Patent Grant
7478166
1. Field of the Invention
The invention described herein relates to publication of media content through a computer network.
2. Related Art
The popularity of the World Wide Web (“Web”) and services provided on the Web has increased the demands placed upon Web sites. Server farms made up of a number of servers are sometimes used to provide computing power for a Web site. The load capacity on a single Internet Protocol (IP) based application server such as a Web server is often fixed. When a load upon a server reaches a certain level, performance of the server can degrade.
Therefore, given a server farm, load balancers can be used to distribute computing work load across different servers.
Load balancer 100 has a number of limitations. First, load balancer 100 does not discriminate among requests. All requests are handled equally regardless of whether a particular request involves more complex computing or a higher workload, relative to other requests. Treating a high workload request in the same manner as a low workload request can create an imbalance across servers. Second, load balancer 100 can manage multiple unrelated pools of servers, but is not aware of the data flow between pools of servers. This hampers load management. Third, the configuration needed in load balancer 100 can make it difficult to retire or reassign servers frequently or automatically. Operator intervention may be required, since the retiring or reassignment of a server must generally be carefully synchronized with changes on the load balanced machines.
Load balancer 100 generally has a limited fault tolerance, based only on server availability. Load balancer 100 may perform a TCP-level check on the availability of Web servers 120-140 and distribute requests to available Web servers. This distribution (based on availability) is fairly crude and does not take into account whether an available server can provide a useful service in light of a particular task.
Conventional load balancers are generally limited by a lack of understanding of higher layer operations being carried out by servers. For example, many conventional load balancers work at a transport layer (OSI Layer 4) using TCP. At this layer, a load balancer can only determine that a server is willing to accept a connection, but not whether the server can satisfy a given request sent on that connection.
Moreover, a request from a user may not require the attention of a server in farm 110. Alternatives may exist for the satisfaction of a user request. A server external to farm 110 may best be able to handle the request, for example. Alternatively, a request for data could be handled by reading the requested data from a memory unit, such as a cache, and delivering the data to the user. In either of these cases, the processing of one of the servers 120-140 is not necessary.
Therefore, a need exists for a system that can intelligently process a user request, either by directing the processing of the request to one of a set of servers so as to effectively balance the load among the servers, or by directing the processing of the request to an appropriate alternative means for satisfying the request.
The inventors have recognized that limitations in conventional load balancers make them unsuitable for handling certain tasks, particularly those related to on-demand media transcoding. Transcoding servers and distribution servers that carry out on-demand media transcoding and publication involve complex server resources that require sophisticated balancing of transcoding tasks on the fly. What is needed is a method and system for intelligently processing a user request for media content, either by directing the processing of the request to one of a set of transcoding servers so as to effectively balance the load among the servers, or by directing the processing of the request to an appropriate alternative means for satisfying the request.
The present invention is directed to a system, method, and computer program product for publishing transcoded media content in response to publishing service requests from end users. The topology of the system is illustrated generally in
Note that the requested content originates as source media content from an origin server of a publisher such as CNN (publisher 217) or PBS (publisher 225). At the time of a publishing service request, the source media content may or may not reside at a publisher's origin server. The source media content may reside at a server at publishing service farm 210 at the time of the request, or may reside somewhere else reachable by publishing service farm 210.
Source media content can be any digital media content including but not limited to audio, video, streaming video, streaming audio, movies, video clips, playlists, pointer files, metafiles, HTML pages, etc. Transcoded media is media content that has been transcoded from a source format into a destination format suitable for a client of end user 215. The publishing service farm 210 can, as necessary, perform this transcoding of the source media content. In one example, not intended to limit the scope of this invention, the source media content would have the following properties:
video encoding: MPEG2,
audio encoding: MPEG2,
data rate: 1.6 megabits per second,
image size: 320×340 pixels,
video content: talking head,
audio content: voice.
The destination format would have the following properties:
video encoding: windows media 8,
audio encoding: windows media 8,
data rate: 300 kilobits per second,
image size: 160×120.
In particular, as shown in
In an embodiment of the invention, the media provider farm 330 comprises a media provider request processor 340, a plurality of transcoding servers 350, a plurality of distribution servers 360, and a cache memory 370. The publishing service request processor 310 generates a media provider request to media provider request processor 340 in the media provider farm 330. The media provider request specifies one of several types of processing, depending on the media content requested, the constraints of the client, and resources available to the media provider farm. First, the media provider request can specify on-demand transcoding, in which source media content is transcoded by one of several transcoding servers 350 in the media provider farm 330, to produce the media content in the requested format. The transcoded media content is then sent to a destination client through a distribution server 360 in the media provider farm. Second, the media provider request can specify a cache access, so that previously transcoded media content, resident in the cache 370 in the media provider farm 330, can be used to fulfill the client's publishing service request. Again, the transcoded media content is sent to the destination client through a distribution server 360 in the media provider farm 330. Third, the media provider request can specify pass-through processing, in which the requesting client is directed to an origin server (not shown) from which the requested media content can be provided directly.
