for the parallel volume rendering system;

Fig. 3 is the division of tasks synoptic diagram

Fig. 4 a, Fig. 4 b are wrong shear deformation (Shear-Warp) algorithm synoptic diagram;

Fig. 5 a, Fig. 5 b, Fig. 5 c are that 2 nodes are drawn time plot;

Fig. 6 a, Fig. 6 b, Fig. 6 c are that 4 nodes are drawn time plot;

Fig. 7 a, Fig. 7 b, Fig. 7 c are that 8 nodes are drawn time plot;

Fig. 8 a, Fig. 8 b, Fig. 8 c are that 16 nodes are drawn time plot;

Fig. 9 a, Fig. 9 b are the drafting effect of wrong shear deformation (Shear-Warp).

Embodiment

Below in conjunction with accompanying drawing and embodiment to further explain of the present invention.

(1) parallel volume rendering system initialization.

At first start host node as server end, start then from node as client, at whole parallel volume rendering run duration, host node keeps respectively from the monitoring of node.

(2) handle from the dynamic adding of node and withdraw from request.

This flow process cooperates by host node with from node to be accomplished;

Receive the request that dynamically adds from node like host node; Then with add from nodal information that host node is responsible for safeguarding to from node listing; These comprise from node IP address from nodal information; Previous frame is drawn the related sweep trace of task and is divided, and previous frame is drawn drafting time, network latency, the T.T. that task spent, and the parameter information that is used for frame rate control.Wherein sweep trace is divided into empty set, and the time is 0.

Receive the request of withdrawing from like host node, then will be somebody's turn to do from node listing, deleting of being responsible for safeguarding from nodal information from host node from node.

(3) accept user interactions.

Host node receives user interactions, obtains information such as viewpoint, modal position and direction of visual lines, calculates the required rendering parameter of volume drawing.This flow process is accomplished by main frame;

The user can accomplish two types alternately through mouse and keyboard operation, and one type is the translation and the rotation of viewpoint, and another kind of is the translation and the rotation of data fields.Main frame calculates new viewpoint position and data fields coordinate according to interaction results, forms new direction of visual lines, is promptly pointed to the direction of data fields centre coordinate by viewpoint, the rendering parameter that the draw calculation of above-mentioned information constituting body is required.

(4) based on the task division of sweep trace and load balancing principle.

Host node is respectively reformulated task division from the drafting task division and the drafting time of node according to rendering parameter and previous frame, and be distributed to that each has connected from node.

Because wrong shear deformation (ShearWarp) algorithm is a kind of intermediary image sequence algorithm, synthetic work is that the sweep trace according to topography carries out, and the synthetic work of each bar sweep trace does not disturb mutually, so the present invention divides respectively the drafting task from node according to sweep trace.Method for allocating tasks commonly used mainly contains according to the mean allocation method of topography's size with by the mean allocation method of actual estimation tasks amount.The former arrives each from node drawing synthetic task by topography's size mean allocation; Although respectively from node draw topography's size identical; But actual respectively from the load of the actual wrong shear deformation Visual calculation of node and unbalanced, this is by the visual characteristics decision of wrong shear deformation.The latter is that each is from the node allocating task by the actual computation load; Be that the general assignment amount is estimated by elder generation when beginning in every frame drafting cycle; Press general assignment amount mean allocation again and give slave; This strategy one side need be carried out extra task load calculation, and the accuracy of its result of calculation receives stochastic factor big on the other hand, and actual effect is undesirable.In view of the foregoing, the present invention has adopted a kind of more practical optimisation strategy, the Task Distribution optimisation strategy of promptly carrying out performance prediction and adjustment according to situation about respectively accomplishing from node previous frame task division.This flow process is cooperated by main frame and slave to be accomplished, and idiographic flow is following:

