ES2354429T3 - SYNCHRONIZED DATA TRANSFER SYSTEM. - Google Patents

Abstract

Synchronized data transfer system (100) comprising: a series of processor nodes (103); a network core switch (101) coupled to said series of processor nodes to enable communication between said series of processor nodes; a series of storage devices (111) distributed on said series of processor nodes and storing a series of titles, each title being divided into a series of sub-fragments (113) that are distributed on said series of storage devices; a series of transfer processes (215), each executed in a corresponding processor node of a series of processor nodes and being operative to send a message (MSG) to a synchronous switch management process (219) to each sub-fragment that has to be transmitted from a local storage device of a source processor node (203) to a destination processor node, each message including a source node identifier (SRC) that identifies one of a series of processor nodes as said source processor node and a destination node identifier (DST) that identifies one of said series of processor nodes as said destination processor node; and said synchronous switching management process, executed in at least one of said series of processor nodes (205), which periodically sends a transmission order (TX CMD) to said series of processor nodes to initiate each of the periods of a series of sequential transmission periods that a series of messages (MSG) receives, and before each transmission period, which selects from said series of messages to ensure that each process node sends to a sub-fragment and receives up to a sub-fragment during a following period, and that sends a series of transmission requests (TXR) corresponding to selected messages; and wherein each transfer process that has sent at least one message and that has received a transmission request (TXR) of said synchronous switching management process, identifying a corresponding sub-fragment, sends said corresponding sub-fragment (SC) during the following period of transmission initiated by a transmission order sent.

Description

REFERENCE TO RELATED PATENT APPLICATIONS

The present application claims the benefit of the provisional US patent application No. 60 / 526,437 filed on 02/12/2003 and is a continuation in part of the US patent application entitled "Interactive 5 Broadband Server System" ("Broadband Server System Interactive ”) serial number 10 / 304,378 filed on 11/26/2002, pending processing, which in turn claims the benefit of provisional US patent application No. 60 / 333,856 filed on 11/28/2001, all of which have a common inventor with the same holder by assignment.

BACKGROUND OF THE INVENTION 10

TECHNICAL SECTOR OF THE INVENTION

The present invention relates to interactive broadband server systems and, more particularly, to a device with interactive content that uses a synchronized data transfer system, to facilitate simultaneous delivery of multiple isochronous high-speed data streams.

DESCRIPTION OF RELATED TECHNIQUES 15

It is desirable to provide a solution to the storage and supply of media flow content. An initial objective for scalability is 100 to 1,000,000 individual simultaneous streams with isochronous content at 4 megabits per second (Mbps) per stream, although different data rates are expected. The total available bandwidth is limited by the greater available switching of the available backplane. The biggest commutes are currently in the range of 20 terabits per second, that is, approximately 200,000 simultaneous output streams. The number of output streams is generally inversely proportional to the bit rate per stream.

The simplest model of content storage is a single disk connected to a single processor that has a single network connector. The data is read from the disk, put in memory and distributed in packets through a network to each user. Traditional data, such as 25 Web pages or the like, can be provided asynchronously. In other words, there are random amounts of data with random time delays. A low volume and low resolution video can be supplied from a web server. Real-time media content, such as video and audio, requires isochronous transmission or transmission with guaranteed supply times. In this situation, there is bandwidth limitation on the disk drive. The disc has rotational arm movement and latency 30 that must be counted on. If the system can support only 6 simultaneous streams of continuous content from the disk to the processor at any time, then a user's 7th request must wait for one of the other 6 previous users to yield a content stream. The advantage of this design is its simplicity. The disadvantage is the disk that, as the only mechanical device in the design, can only access and transfer data with this speed. 35