An embodiment of the invention also allows for transcoding independent of any particular client or user. Here, a media provider request scheduler 320 creates and sends a batch media provider request to the media provider request processor 340, specifying the source media content to be transcoded and a format. Like a media provider request made in the on-demand mode, such a request generated in the batch mode can include, but is not limited to, one or more of the following constraints relevant to clients: client media format, bit rate of the client connection, client player version, client locale (country of residence), client native language, requested height and/or width of a display window (e.g. 320×240 pixels), etc. As in the case of on-demand transcoding, a transcoding server 350 transcodes the media content (obtained from an origin server) in the specified format. The resulting transcoded media content is then cached, in anticipation of future publishing service requests from end users.
When a transcoding task is to be performed in the media provider farm (i.e., in the on-demand and batch cases), the invention provides a process for selecting a transcoding server. The process seeks to select a particular transcoding server in a manner that balances the totality of ongoing transcoding tasks across all such servers. The process considers the processing load that will be created by the transcoding task and the current load being borne by each transcoding server. According to one embodiment of the invention, the media provider request processor has a detailed understanding of the application layer (OSI layer 7) including application(s) performed by each type of transcoding server. This understanding is used to allocate transcoding tasks such that the work of servers is balanced more effectively. For example, unavailable servers are culled more effectively. This in turn provides a better quality of service to the end user. According to a further feature, the media provider request processor can even understand and balance transcoding servers which do not implement an industry-standard protocol like HTTP.
In an embodiment of the invention, each transcoding task is assigned a priority that can be based, for example, on the number of viewers seeking the associated media content. A higher priority transcoding task will displace a task of lower priority. The lower priority task can be “killed.”
Additional features and advantages of the invention will be set forth in the description that follows, and in part will be apparent from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the system and method particularly pointed out in the written description and claims hereof as well as the appended drawings.
The foregoing and other features and advantages of the invention will be apparent from the following, more particular description of a preferred embodiment of the invention, as illustrated in the accompanying drawings:
Table of Contents
The invention described herein is a system, method, and computer program product that allows an end user to access media content across a network. In particular, the invention accepts a request from the user's client machine and sends the requested media content to a destination client in a format that is usable in light of the destination client's configuration. Note that the requesting client and the destination may or may not be the same machine. The requested media content in its original form (denoted herein as source media content) may be of a different format than what is desired or required by the end user. In this situation, the source media content is transcoded, by transcoding servers, into a format usable by the user at the destination client. In some situations, the requested media content will have already been transcoded and stored in a cache. In such a case, the already transcoded media content is simply distributed to the client. In other cases, if the source media content is already in an appropriate format for the end user, the transcoding process will effectively be bypassed, so that the client is simply directed to the origin server on which the source media content resides.
Source media content may originate from a storage device or another distribution server. The storage device or server may reside at any network location reachable by the transcoding servers. For example, the storage device or server may be housed in the same facility as the transcoding servers, or it may reside elsewhere on the Internet. The source media content may be stored (pre-recorded), or it may consist of live content being digitized as it is being distributed.
One or more transcoding servers can be used to transcode source media content into a format usable by the user at the destination client. The invention includes a process by which the load placed on the transcoding servers is generally balanced. This facilitates efficient usage of available transcoding capacity. In an embodiment of the invention, transcoding tasks can be prioritized, so that more important transcoding tasks are performed, while less important transcoding tasks are deferred or killed.
In some situations, it may be desirable to allow some variants of the constraints that would otherwise be placed on a transcoding task. Such constraints include those related to the format requested or required by a client. “Loosening” these constraints may allow a faster or more efficient completion of a transcoding task. This would then allow a faster response to the end user. Relaxing the constraints may also allow delivery of the media content in a superior format than what was requested. Moreover, if transcoded media content is already available in cache memory, such content may be provided to the client even if the format of such material does not exactly match that specified by the requesting client.
As described in this section, and as will be seen in greater detail below, a number of options exist as to how a user's request for media content can be filled. It is a feature of this invention, that, given a request from an end user, all options are evaluated and an appropriate response is made in light of the request and the resources available for fulfilling the request.
The invention is illustrated generally in
Publishing service request processor 310 may, for example, determine that source media content needs to be obtained from an origin server and transcoded. Publishing service request processor 310 may alternatively decide that the requested transcoded media is already available, and in storage within publishing service farm 210. In this case, the requested transcoded media content would simply be read from memory and provided to the destination client. Publishing service request processor 310 can alternatively decide that the requesting client should be redirected to the origin server itself so that the desired media content can be delivered directly from the origin server.