The first step: confirm the sweep trace quantity that next frame is drawn according to viewpoint and data fields coordinate range, be designated as S, as shown in Figure 3.If i the sweep trace quantity of being responsible at previous frame from node is PS _i, the T.T. that is spent (promptly drawing time and network latency) is TT _i, then i is from averaging time that the every sweep trace of node is spent

If i is initiate node from node, dt is set directly then _i=1;

Second step: according to the task loading condition of previous frame adjustment next frame respectively from the task division of node; Because the variation of front and back two frames can be very not big; Respectively under the situation of node task load balancing; All all should be T from node in the T.T. of the required cost of next frame, and establishing next frame i the sweep trace quantity of being responsible for from node is S _i, the Task Distribution results of optimization hopes to find the solution S _iAnd T, satisfy

While dt _iS _i=T is so solve

The 3rd step: the task division that next frame is new is distributed to respectively from node.

(5) AIAMD time prediction parameter adaptive dynamically updates.

In the parallel volume rendering system that does not add frame synchronization control; Respectively adopt the strategy of drawing and transmitting as possible as possible from node; This will cause respectively repeating identical Visual calculation and identical topography's transmission, meaningless computational load and the offered load that increases the weight of from node from node.The present invention is for fear of the problems referred to above; Consider getting in touch and distinguishing of drawing frames synchro control and the two types of problems of Frame-rate Control in the video flowing transmission course in the parallel volume rendering system; The present invention is based on the thought of AIAMD in the video stream frame rate control strategy, and expand and improve, provided one and overlapped the frame control strategy that is applicable to the parallel volume rendering system; Through to respectively sending the next frame start time, avoid the Visual calculation and the Network Transmission of repetition from node.

At first introduce some notions in the frame control strategy:

The drafting time (Render Time): RT _i(j), i=1,2 ..., N, j=1,2 ..., represent that i is utilized wrong shear deformation method for visualizing to draw the drafting time of j frame topography from node;

Network latency (Send Time): ST _i(j), i=1,2 ..., N, j=1,2 ..., represent that i is transmitted the time of j frame topography to host node from node;

T.T. (Total Time): TT _i(j), i=1,2 ..., N, j=1,2 ..., represent that i the j frame topography from node sends the T.T. that the host node buffer zone spends to from being generated to.

Frame frequency manager (Frame Rate Manager) runs on the host node, respectively confirms respectively the drafting zero hour from the node next frame from the drawing ability of node and network capacity through statistical study.

Frame display buffer (Display Buffer) is a buffer zone on the host node, is used for storing respectively the pixel data of the drawing image that transmits from node.

In the video flowing control method, AIAMD time prediction model is: T _K+1=alpha+beta T _k, k=1,2 ..., parameter alpha, the β that realizes the AIAMD strategy is through repeatedly testing the parameters optimization that draws, in whole frame per second control procedure, immobilizing.In the parallel volume rendering system that the present invention was directed against, there is bigger limitation in above-mentioned strategy.AIAMD time prediction model of the present invention adopts the mode of every frame adaptive Prediction Parameters update time to define, thereby the time prediction model modification is: T (j+1)=α _j+ β _jTT _Max(j), j=1 wherein, 2 ... Be frame number, α _j, β _jBe the time prediction parameter of j frame, need dynamically update that idiographic flow is following according to the different situations self-adaptation of each frame:

The first step, host node receive the every frame that respectively sends from joint and draw information, comprising:

The i node is at the drafting time of j frame: RT _i(j), i=1,2 ..., N, j=1,2,

The i node is at the data transfer time of j frame: ST _i(j), i=1,2 ..., N, j=1,2,

Calculate the T.T. that i node j frame drafting task is accomplished simultaneously:

TT _i(j)＝RT _i(j)+ST _i(j)，i＝1，2，…，N，j＝1，2，…；

Thereby try to achieve the T.T. of the maximum of all nodes in T.T. of present frame:

${\mathrm{TT}}_{\max }\left(j\right)=\underset{i}{\max }\left\{{\mathrm{TT}}_i\left(j\right)\right\},$ i＝1，2，…，N，j＝1，2，…；

Second step is according to the time prediction parameter alpha of the situation of change between adjacent two frame time informations to the employing of j frame _j, β _jCarry out span and define, as shown in table 1.