An improvement can be achieved by adding another disk or disk drive, and interleaving disk accesses. Also, duplicate content can be stored on each disk gaining redundancy and performance. This is a better solution, but there are still several problems. Only a certain content can be placed on the disk or local disks. The disk units, CPU and memory are, each of them, individual points of failure that can be catastrophic. System 40 can only be scaled to the number of units that the disk controller can manipulate. Even with many units there is the problem of the distribution of titles. In the real situation, everyone wants to see the latest movies. As a usual rule, 80% of content requests go to only 20% of the titles. All the bandwidth of a device cannot be consumed by a title, since it would block access to less popular titles stored only on this equipment. As a result, the "high demand" titles would have to be loaded on most of the equipment. In a nutshell, if a user wanted to watch an old movie, this user could have difficulties, even if the movie was loaded into the system. With a large library, the proportion can be much higher than the 80/20 standard used in this example.

If the system were based on the standard Local Area Network (LAN) used in the data processing, 50 other inefficiencies would arise. TCP / IP systems based on Ethernet modem are a marvel of guaranteed supply, but include a time penalty caused by packet collisions and retransmissions of partially lost packets and the management necessary to make it work. There is no guarantee that a set of content streams will be available at the right time. Also, each user consumes a switching port and each content server consumes a switching port. Therefore, the number of switching ports has to be twice the server number, limiting the total bandwidth online.

The basic architecture shown and described in US 6,134,596 has two limiting characteristics.

First, the network switch is the main point for the reconstruction of data segments provided by the data servers coupled to it. This configuration requires significant additional processor capacity in the network switch for reconstruction and limits total production. Second, the network switch ports must be shared between the data servers and the devices attached to the network server to provide content to the clients.

US 5,367,520 discloses a means for reducing the internal complexity of an Asynchronous Transfer Mode (ATM) increasing its reliability through modularity and internal redundant routes. It uses both input and output buffers and deals with a network that already works asynchronously. This approach is not useful or relevant for a system, according to the present invention, particularly for isochronous information. An autonomous buffer input switch cannot be used reliably to recombine data in the correct order for a given output when the data stream appears in sequence in multiple inputs. The internal switching content on individual routes may cause the data of input 1 to appear on the designated output after 15 that the data of input 2 has begun to appear. For an isochronous video, this is not favorable since packets They must be sent in the correct order to produce a consistent image.

Document US 6,415,328 describes a system for recovering data on a video server in which the process is centralized in a controller and programmer, and the data is stored within a centralized storage medium. This system is specific to a video system in which each data stream has an identical maximum data rate (data stream bandwidth).

Since the system known from US 6,415,328 requires redundancy for load balancing, there are likely situations in which the failure of a disk will result in the failure to deliver a title to a user. In addition, the magnitude of the redundancy required for a title depends on the popularity of the title, acquiring extensive management. The system known from US 6,415,328 does not provide for scaling the system beyond the bandwidth accepted by the bus that connects the storage devices or the bus that connects the output buffers. The flow capacity of the known system is limited by the maximum bus capacities and the known system achieves maximum use of the bus by storing redundant copies of the content. 30

The present invention has been defined in the independent claim of system 1 and the independent claim of method 12.

BRIEF DESCRIPTION OF THE DRAWINGS

The advantages, features and benefits of the present invention will be better understood with respect to the following description and accompanying drawings, in which:

Figure 1 shows a simplified block diagram of a part of an Interactive Content Device (ICE) implemented in accordance with an exemplary embodiment of the present invention; Y

Figure 2 is a logical block diagram of a part of the ICE of Figure 1, showing a synchronized data transfer system implemented according to an embodiment of the present invention.

DETAILED DESCRIPTION

The following description aims to enable an ordinary person skilled in the art to carry out and use the present invention, as disclosed within the context of a particular application and its requirements. However, different modifications of the preferred embodiment will be apparent to those skilled in the art and the general principles defined may be applied to other embodiments. Therefore, the present invention is not intended to be limited to the particular embodiments that have been shown and described, but should receive the broadest scope consistent with the principles and new features that are disclosed.