The specific option identified by the publishing service request processor 310 is then conveyed in a media provider request to media provider farm 330. In particular, the media provider request is sent to media provider request processor 340. If the media provider request specifies that a transcoding task is to be executed in response to the client's request (i.e., “on-demand”) than media provider request processor 340 initiates a transcoding task at one of several transcoding servers, such as transcoding server 350. As will be described in greater detail below, the selection of a transcoding server is based on the transcoding task and on the attributes of the assorted transcoding servers. These attributes include the available transcoding capacity on each transcoding server. In initiating a transcoding task, media provider request processor 340 will also consider the load that the transcoding task is expected to create on a transcoding server. An effort is made to select a transcoding server such that the total processing load across all servers is generally balanced. In addition, the priority of the transcoding task can also be a factor, in that a task of higher priority will be allocated to the transcoding server best able to execute the transcoding task in a timely manner. In an embodiment of the invention, a higher priority transcoding task can displace a lower priority transcoding task in a manner described in greater detail below.
If the media provider request sent by publishing service request processor 310 specifies that the media provider request is to be fulfilled by media content that has already been transcoded, media provider request processor 340 accesses the already transcoded media content from a cache memory 370. Such transcoded media content can then be sent to the destination client. Cache 370 is used to service requests that would otherwise consume transcoding resources. To facilitate access, the client constraints used to create the cache entry are associated with each cache entry.
All transcoded media that is provided by media provider farm 330 to the client is distributed to the client through one of several distribution servers, such as distribution server 360. This applies to media content that has been accessed from cache 370 or media content that has been transcoded by one of the transcoding servers. Distribution is initiated by a request from media provider request processor 340 to distribution server 360. The selection of a particular distribution server is based on availability and on the type of delivery service required. For example, in some cases a hypertext transfer protocol (HTTP) server is appropriate; in other cases, a RealServer (RTSP protocol), Windows Media Server (MMS protocol), or QuickTime server (RTSP protocol) is appropriate. Distribution server 360, which can be a streaming or download server, serves transcoded media content by either streaming it or downloading it to the destination client or to another distribution server. Examples of distribution processes include, but are not limited to, RealNetworks Real audio/video streaming, Apple QuickTime audio/video streaming, Microsoft Windows Media audio/video streaming, shoutcast/MP3 streaming, and HTTP downloading of RealNetworks, Microsoft, QuickTime, or Generic Media gMovie player movie files.
Media provider request processor 340 can also accept media provider requests from media provider request scheduler 320. Scheduler 320 acts autonomously from any particular client. Such a request from media provider request scheduler 320 can be viewed as a pre-encode batch media provider request. Based on such a request, media provider request processor 340 initiates a transcoding task, resulting in transcoded media content that is then stored in cache 370. This results in the ready availability of transcoded (pre-encoded) media content in media provider farm 330, so that future publishing service requests can be handled without transcoding on-demand.
II. System
The system of
A. On-Demand
Transcoding server 350 then performs the transcode task in a manner specified by the media provider request. The resulting transcoded media content is then sent to a distribution server such as distribution server 360. Note that in an embodiment of the invention, the transcoded media content is also sent to cache 370. This allows subsequent publishing service requests from client 402 or from any other client to be serviced by media provider farm 330 without having to repeat the transcode task. Distribution server 360 passes the transcoded media content to publishing service request processor 310, which forwards the transcoded media content to client 402, or to whatever client was specified in the publishing service request.
B. Cache
The system of the invention is illustrated in
Media provider request processor 340 then chooses a distribution server, such as distribution server 360. A distribution request is sent from media provider request processor 340 to distribution server 360, instructing distribution server 360 to distribute the requested transcoded media content from a specified file in cache 370 to the client. The transcoded media content is then read from cache 370 and forwarded to publishing service request processor 310 through the chosen distribution server 360. The requested content is then forwarded to client 402, or to whatever client was specified in the publishing service request.
C. Pass-Through
The system of the invention can also operate to direct a client to an origin server, from which the client can obtain the desired media content directly. This is illustrated in
D. Batch
In addition, the system of the invention can also perform transcoding in anticipation of future client requests. This is illustrated in
E. Intelligent Publishing Agent and Load Monitors
The system illustrated in
A media provider request is then sent from publishing service request processor 810 to media provider request processor 840. If the media provider request specifies that transcoding is to be performed, then a transcoding task must be initiated at a transcoding server. To facilitate the allocation of the transcoding task to a particular transcoding server, the load to be placed on the transcoding server by the particular transcoding task must be evaluated. This is the job of the task load monitor 844. The nature of on-demand media transcoding is such that it is difficult to predict the load an individual task will put on the system. In the case of the transcoding servers, the load is a function of the source media content, requested destination format, and server hardware. Determining the resulting load is sufficiently complex that one cannot reliably compute it before the transcode begins. This is primarily due to the wide variety of source and destination properties that the transcoding servers collectively support. In addition, variations in server hardware amongst the transcoding farm servers (CPU clock speed, hard drive performance, hardware acceleration boards, etc.) make some servers more efficient than others at certain tasks (decodes or encodes of specific formats, color coordinate transformations, etc.).