The tentative prediction of table 1 auto-adaptive parameter

Can construct the corresponding α of j frame according to last table _j, β _jSecond-order matrix:

X _2×1(j)＝[β _j?α _j] ^T；

In the 3rd step, construct j frame time corresponding matrix

${\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HSA00000337529300035.png"\; state="\{\{state\}\}">T</figure-callout>}}_{2\&times;2}\left(j\right)=\left[\begin{array}{cc}T{T}_{\max }\left(j\right)& 1\&times;\mathrm{sign}\left({\mathrm{\&alpha;}}_j\right)\\ -T{T}_{\max }\left(j\right)& -1\&times;\mathrm{sign}\left({\mathrm{\&alpha;}}_j\right)\end{array}\right];$

Construct the corresponding constant constraint matrix b of j frame _{2 * 1}(j), wherein C=0.001 is a fixed constant

${\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HSA00000337529300035.png"\; state="\{\{state\}\}">b</figure-callout>}}_{2\&times;1}\left(j\right)=\left[\begin{array}{c}T{T}_{\max }\left(j\right)+C\\ C-{\mathrm{TT}}_{\max }\left(j\right)\end{array}\right];$

In the 4th step, construct j frame corresponding relationship constraint function T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}And under the prerequisite of above-mentioned constraint, find the solution α (j), _j, β _jMake objective function minimum, promptly

min?Z：T(j)＝α _j+β _j·TT _max(j-1)，j∈{1，2，…}，

Problem is converted into the problem of in given range, finding the solution optimal value thus, promptly typical linear programming problem, and the present invention utilizes simplex algorithm to accomplish finding the solution of the problems referred to above.Simplex algorithm is the mathematical optimization algorithm by the numerical solution that is directed against linear programming problem of George Dantzig proposition; Simplex algorithm is according to the linear programming problem that provides; Certain feasible solution begins from feasible zone; Constantly change feasible solution, reach maximal value up to objective function, corresponding basic feasible solution is exactly an optimum solution.Concrete algorithm steps is following:

1) finds out variable, the objective function constraints of problem, and variable, the objective function constraints of problem turned to standard form, promptly set α _j, β _jSpan, structure equation of constraint T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}(j), target setting function Z={TT _Max(j)-[α _j+ β _jTT _Max(j-1)] } ²

2) root is according to T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}(j) list the matrix of coefficients that comprises 2 equation of constraint;

3) find out in the above-mentioned matrix of coefficients the independently base that constitutes of constraint factor vector of 2 linearities;

4) find out corresponding basic variable according to first base;

5) make in the matrix of coefficients that the nonbasic variable coefficient is zero entirely, obtain a basic feasible solution, obtain first objective function simultaneously;

6) analyze the check number that the nonbasic variable place is listed as; If also there is the nonbasic variable of positive test number; The possibility that the expression target function value possibly increase in addition, need exchange certain nonbasic variable and certain basic variable of confirming last time this moment, so that obtain more excellent target function value;

7) the corresponding nonbasic variable of selecting to have optimal value of check number is designated as x, changes to as new basic variable;

8) definite variable that need from base, swap out makes it to become nonbasic variable;

9) repeating step 5), 6), 7), 8), in the check number of nonbasic variable row, do not have positive number till;

10) be exactly this Optimal Solution of Linear Programming with maximum each corresponding decision variable value of target function value.

(6) frame synchronization control.