The architecture described accepts individual components of different capacity to prevent the installation from being limited to the time at which the initial system was acquired. The use of trade components guarantees well-proven recent technology, avoiding unique sources of supply and lower costs for data flow or delivery. Individual component failures are tolerable. In many cases there are no sensitive behavioral changes from the user's perspective. In other cases there is a brief cycle of "self-repair." In many cases 55 multiple faults can be tolerated. Also, in many cases but in the whole, the system can

recover without requiring immediate attention, making it ideal for "unattended" operation.

Content storage allocation and internal bandwidth are automatically controlled by algorithms of the Lesser Recent Use (LRU) type, which ensures that the content in RAM and the hard disk device are appropriate for current demand and that The 5-band width of the backplane switch is used most effectively. The bandwidth within the system is hardly, or never, overused or excessively required, so it is not necessary to discard or delay packet transmission. The architecture provides the ability to take full advantage of the composite bandwidth of each component, so that guarantees can be fulfilled, and the network is private and under control, so that even in a peak demand situation In advance there is no overhead of data paths. Streams can be tolerated with any bit rate, but typical streams can be expected to remain within the range of 1 to 20 Mbps. Asynchronous content is tolerated based on the available bandwidth. Bandwidth can be reserved for this purpose if necessary for the application. Files can have any dimension with a minimum of inefficiency of storage. fifteen

Figure 1 is a simplified block diagram of a part of an Interactive Content Device (ICE) (100) implemented in accordance with an exemplary embodiment of the present invention. Parts not applicable for a complete understanding of the present invention have not been shown for clarity. The ICE (100) comprises an appropriate Gigabit Ethernet (GbE) multiport switch (101) as a backplane fabric having multiple ethernet ports coupled to a series of 20 Storage Processor Nodes (SPN) (103). Each SPN (103) is a simplified server that includes two Gigabit Ethernet ports, one or more processors (107), a memory (109) (for example, a random access memory (RAM)) and an appropriate number (for example four to eight) disk drives (111). A first Gb port (105) of each SPN (103) is connected to a corresponding switch port (101) for full duplex operation (simultaneous transmission and reception at every SPN connection / port) and is used to shift data within the ICE 100. The other Gb port (not shown) sends the content output to users below (not shown).

Each SPN (103) has high speed access to its local disk drives and the other disk drives of the other four SPNs in each group of five SPNs. The switch (101) is a "backplane" for the ICE (100) instead of just a communication device between the SPNs (103). Only five SPNs (103) have been shown for the purpose of illustration, it being understood that the ICE (100) typically comprises a larger number of servers. Each of the SPN (103) acts as storage, process and content transmitter. In the configuration shown, each SPN (103) is configured using commercial components and is not a computer in the usual sense. While standard operating systems are provided, such operating systems operated with interruption may present unnecessary bottlenecks.

Each title (for example, video, movie or other media content) is not completely stored on any single disk drive (111). Instead, the data for each title is divided and stored among several disk units within the ICE (100) to achieve the advantages of interleaved access speed. The content of a single title is extended in multiple 40 disk units of multiple SPNs (103). Short “time frames” of carousel-shaped title content from each disk drive in each SPN (103) meet. In this way the physical load is extended by saving the limits of the number of SCSI and IDE disk drives, a safe operation mode is achieved, and a large number of titles are organized and controlled.