Task load monitor 844 estimates task load by constructing a model of required load for a given request based on previous transcodes. Upon completion of any transcode, transcoding tasks record the source, destination, and server properties along with statistics on their load (e.g., average CPU load, maximum CPU load) into a database 843. During the server selection process, task load monitor 844 estimates the prospective load of a requested task by comparing source, destination and potential server properties with that of previous transcodes. If an exact properties match is found, the load of the previous matching transcode is used as the load estimate for the new task. If an exact match cannot be found, the closest properties match is used as the load estimate for the new task. In the difference calculation, certain properties are given more weight than others, as will be described below.
In addition, the current processing load on each of the transcoding servers (i.e., load value) must be considered. One aim of the present invention is to maintain an accurate load value for all servers, so that the media provider request provider 840 does not mistakenly overload or under utilize a server. This is the job of the server load monitor 842. Transcoding servers periodically report their load to the server load monitor 842. Between the time at which a task is allocated on a transcoding server and the time at which the server next reports an existing load that includes that task, it will appear to the media provider request provider 840 as if the new task has placed no load whatsoever on the server. One can imagine a scenario in which a task is assigned to a transcoding server, and the media provider request provider's load value for that server is not updated in a timely manner. The media provider request provider 840 may mistakenly assign more tasks to the server, eventually overloading it.
To overcome this race condition, server load monitor 842 tracks the state of new tasks and assigns an estimated load to tasks until they reach a mature, or running, state. During this time, server load monitor 842 calculates the server load as the measured current server load plus the load estimate for each of the newly allocated tasks on that transcoding server. Once the task has reached the mature, running state, a server handling the task updates its reported state to “running” to indicate that its load is accurately reflected in the returned measured server load. This has the effect of reserving a portion of the transcoding server for the new task until the task can start up and report its actual measured load, avoiding the race condition.
In the embodiment of
The choice of which transcodes to perform speculatively can be motivated in several ways:
Source media content may also, for example, be transcoded to meet a schedule specified by a publisher.
F. Environment.
Referring again to
The illustrated operating environment includes an end user 215, origin servers 217 and 225, a publishing service farm 210, and a network 201. Only one user and two origin servers are shown for clarity. In general, any number of these components can be included in the system of the present invention. Moreover, in
The end user 215, the origin servers 217 and 225, and the publishing service farm 210 are all connected via a network 201. The network 201 connects all the components of the present invention, and can be any type of computer network or combination of networks including, but not limited to, circuit switched and/or packet switched networks, as well as wireless networks. In one example, the network 201 includes the Internet.
Any conventional communication protocol can be used to support communication between the components of the system. For example, a Transmission Control Protocol/Internet Protocol (TCP/IP) suite can be used to establish links and transport data and Real-Time Streaming Protocol (RTSP) can be used to stream data between components of the system. A World Wide Web-based application layer and browser (and Web server) can also be used to further facilitate communication between the components shown in
The destination client is capable of receiving and playing various types of media content. For example, the user client may receive and/or play media content in various well-known encoded formats including, but not limited to, MPEG, AVI, MP3, REAL, WINDOWS MEDIA, QUICK TIME, H.263 video coding, and PALM-compatible formats.
The origin servers 217 and 225 are used by their respective content providers to publish and/or transmit media content over the network 201. An origin server can provide media content using a variety of media input devices and programs. For example, media content can be provided using cameras (8 mm, Hi-8, or any video digitizing device), line-in/microphone (either attached to any of the camera devices, or stand-alone audio input devices), digital cameras, devices that upload slide shows with voice-over illustrations, files previously encoded in a client-chosen format, or files available via a network accessible mount point (such as, but not limited to, Hypertext Transfer Protocol (HTTP), File Transfer Protocol (FTP), or remote servers). These examples are not limiting, and one skilled in the art will appreciate that a wide variety of client devices and programs can be used to publish and/or transmit media content via the network 201, and that the invention is directed to such client devices and programs.
An origin server is capable of publishing and/or transmitting various types of media content. For example, the origin server 217 can provide multimedia files in various well-known encoded formats including, but not limited to, MPEG, AVI, MP3, REAL, WINDOWS MEDIA, QUICK TIME, H.263 video coding, and PALM-compatible formats.