The Forecasting Methodology that the present invention is directed to T (j+1) has been used for reference Panagiotis; The thought of the AIAMD that people such as Vassilis propose; Compare the AIAMD method; Expanding to some extent aspect the condition that the present invention changes at T (j+1), comprising taking all factors into consideration respectively from drafting time, Same Scene image information that node is drawn same frame scene being sent to time of host node and the total three types of time information such as time sum of the two through network.

Host node is according to respectively spending situation from task time of node previous frame, calculate respectively from the zero hour that the node next frame is drawn, and be distributed to respectively from node, respectively controls the drafting frame frequency of this node in view of the above from node, and the idiographic flow of frame frequency synchro control is following:

The first step, the TT that host node obtains statistics _Max(j) give the frame frequency manager to handle, so that obtain the time block T (j+1) of next frame, and according to total temporal information of the temporal information performance prediction j+1 frame of j frame, wherein α _j, β _jBe that the self-adaptation that the j frame adopts is drawn the time prediction parameter, every frame is dynamically adjusted, and to the prediction of T (j+1), the present invention divides following three kinds of situation to handle:

Situation 1:, take additivity to reduce the T.T. that strategy leniently reduces the j+1 frame when T.T. of j frame during less than the j-1 frame:

T (j+1)=TT _Max(j)-| α _j|, j=1,2 ... For drawing frame number;

Situation 2: when the equating T.T. of T.T. of j frame and j-1 frame, parameter current is constant, and j+1 frame T.T. remains unchanged;

T (j+1)=TT _Max(j), j=1,2 ... For drawing frame number;

Situation 3: when T.T. of j frame during, if the drafting time strategy of greater than the network delivery time, then taking additivity to increase increases the T.T. of j+1 frame greater than T.T. of j-1 frame:

T (j+1)=TT _Max(j)+| α _j|, j=1,2 ... For drawing frame number;

If the network delivery time, is then explained the charge capacity of network greater than the drafting time increase,, should increase the T.T. of j+1 frame fast for fear of producing network congestion, so take the strategy of the property taken advantage of increase:

T (j+1)=β _jTT _Max(j), j=1,2 ... For drawing frame number;

Second step, respectively receive the predicted time piece of j+1 frame from node after, begin Visual calculation according to the sweep trace task division of i node based on wrong shear deformation, add up the drafting time RT of i node j+1 frame simultaneously _i(j+1) and delivery time ST _i(j+1), calculate the T.T. TT of the actual cost of j+1 frame _i(j+1) and the T.T. T (j+1) that sets with host node relatively, if TT _i(j+1)>=and T (j+1), then change the drawing of next frame immediately over to; Otherwise, if TT _i(j+1)＜and T (j+1), dormancy is full to T (j+1) time block, changes the drawing of next frame then over to.

(7) from parallel visual drafting of node and data transmission, respectively carry out the Visual calculation of partial sweep line according to the sweep trace task division and according to wrong shear deformation algorithm from node, form topography, and view data is passed to host node.

This flow process is by respectively accomplishing from node, and specific practice is as shown in Figure 4, has shown a data fields space (volume data) and projection plane among Fig. 4 a, and according to the direct projection of classic method, the image that obtains is AB.Fig. 4 b has illustrated the basic thought of wrong shear deformation (ShearWarp).At first; According to direction of visual lines (projecting direction); Determine a mid-plane, this plane is perpendicular to certain direction of principal axis in data fields space, and the data hierarchy in data fields space is carried out wrong contact transformation (Shear); Again with data fields spatial data vertical projection to mid-plane, obtain the A ' B ' of topography.Because the projection pattern that wrong shear deformation (ShearWarp) is adopted is a vertical projection, its calculated amount is far below the oblique projection of relative perspective.At last, the A ' B ' of topography that obtains is out of shape (Warp), obtains final image AB.

Above-mentioned Visual calculation sends it back main frame from node with Visual calculation result (topography) after accomplishing, and whether drops into next frame immediately by the frame synchronization control strategy decision that (6) step provided and draw or dormancy a period of time.