In the specific configuration shown, each content title is divided into 45 fragments of a fixed size (typically about 2 megabytes (MB) per fragment). Each fragment is stored in a different set of SPN (103) in carousel arrangement. Each fragment is divided into four subfragments and five subfragments represent that parity has been created. Each subfragment is stored in a disk drive of different SPN (103). In the configuration shown and described, the subfragment dimension of approximately 512 kilobytes (KB) (where "K" is 50 1024) corresponds to the nominal data unit of each of the disk drives (111). SPNs (103) are arranged in groups of five and each group or set of SPN stores a piece of data from a title. As shown, the five SPN (103) are marked 1-4 and "Parity", which collectively store a fragment (113) as five separate sub-fragments (113a, 113b, 113c, 113d and 113e) stored in the SPN 1 , 2, 3, 4 and Parity, respectively. The sub-fragments (113a-55 113e) have been shown distributed in a different way for each different SPN (for example, SPN1 / DRIVE1, SPN2 / DRIVE2, SPN3 / DRIVE3, etc.), but can be stored with any other possible combination (for example, SPN1 / DRIVE1, SPN2 / DRIVE1, SPN3 / DRIVE3, etc.). Subfragments 1-4 comprise the data and the parity subfragment comprises the parity information for the data subfragments. The dimensions of each SPN set, although typically five, are 60 arbitrary and could also be any other number, for example, 2 SPN to 10 SPN. Two SPNs would use 50% of their capacity for redundancy and ten would use 10%. Five is a compromise between storage efficiency and the probability of failure.

By distributing the content in this way, at least two objectives are achieved. First, the number of users who can view a given title is not limited to the number that can be served by a single set of SPNs, but by the bandwidth of all SPN sets considered together. Therefore, only one copy of each content title is required. The commitment is to limit the number of new viewers for a given title 5 that can be launched every second, which is much less limiting than the wasted space load and redundant storage management. A second objective is to increase the overall reliability of the ICE (100). The failure of a single disk is masked by the real-time regeneration of its contents, using the Parity disk, similar to a redundant set of independent disks (RAID). The failure of an SPN (103) is masked by the fact that it contains a disk of each of several RAID sets, each of which continues to function. Users connected to a failed SPN are recovered very quickly by shadow processes that work in other SPNs. In the event of a disk drive or SPN assembly failure, the operator receives notification to repair or replace the equipment that has failed. When a subfragment that fails is rebuilt by the user's process, it is transmitted in return to the SPN that has arranged it, where it is kept in a RAM (as it would have been if it had been read from the local disk). This avoids wasting time from other user processes when performing the same reconstruction of a popular title, by completing subsequent requests from RAM, provided that the subfragment is popular enough to remain saved or reserved.

The objective of a user process (UP) operating in each user SPN (103) is to gather the subfragments of its own disk plus the four corresponding subfragments of another user SPN to gather a fragment of video content for its Shipping. User SPNs are distinguished from one or more MGMT management SPNs, which are configured in the same way but perform different functions, as will be described later. A pair of redundant MGMT SPNs is expected to increase reliability and performance. The meeting and coupling functions 25 carried out by each UP are often carried out taking into account many users in each user SPN (103). As a consequence, there is a significant amount of data traffic between the user SPN (103). The typical ethernet protocol, with packet collision detection and new attempts, would otherwise be overcome. Typical protocols are designed for random transmissions and depend on the time intervals between such events. Therefore, this system is not used. In 30 ICE (100) collisions are avoided by using a complete duplex, completely switched architecture and managing bandwidth carefully. Most of the communication is done synchronously. The switch 101 itself is controlled in a synchronized manner, as will be described later, so that the transmissions are coordinated. Since it is determined which of the SPNs (103) happens to transmit and when it does, the ports are not exceeded by more data than they can manipulate for a certain period of time. Certainly, the data is first collected in the memory (109) of the user SPN (103) and then its transfer is controlled synchronously. As part of the arrangement there are situation signals between user SPNs (103). Unlike the actual content that passes to the end users, the data dimensions for signaling between the user SPN units are very small. 40

The length of each sub-fragment (approximately 512K, being "K" 1024) would otherwise exceed any buffer effect available on the GbE switch (101) if transmission of sub-fragments was allowed randomly or asynchronously. The period to transmit this amount of information is about 4 milliseconds (ms) and it is desirable to ensure that several ports do not attempt to transmit and do not transmit to a single port simultaneously. Therefore, as described below, the switch (101) is managed in a manner that causes its asynchronous operation, with all ports fully used under full load conditions.