The publishing service farm 210 acts as an intermediate between the origin servers and the user client 215. As will be described in more detail below, the publishing service farm 210 receives requests for media content from the user client 215 and obtains the requested media content from an origin server. If necessary, the publishing service farm 210 then transcodes the media content received from the origin server from a source type to a destination type that can be accommodated by the user client 215, and delivers the transcoded media content to the user client 215 or other destination client. The publishing service farm 210 performs the transcoding and/or delivery of the requested media content in a manner that is transparent to the content provider as well as the viewer of the media content.
In accordance with the present invention, because the publishing service farm 210 can distribute media content in a variety of formats, i.e., destination types, the origin server can provide media content using a single media input device and still deliver the content to viewers using a variety of media players, each of which requires a different destination type. Additionally, the present invention permits users to access a variety of media content published in different source types no matter what media player they are using.
II. Process
A. On-demand and Cache Processing
One process for handling a client's publishing service request is illustrated in
Note that here, and in all other situations in which media content is distributed using this invention, the media content may be sent to a client other than that of the end user that originated the publishing service request. The receiving client will be specified in the publishing service request.
If, in step 515, the determination is made that the requested media is not available in cache memory, then processing continues at step 530. In this step, the publishing service request processor determines whether there is transcoding capacity available in the media provider farm. If not, then the end user is informed that the system is too busy to process the request. If, however, capacity is available, then processing continues at step 532. Here, the publishing service request processor generates and sends an on-demand media provider request to the media provider request processor. Referring to
Note that in processing any given media provider request, the availability of transcoding servers must be determined accurately and unambiguously. If, for example, two media provider requests are being processed more or less simultaneously, there is a risk that each request would be unaware of the prospective processing load implied by the other. This could result in a resource contention problem, where each media provider request could lay claim to the same transcoding server. To prevent conditions such as these, database access must be controlled. In one embodiment of the invention, a media provider request is granted exclusive access to the database for purposes of determining server load (step 540) while other pending media provider requests wait. When the task and load associated with the first request have been assigned to a server, then the processing of another media provider request can begin by determining the current server load, etc. This and other mechanisms for preventing resource contention and database coherency problems are known to persons of skill in the art.
In step 550, the selected transcoding server transcodes source media content to obtain the media content that was requested by the client's publishing service request. In step 552, the resulting transcoded media content is sent to the selected distribution server. Note that in an embodiment of the invention, the transcoded media content is also sent to cache memory, where it is stored in the event that subsequent publishing service requests seek the same transcoded media content. In step 553, the database is updated to reflect completion of transcoding, and to record historical data relating to the transcode, such as processing load statistics, and source, destination, and transcoding server properties. This data can be used subsequently in estimating the load that will be required for future, similar tasks. In step 554, a distribution request is sent to the selected distribution server. In step 560, the transcoded media content is delivered from the selected distribution server to the destination client.
Note that steps 550 through 560 are shown in
B. Pass Through Processing
In the event that the publishing service request processor decides that the publishing service request should be handled in a pass-through manner, the resultant processing takes place as illustrated in
C. Batch Processing
As described briefly above, the invention can also perform batch transcoding independent of any particular client or publishing service request. The process for batch transcoding is illustrated in
D. Determining Type Of Media Provider Request
The process of determining what kind of media provider request to generate (and, therefore, what kind of request processing to perform) is illustrated in
Steps 920 and 930, the steps of determining the optimum publishing option and generating the corresponding media provider request, respectively, are illustrated in greater detail in
Option 1040 represents publication based on an on-demand transcode operation wherein the transcoding matches the constraints identified in the publishing service request. Option 1050 also represents publication based on a transcode operation wherein the transcoding matches the constraints identified in the publishing service request. In option 1050, however, the transcode task has a certain priority that is then compared with the priorities of ongoing transcoding tasks. Option 1060 represents publication based on a transcode operation wherein the constraints identified in the publishing service request are varied, so that the originally identified constraints are not matched exactly in transcoding. Option 1065 also represents publication based on a transcode operation wherein the originally identified constraints are not matched exactly in transcoding. Like option 1050, however, option 1065 includes assessment of the priority of the transcoding task relative to other ongoing tasks. Option 1070 represents publication based on a pass through operation. In step 1080, the ideal option is selected.
With respect to step 930, generation of the appropriate media provider request is performed in step 1090. The generated request will be one of requests 1091 through 1099. Media provider request 1091 specifies cache access and corresponds to publishing option 1020.
Media provider request 1092 specifies cache access wherein the accessed media content represents a near match of the contents requested in the publication service request. Request 1092 corresponds to publication option 1030. When all attempts at scheduling new transcoding tasks have been exhausted it is possible, in this case, to respond to the client request with a piece of cached transcoded media that closely, but not exactly, matches the original request. Most users would rather see their request serviced with slightly modified constraints than not at all.