(8) final image generates.

Host node is pressed the sweep trace preface and is merged respectively from the node partial image data, synthetic final image output.

The test environment that following experimental result and analysis result rely on is: operating system Windows XP; CPU Intel Xeon4.0GHz; Video card NVIDIA GeForce GTX260; Network interface card ntel 82567LF-2 Gigabit Network Connection; Switch TP-LINK TL-SG1048 gigabit ethernet switch; The test body data are data fields 1M sampled data points.

Aspect the frame synchronization control performance; The frame synchornization method of application-level generally uses MPI at present; Adopt respectively in the experiment that the present invention provides based on AIAMD self-adaptation frame per second control method with adopt the MPI method and do not adopt any method for synchronous; The performance of frame synchronization under the checking equal conditions, when adopting 2,4,8,16 respectively during from node, experimental result such as Fig. 5-shown in Figure 8.Thus it is clear that, when not adopting any synchronization policy, draw the task different nodes and draw and have nothing in common with each other required T.T. of every frame; And when adopting the MPI method synchronous, it is close that the node that the drafting task is close is drawn required T.T. of every frame, and to draw required T.T. of every frame still obviously identical but draw the task different nodes; And adopt that the present invention provides based on the self-adaptation frame per second control method of AIAMD, embody stronger adaptive ability and can make being consistent basically of frame per second of each node.

Aspect control strategy; The contrast of frame synchronization control strategy of the present invention and video flowing control strategy is adopted 2,4,8,16 respectively during from node, is 10 with step-length; Required T.T. of every frame when recording the 10th frame to the 100 frames (unit: millisecond), experimental result is as shown in table 2:

The every frame T.T. contrast of the following two kinds of methods of the different interstitial contents of table 2

According to table 2 data, statistical separate out the following two kinds of methods of different interstitial contents every frame averaging time (unit: millisecond) and variance, as shown in table 3:

Two kinds of method time series analyses on the different nodes of table 3

Can find out that from above-mentioned experimental result the strategy that the video stream frame rate control strategy provided with the present invention on the averaging time of every frame is close basically, promptly the two all can play the frame synchronization effect in the parallel volume rendering system.But relatively variance can be found, adopts the time variance of video stream frame rate control strategy bigger, and the T.T. of different frame numbers changes greatly; And adopt the variance of self-adaptation frame per second control method of the present invention less, i.e. the needed time variation of each node is milder, has stationarity preferably.

The drafting effect is as shown in Figure 9, and parallel volume rendering system proposed by the invention compares the image quality no significant difference with the drawing result of classical wrong shear deformation (Shear-Warp) algorithm.

Claims (3)

1. based on the parallel volume rendering method of wrong shear deformation; It is characterized in that being the basis with wrong shear deformation (Shear-Warp) volume rendering algorithm; Construct distributed parallel volume rendering system; This system is made up of from node a host node and some, and host node and all adopt ordinary PC from node connects through LAN between node; This parallel volume rendering system is by being the self-adaptation frame per second control strategy of Additive Increase/Additive Decrease, Additive Increase/Multiplicative Decrease thought based on the task division method of sweep trace and load balancing principle with based on AIAMD thought; Under the prerequisite that guarantees visual image quality quality, further improve internodal load balancing property and render speed; Specifically may further comprise the steps:

(1) starts host node as server end,, start from node as client to monitor from from the connection request of node as client;

(2) handle from the dynamic adding of node and withdraw from request;

(3) host node receives user interactive, obtains in viewpoint, modal position, the direction of visual lines information at least one, calculates rendering parameter;

(4) rendering parameter that obtains based on (3) step of host node and respectively from the visual required scan line quantity of carrying out of drafting time, the data fields of node previous frame, and respectively from drafting time of node previous frame; Again for respectively distributing the drafting task from node; Scan line quantity and sequence number that said drafting task is the required calculating of next frame, and be distributed to that each has connected from node;