The redundant directory or directory process that controls the file system (or virtual or VFS file system) is responsible for informing where a particular content title is stored when requested by a user. It is also responsible for allocating the 50 storage space required when a new title has to be loaded. All attributions take place in integral fragments, each of which is composed of five subfragments. The space on each disk drive is controlled within the drive by a Logical Block Address (LBA). A subfragment is stored in a disk drive in contiguous sectors or LBA addresses. The capacity of each disk drive in the ICE (100) is represented by its maximum 55 LBA addresses divided by the maximum of sectors per sub-fragment.

Each title map or “directory inscription contains a list indicating where the fragments of its title are stored and, more specifically, where each sub-fragment of each fragment is located. In the embodiment shown, each item in the list representing a sub-fragment contains an SPNID that identifies the SPN (103) of a specific user, a 60 disk unit number (DD #) identifying a specific disk (111) of the identified user SNP (103) and a sub-fragment pointer (or Logical Block Address or LBA) packaged as a 64-bit value. Each directory entry contains a list of sub-fragments of approximately

half an hour of content for the nominal 4 Mbps. This is equal to 450 fragments or 2250 subfragments. Each directory record has approximately 20 KB of auxiliary data. When a UP running in an SPN requests a directory entry, the entire entry is sent and stored locally for the corresponding user. Even if an SPN supports 1000 users, only 20 MB of memory is consumed for local lists or directory entries. 5

The ICE (100) maintains a database of all the titles available to a user. The list includes the library of the local optical said, real-time network programming and titles in remote locations, with which license and transport agreements have been made. The database contains all metadata for each title, including management information (license period, bit rate, resolution, etc.), as well as information of interest to the user (producer, director, interpreters, staffing 10 , author, etc.). When the user makes a selection, a directory of a virtual file system (VFS) (209) (figure 2) is asked to determine if the title has already been loaded on the disk device. Otherwise, a loading process (not shown) for said piece of content is initiated and the UP is notified if necessary when it will be available for viewing. In most cases, the latency is not greater than the mechanical latency of the optical disk recovery robot (not shown), that is, about 30 seconds.

The information stored on the optical disc (not shown) includes all metadata (which is read in the database when the disc is first loaded into the library) as well as compressed digital video and audio representing the title and all the information that can be extracted in advance with respect to such data flows. For example, it contains pointers for all relevant information in data flows, such as clock values and time stamps. It is already divided into subfragments with the precalculated parity subfragment and stored on the disk. In general, anything that can be done in advance to save loading time and process structure is included in the optical disk.

A sender (not shown) that the UP consults 25 to receive the start time for the data flow (usually within milliseconds of the request) is included in the resource management system. The shipper ensures that the system load remains regular, that the latency is minimized and that at no time the bandwidth required within the ICE (100) exceeds that available. In the case where a user requests an interruption, pause, fast forward, rewind or other operation that interrupts the flow of their data flow, their bandwidth is deallocated and a new assignment is made for any new requested service ( for example, a fast forward flow).

Figure 2 is a logic block diagram of a part of the ICE (100) showing a synchronized data transfer system (200) implemented in accordance with an embodiment of the present invention. The switch (101) has been shown coupled to several SPN (103) by way of example, including an SPN (201) of a first user, an SPN (203) of a second user and a management SPN (205) 35 ( MGMT). As previously indicated, many of the SPNs (103) are coupled to the switch (101) and only two user SPNs (201, 203) have been shown for illustrative purposes and have been physically implemented the same than any SPN (103) same as described above. The SPN (205) MGMT is physically implemented just like any other SPN (103), but it generally performs management functions instead of user-specific functions. The SPN (201) shows certain functions and the SPN (203) shows other functions of each user SPN (103). However, it will be understood that each user SPN (103) is configured to perform similar functions, so that the functions (and processes) described for the SPN (201) are also arranged in the SPN (203) and vice versa.