Here, existing cached transcoded media is sorted based on weight, where their weight is a function of their associated request constraint's variance from the client's original request constraints. More weight is given to constraints of greater importance, such as the source media content, player type, and player version. Less weight is given to constraints of less importance such as bitrate, height, width, etc. The cache entry of least weight variance is returned to the client. If the minimum variance is sufficiently large (e.g. no transcode of a particular source media within an acceptable variance could be found in the cache), option 1092 will not be chosen.
There may be some circumstances under which previously transcoded media content will be chosen, where the chosen media content does not exactly match the client's constraints, even when there is enough spare capacity to launch a new transcoding task that exactly matches the client's constraints. One example where this may occur is if previously transcoded media exists that has been transcoded in non-real-time as in the case of speculative transcoding operations; its quality may be so much higher than the real-time transcode that could be produced with an exact constraint match, that the previously transcoded media will provide an overall superior experience for the end user.
Returning to
Killing a task is based on priority. Each task is responsible for periodically updating its own entry in a table of tasks and respective priorities, providing both load and priority information. Task priority is determined by several factors including number of viewers (since content in high demand makes for a more urgent task) and contracted publisher uptime (since an obligation to publish also makes a task more urgent). Transcoding servers handling transcoding tasks actively modify a task's priority when the number of viewers changes. In the simplest case, a transcoding task whose viewer count has fallen to zero lowers its priority. If a new user joins, the server handling a task raises its priority.
Determination of which task(s) to kill is accomplished by iterating through the servers of the needed type until a server is found containing a set of lower priority tasks whose total load is less than the load estimate of the new task to be initiated. Once a server with a set of killable tasks has been identified, the media provider request processor sends kill message(s) to the identified killable tasks, then allocates the new task to the selected server. Note that the media provider request processor can kill tasks on transcoding and/or distribution servers, as needed.
In one example not intended to limit the scope of the invention, a transcoding task A is spawned in response to a request A. Initially it is assigned a priority of 1 because it has a single viewer. If, for example, the viewer disconnects, then the priority is lowered to zero to indicate lack of viewers. As task A proceeds, it reports an associated load of 50 units. Given a new task B with an estimated load of 40, the media provider request processor might then search for a transcoding server with a set of tasks of lower priority than B's priority, whose reported load is greater than or equal to 40. In this case, the media provider request processor will find that server A and task A satisfy its search. The media provider request processor reserves task B's estimated load on the server, then issues a kill command to task A, followed by initiation of task B at server A.
Media provider request 1096 specifies on-demand transcoding, wherein the constraints identified in the publishing service request can be varied. Request 1096 corresponds to publication option 1060. Such a media provider request may be appropriate, for example, when matching a requested format exactly is less important than a timely response. If, for example, none of the other options are feasible, then option 1096 could be the only alternative that satisfies the request. In the case of option 1096, the request is fulfilled in a slightly different format that what was requested. Alternatively, media content of a quality higher than what was requested may be provided. Again, matching the requested format exactly would be a secondary consideration. Note that by providing option 1096, system efficiency is enhanced, since the number of unique sets of constraints serviced by the system is reduced. This can allow the use of fewer transcoding servers by increasing the probability that a request can be handled by a cache.
In one example not intended to limit the scope of the invention, a transcoded task (with a requested bit rate of 300 k) has an estimated load of 80 units, but at most only 60 units of load are available on any particular transcoding server. In this case the media provider request processor could conform the estimate load to the available load by reducing the requested output bit rate from 300 K to 128 K. The media provider request processor next determines if this optimization is within the acceptable variance of the original request. In this case, reducing the output bit rate may not significantly reduce the user experience and, if so, is judged acceptable. Next, the task is initiated at the appropriate server.
Returning to
Finally, returning to
E. Estimating Task Load
As discussed above, the media provider request processor can comprise a task load monitor that estimates the processing load imposed by a given transcoding task. This estimation is necessary when determining which transcoding server will be assigned the task. The process for estimating the load associated with a transcode task is illustrated in
In step 1110, information is collected pertaining to the nature of the required transcode task. In particular, this information includes properties of the source and destination formats. This represents consideration of format of the source media content, and the format of the requested media content. Step 110 also includes determination of transcoding server properties. This is necessary because the processing load created by a transcoding task depends in part on the properties of a prospective transcoding server.
In step 1120, a search is performed over historical transcode load data, in an attempt to find the closest match for the transcode task at hand with respect to source, destination, and server properties. In step 1130, the difference between the transcoding task at hand and the closest match is determined. If, in step 1140, the match is sufficiently close, then in step 1150, it is assumed that the load of the task at hand is approximately that of the closest historical match. The load of the historically matching task is therefore returned in step 1150.
If, in step 1140, the closest match is not sufficiently close to the task at hand, then in step 1160, a conservative, fixed, predetermined load estimate is returned.
In either case, the result of process 100 is an estimate of the load that would be imposed by the task on a given transcoding server. After this process is performed with respect to all possible transcoding servers, and after current server loads are determined (see Section III. F. below), a decision can be reached as to the server to which the task should be assigned.