(5) host node upgrade with respectively from the relevant AIAMD time prediction model parameter of node; Said AIAMD time prediction model is: T (j+1)=α _j+ β _jTT _Max(j), j=1 wherein, 2 ... Be frame number, α _j, β _jIt is the time prediction parameter of j frame; I=1,2 ..., N, j=1,2 ... Be the T.T. of the maximum of all nodes in T.T.; T (j+1) is the time block of j+1 frame;

(6) host node calculated respectively from the zero hour that the node next frame is drawn, and was distributed to respectively from node, respectively from node based on the drafting frame frequency of controlling this node the said zero hour;

(7) respectively carry out the Visual calculation of partial sweep line according to the sweep trace task division and according to wrong shear deformation algorithm, form topography, and partial image data is passed to host node from node;

(8) host node is pressed the merging of sweep trace preface respectively from the node partial image data, synthetic final image and output.

2. the parallel volume rendering method based on wrong shear deformation according to claim 1 is characterized in that: the AIAMD time prediction parameter update in the said step (5) is that self-adaptation dynamically updates, and specifically comprises:

(5.1) host node receives the every frame that respectively sends from joint and draws information, comprising:

The i node is at the drafting time of j frame: RT _i(j), i=1,2 ..., N, j=1,2,

The i node is at the data transfer time of j frame: ST _i(j), i=1,2 ..., N, j=1,2,

Calculate the T.T. that i node j frame drafting task is accomplished simultaneously:

TT _i(j)＝RT _i(j)+ST _i(j)，i＝1，2，…，N，j＝1，2，…；

Thereby try to achieve the T.T. of the maximum of all nodes in T.T. of present frame:

${\mathrm{TT}}_{\max }\left(j\right)=\underset{i}{\max }\left\{{\mathrm{TT}}_i\left(j\right)\right\},i=\mathrm{1,2},...,N,j=\mathrm{1,2},...;$

(5.2) the time prediction parameter alpha that according to the situation of change between adjacent two frame time informations j frame is adopted _j, β _jCarry out span and define, divide four kinds of situation:

Situation 1:TT _Max(j)＜TT _Max(j-1), α then _j＜0; β _j=1

Situation 2:TT _Max(j)=TT _Max(j-1), α then _j=0; β _j=1

Situation 3:TT _Max(j)＞TT _Max(j-1)

And there be i node R T _i(j)-RT _i(j-1)＞ST _i(j)-ST _i(j-1)

α then _j＞0; β _j＞1

Situation 3:TT _Max(j)＞TT _Max(j-1)

And there be i node R T _i(j)-RT _i(j-1)＜ST _i(j)-ST _i(j-1)

α then _j＞0; β _jThere is not constraint

Can construct the corresponding α of j frame according to said method _j, β _jMatrix

X _2×1(j)＝[β _j?α _j] ^T；

(5.3) structure j frame time corresponding matrix

$T_{2\&times;2}\left(j\right)=\left[\begin{array}{cc}{\mathrm{TT}}_{\max }\left(j\right)& 1\&times;\mathrm{sign}\left({\mathrm{\&alpha;}}_j\right)\\ {-\mathrm{TT}}_{\max }\left(j\right)& -1\&times;\mathrm{sign}\left({\mathrm{\&alpha;}}_j\right)\end{array}\right];$

Construct the corresponding constant constraint matrix b of j frame _{2 * 1}(j), wherein C=0.001 is a fixed constant

$b_{2\&times;1}\left(j\right)=\left[\begin{array}{c}{\mathrm{TT}}_{\max }\left(j\right)+C\\ C-{\mathrm{TT}}_{\max }\left(j\right)\end{array}\right];$