As described above, the switch (101) operates at 1 Gbps per port, so that each sub-fragment (approximately 512 KB) requires approximately 4 ms to pass from one SPN to another. Each user SPN (103) executes one or more user processes (UP), each of which supports a user located below. When a new fragment of a title is needed to fill a user exit buffer (not shown), the following five sub-fragments of the list of other user SPNs that store these sub-fragments are requested. Since many UPs require 50 potentially multiple sub-fragments substantially at the same time, the transmission duration of the sub-fragment would otherwise exceed the buffer capacity of almost any GbE switch with a single port and much more for the entire assembly. This is true for the switch (101) shown. If the transmission of subfragments is not controlled, this would result in all five subfragments of each UP being returned simultaneously, exceeding the bandwidth of the output port. It is desirable to adjust the timing of ICE SPN transmissions (100), so that the most critical data is transmitted first and intact.

The SPN (201) has been shown performing a UP (207) for the service of a corresponding user located below. The user asks for a title (for example, a movie), whose request is sent to the UP (207). The UP (207) transmits the title request (TR) to the VFS (209) (described below 60) located in the SPN (205) MGMT. The VFS (209) returns a directory address (DE) to the UP (207), which normally stores the DE shown in (211). The DE (211) comprises a list that locates

each sub-fragment of the title (SC1, SC2, etc.), including each entry the SPNID that identifies a user-specific SPN (103), identifying the disk unit number (DD #) a specific disk unit (111) of the SPN identified (103) and an address or LBA that provides the specific location of the sub-fragment in the identified disk drive. The SPN (201) initiates a time stamp read request (TSRR) for each subfragment in the DE (211), one at a time. In ICE (100) requests are made immediately and directly. In other words, the SPN (201) begins carrying out the subfragment requests immediately and directly to the specific user SPN (103) storing the data. In the configuration shown, the requests are made in the same way, even if they are stored locally. In other words, even if the requested sub-fragment is in a local disk drive in the SPN (201), it sends the request through the switch (201) even if it is located in a remote position. The network is the location that can be configured to recognize that a request is being sent from an SPN to the same SPN. It is simpler to handle all cases the same, especially in large facilities where the request is less likely to be really local.

While requests are sent immediately and indirectly, each sub-fragment is returned in a fully managed manner. Each TSRR is sent to the specific user SPN 15 using the SPNID and includes the DD # and LBA for the target user SPN to retrieve and return the data. The TSRR may also include any other identifying information sufficient to ensure that the requested sub-fragment is properly returned to the appropriate petitioner and to enable the petitioner to identify the sub-fragment (for example, UP identifier to distinguish between many UPs that are executed in the SPN of destination, a subfragment identifier to distinguish between the 20 subfragments for each piece of data, etc). Each TSRR also includes a time stamp (TS) identifying the specific time when the original request was made. The TS identifies the priority of the request for synchronous retransmission purposes, in which the priority is based on time, so that the previous requests have a higher priority. Once received, the subfragments returned from the requested title are stored in a local title memory (213) for further processing and supply to the user who has requested the title.

The user SPN (203) shows the operation of a transfer process (TP) (215) and support functions that are executed in each user SPN (for example, (201, 203)) to receive the TSRRs and to return Subfragments ordered. The TP (215) comprises a storage process (not shown) or is in interface with it, which interconnects the local disk drives 30 (111) in the SPN (203) to request and access the stored sub-fragments. The storage process can be implemented in any desired manner, such as a machine state (state machine) or the like and can be a separate process interconnected between the TP (215) and the local disk drives (111), such as It is known by those skilled in the art. As shown, the TP (215) receives one or more TSRRs from one or more UPs running in the SPN (103) of another user and stores each request in a queue of read requests (RRQ) (217) in your local memory (109). The RRQ (217) stores a list of requests for subfragments SCA, SCB, etc. The disk drive that stores the requested subfragments removes the corresponding RRQ requests (217), classifies them by physical order and then executes each reading in the classified order. Access to subfragments on each disk are managed in groups. Each group is classified by physical order, according to an “elevator search” operation (sweep from bottom to top, next sweep from top to bottom, etc.), so that the disk head sweeps back and forth ahead on the surface of the disk stopping to read the next sequential subfragment). Requests for satisfactory readings are stored in a satisfactory read queue (SRQ) (218) classified in TS order. Read requests that have failed (if any) are stored in a 45-fault read queue (FRQ) (220) and the information that has failed is sent to a network management system (not shown) that determines the error and appropriate corrective action. It should be noted that in the configuration shown the queues (217, 218 and 220) store request information instead of the true sub-fragments.