F. Estimating Server Load
The media provider request processor can also comprise a server load monitor for determining the current processing load on any given transcoding server. The process for this determination is illustrated in
G. Speculative Batch Processing
Also as described above, media provider request scheduler can include a speculative transcode manager. The process of speculatively issuing a batch media provider request is illustrated in
In one example not intended to limit the scope of this invention, the speculative transcoding manager would re-encode a previously real-time-transcoded content at slower than real-time in order to generate a higher quality output. Techniques employed to generate higher quality output could also include allowing more time for decompression and compression, filtering of the decompressed video and audio signals (e.g., noise reduction), and multipass encoding, to name a few.
In another example, the speculative transcoding manager would examine usage patterns for a particular piece of media content, and find that users had made numerous Real media requests at bit rates of 28.8 k, 128 k, 300 k, but not at 56 k. In this case the lack of a 56 k transcode would be seen as anomalous and, in reaction, the speculative transcoding manager would spawn a request for the content at the 56 k bitrate.
In another example not intended to limit the scope of the invention, the speculative transcoding manager would pre-transcode content based on a publisher's request. If for example, CNN produced daily content, CNN could specify that their content is to be pre-transcoded in real time or slower than real time into Windows Media, Real, and QuickTime formats at 56 k, 128 k, and 300 k. In doing this, CNN could guarantee the availability of their content at the moment they published it externally.
H. Zero-Downtime Reassignment of Servers to Different Tasks, Fault Tolerance, and Zero-Downtime Server Retirement
A distributed on-demand media transcoding system according to the present invention can include pools of servers with different capabilities (e.g., transcoding servers, distribution servers serving media content in Real format, distribution servers serving media content in a Windows Media format, distribution servers serving media content in a Quicktime format, distribution servers serving media content in a gMovie format, etc.). In many implementations, these differences in capability are strictly due to installed software. So it is practical to reconfigure a server to perform a different task via automated software that does not require operator intervention.
One variable, when optimizing the quality of service and efficiency of a distributed on-demand media transcoding system, is the allocation of farm servers to these varying tasks. Being able to reallocate available servers on the fly to different tasks allows a media provider request processor to adapt to changing usage patterns while maintaining the best quality of service and efficiency. However, this is difficult to do without interrupting service. A related problem is that of fault tolerance and upgrade. If a server goes down unexpectedly, this must not interrupt service in any other part of the farm. If a server needs to be retired so that it can be upgraded, this too must not interrupt service.
In one embodiment, illustrated in
If the registration information for a server is current, then the server is presumably available, as indicated in step 1410. In step 1415, the media provider request processor then chooses a server from the remaining pool and performs application-level checks to verify that the chosen server is in fact usable. The media provider request processor repeats this process until it finds a server it can use to satisfy the user's request.
In an embodiment of the invention, each server supports a “shutdown” primitive which will cause it to remove its own entries from the database, thus preventing the media provider request processor from scheduling any new tasks on that server. Then, after all currently running tasks have completed, the server can be taken down without client-visible interruption of service. This process is illustrated in
IV. Data
As described above, the media transcoding system includes one or more transcoding engines that convert certain types of media content (referred to herein as a source type) to another type of media content (referred to herein as a destination type). Transcoding can involve a number of different conversion operations. The particular conversion operations used depend upon the media content and associated publishing variables being converted. “Publishing variables” as used herein refers to different characteristics of media content.
According to the present invention, media content is digital data being published over a network. In this case, publication refers to digital data which has been formatted for delivery over a network and for viewing by a destination media player. Publishing variables for media content can include, but are not limited to, the file format, bit rate, communication protocol(s), physical medium, compression algorithm, and/or digital rights management information.
The digital data can be any type of file format including but not limited to container formats, bitmap formats, video formats, audio formats, vector formats, metafile formats, scene formats, animation formats, multimedia formats, hybrid formats, hypertext and hypermedia formats, three-dimensional data (3D) formats, virtual reality modeling language (VRML) formats, font formats (bitmap fonts, stroke fonts, spline-based outline fonts), page description language (PDL) formats, and any other type of graphics file format or other file format. Table 1 lists examples of such file formats that can be used in embodiments of the present invention:
See, Murray and vanRyper, pp. 12-26. These examples are illustrative and not intended to necessarily limit the present invention. Other file formats (now known or developed in the future) can be used as would be apparent to a person skilled in the art given this description.