(5.4) structure j frame corresponding relationship constraint function T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}And under this constraint condition, find the solution α (j), _j, β _jMake objective function minimum, promptly

min Z∶T(j)＝α _j+β _j·TT _max(j-1)，j∈{1，2，…}

α thus _j, β _jThe adaptive prediction problem be converted into and in given range, find the solution optimum α _j, β _jValue makes the minimum problem of objective function, and promptly typical linear programming problem utilizes simplex algorithm to accomplish finding the solution of the problems referred to above; Concrete step is following:

1) sets α _j, β _jSpan, structure equation of constraint T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}(j), target setting function

Z＝{TT _max(j)-[α _j+β _j·TT _max(j-1)]} ²；

2) according to T _{2 * 2}(j) X _{2 * 1}(j)≤b _{2 * 1}(j) list the matrix of coefficients that comprises 2 equation of constraint;

3) find out in the above-mentioned matrix of coefficients the independently base that constitutes of constraint factor vector of 2 linearities;

4) find out corresponding basic variable according to first base;

5) make in the matrix of coefficients that the nonbasic variable coefficient is zero entirely, obtain a basic feasible solution, obtain first target function value simultaneously;

6) analyze the check number that the nonbasic variable place is listed as; If also there is the nonbasic variable of positive test number; The possibility that the expression target function value possibly increase in addition, need exchange certain nonbasic variable and certain basic variable of confirming last time this moment, so that obtain more excellent target function value;

7) the corresponding nonbasic variable of selecting to have optimal value of check number is designated as x, changes to as new basic variable;

8) definite variable that need from base, swap out makes it to become nonbasic variable;

9) repeating step 5), 6), 7), 8), in the check number of nonbasic variable row, do not have positive number till;

10) be exactly this Optimal Solution of Linear Programming with maximum each corresponding decision variable value of target function value.

3. the parallel volume rendering method based on wrong shear deformation according to claim 1 is characterized in that: in the said step (6) respectively from node according to the drafting frame frequency of controlling this node the said zero hour, be specially:

(6.1) host node TT that statistics is obtained _Max(j) give the frame frequency manager to handle, so that obtain the time block T (j+1) of next frame, and according to total temporal information of the temporal information performance prediction j+1 frame of j frame, wherein α _j, β _jBe that the self-adaptation that the j frame adopts is drawn the time prediction parameter, every frame is dynamically adjusted, and to the prediction of T (j+1), the present invention divides following three kinds of situation to handle:

Situation 1:, take additivity to reduce the T.T. that strategy leniently reduces the j+1 frame when T.T. of j frame during less than the j-1 frame:

T (j+1)=TT _Max(j)-| α _j|, j=1,2 ... For drawing frame number;

Situation 2: when the equating T.T. of T.T. of j frame and j-1 frame, parameter current is constant, and j+1 frame T.T. remains unchanged;

T (j+1)=TT _Max(j), j=1,2 ... For drawing frame number;

Situation 3: when T.T. of j frame during, if the drafting time strategy of greater than the network delivery time, then taking additivity to increase increases the T.T. of j+1 frame greater than T.T. of j-1 frame:

T (j+1)=TT _Max(j)+| α _j|, j=1,2 ... For drawing frame number;

If the network delivery time, is then explained the charge capacity of network greater than the drafting time increase, taked the strategy of the property taken advantage of increase:

T (j+1)=β _jTT _Max(j), j=1,2 ... For drawing frame number;

(6.2) respectively receive the predicted time piece of j+1 frame from node after, begin Visual calculation according to the sweep trace task division of i node based on wrong shear deformation, add up the drafting time RT of i node j+1 frame simultaneously _i(j+1) and delivery time ST _i(j+1), calculate the T.T. TT of the actual cost of j+1 frame _i(j+1) and the T.T. T (j+1) that sets with host node relatively, if TT _i(j+1)>=and T (j+1), then change the drawing of next frame immediately over to; Otherwise, if TT _i(j+1)＜and T (j+1), dormancy is full to T (j+1) time block, changes the drawing of next frame then over to.

Images