Even within the same file format, digital data can be compressed according to different compression algorithms. In a QUICK TIME formatted file, for example, video can be compressed in accordance with H.263, CINEPAK, JPEG, QT ANIMATION, or QT VIDEO standards. As a further example, in a WIDOWS MEDIA ASF formatted file, audio can be compressed in accordance with the MICROSOFT AUDIO FORMAT, ACELP, VOXWARE, or MP3 standards. Compression algorithm choices can be made based on optimization according to bit-rate choices, or according to the nature of the content. For example, video files in which little motion occurs (“talking heads”) and video files in which there is a substantial amount of motion (“high-motion” video) may each be more efficiently compressed using different compression algorithms.
Within any one compression algorithm, there can be further variations. For example, files compressed according to the JPEG standard can be either YUB-based or RGB-based,
In addition to the publishing variables set forth above, there are also publishing variables unique to video data and audio data. Publishing variables for video data include the width and height of the video image in pixels as well as the frame rate of the video. Depending on the bit-rate requirements and the nature of the data, different settings may be necessary in order to ensure the best picture quality. For example, some video may be better viewed at 15 frames per second at 160×120 pixels, while some others may be better viewed at 5 frames per second at 320×240 pixels, even at the same bit-rate. Where the bit-rate is 56 K bps, picture quality becomes very limited, and it is almost never optimal to deliver video in 640×480 pixel resolution. Yet another publishing variable for video data is the number of bits per component.
Publishing variables for audio data include the number of samples per second, the number of channels (e.g., mono, stereo, 5-channel) and the sample size (8-bit, 16-bit, etc.). Different settings may be necessary to ensure audio quality in light of a particular content type and bit-rate.
Publishing variables may also include the size of data packets being sent and the choice of transmission protocol (e.g., TCP vs. UDP).
Transcoding can therefore be viewed as conversion of data having a source type (defined by a set of publishing variables) to data having a destination type (defined by a second set of publishing variables). Examples are shown in the following tables:
These examples are illustrative and not intended to limit the present invention. Other types of on-demand transcoding operations that are known now or developed in the future can be used as would be apparent to a person skilled in the art given this description.
V. Conclusion
The foregoing description of the specific embodiments will so fully reveal the general nature of the invention that others can, by applying knowledge within the skill of the art (including the contents of the references cited herein), readily modify and/or adapt for various applications such specific embodiments, without undue experimentation, without departing from the general concept of the present invention. Therefore, such adaptations and modifications are intended to be within the meaning and range of equivalents of the disclosed embodiments, based on the teaching and guidance presented herein. It is to be understood that the phraseology or terminology herein is for the purpose of description and not of limitation, such that the terminology or phraseology of the present specification is to be interpreted by the skilled artisan in light of the teachings and guidance presented herein, in combination with the knowledge of one of ordinary skill in the art.
This application is a Division of U.S. application Ser. No. 10/076,090, filed Feb. 15, 2002, now U.S. Pat. No. 7,155,475, which is related to U.S. application Ser. No. 09/742,294, filed Dec. 22, 2000, and to U.S. application Ser. No. 09/986,683, filed Nov. 9, 2001 which are all incorporated by reference in their entirety herein.
| Number | Name | Date | Kind |
|---|---|---|---|
| 3394352 | Wernikoff et al. | Jul 1968 | A |
| 3913093 | De Vincentiis et al. | Oct 1975 | A |
| 5526397 | Lohman | Jun 1996 | A |
| 5541852 | Eyuboglu et al. | Jul 1996 | A |
| 5657015 | Nakajima et al. | Aug 1997 | A |
| 5796829 | Newby et al. | Aug 1998 | A |
| 5818933 | Kambe et al. | Oct 1998 | A |
| 5838927 | Gillon et al. | Nov 1998 | A |
| 5848134 | Sekiguchi et al. | Dec 1998 | A |
| 5949876 | Ginter et al. | Sep 1999 | A |
| 5991795 | Howard et al. | Nov 1999 | A |
| 6032172 | Kutcher | Feb 2000 | A |
| 6061732 | Korst et al. | May 2000 | A |
| 6122290 | Kawamata | Sep 2000 | A |
| 6247050 | Tso et al. | Jun 2001 | B1 |
| 6751673 | Shaw | Jun 2004 | B2 |
| 20020032777 | Kawata et al. | Mar 2002 | A1 |
| 20020059170 | Vange | May 2002 | A1 |
| 20020069420 | Russell et al. | Jun 2002 | A1 |
| 20020116293 | Lao et al. | Aug 2002 | A1 |
| 20020184159 | Tadayon et al. | Dec 2002 | A1 |
| 20030014630 | Spencer et al. | Jan 2003 | A1 |
| 20030110234 | Egli et al. | Jun 2003 | A1 |
| 20030112792 | Cranor et al. | Jun 2003 | A1 |
| Number | Date | Country |
|---|---|---|
| 1 026 872 | Aug 2000 | EP |
| Number | Date | Country | |
|---|---|---|---|
| 20060294183 A1 | Dec 2006 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10076090 | Feb 2002 | US |
| Child | 11510911 | US |