Each subfragment that is read successfully is placed in memory reserved for a reserve LRU of recently requested subfragments. For each recovered sub-fragment, TP 50 (215) creates a corresponding message (MSG) that includes the TS for the sub-fragment, the source (SRC) of the sub-fragment (for example, the SPNID from which the sub-fragment is being transmitted and its location). physical memory, along with any other identification information) and the destination SPN (DST) to which the sub-fragment has to be transmitted (for example, SPN (201)). As shown, the SRQ (218) includes messages MSGA, MSGB, etc., for sub-fragments SCA, SCB, etc., respectively. After the requested subfragments have been read and reserved, the TP (215) sends corresponding MSGs to a synchronized switch manager (SSM) (219) running on the MGMT SPN (205).

The SSM (219) receives and prioritizes multiple MSGs received from the TP from the user's SPN and eventually sends a transmission request (TXR) to the TP (215) identifying one of the MSGs in its SRQ (218), such as using a message identifier (MSGID) or similar. When the SSM (219) sends 60 a TXR to the TP (215) with an MSGID that identifies a sub-fragment in the SRQ (218), the request list is moved from the SRQ (218) to a network transfer process (NTP) (221) which constitutes the packets used to transfer the subfragment to the SPN of the destination user (designating the term "displace"

the withdrawal of the SRQ request (218)). The order in which the subfragment request lists are removed from the SRQ (218) is not necessarily sequential, although the list is in the order of the time stamp, since only the SSM (219) determines the appropriate order. The SSM (219) sends a TXR to each of the other SPN (103) having at least one sub-fragment to send, except that the sub-fragment has to be sent to a UP in an SPN (103) already programmed to receive 5 a subfragment with an equal or higher priority, as described below. The SSM (219) then sends a single transmission order (TX CMD) to all user SPNs (103). The TP (215) orders the NTP (221) to transmit the subfragment to the UP that has requested it from the SPN (103) of the user in response to the TX CMD command sent by the SSM (219). In this way, each SPN (103) that has received a TXR from the SSM (219) simultaneously transmits to another user SPN (103) that requests it. 10

The VFS (109) in the SPN (205) MGMT manages the list of titles and their locations in the ICE (100). In typical computer systems, directories (data information) usually reside on the same disk on which the data is located. However, in ICE (100), the VFS (209) is centrally located to manage distributed data, since the data of each title is distributed by multiple disks in the disk device, which in turn are distributed by multiple SPN (103) of 15 users. As described above, the disk drives (111) of the user SPNs (103) primarily store the sub-fragments of the titles. The VFS (209) comprises identifiers for the location of each sub-fragment through SPNID, DD # and the LBA, as described above. The VFS (209) also comprises identifiers for other parts of the ICE (100) that are external, such as optical storage. When the user requests a title, it is made available to the 20 UP, which is executed in the user SPN (103) that has accepted the user request, a complete set of directory information (ID / addresses). From there, the task is to transfer the subfragments from the disk drive to memories (buffers), moving them through the switch (101) to the user SPN (103) that asks for it, which gathers a complete fragment in a buffer, supply the user and repeat until termination. 25

The SSM (219) creates a list of “prepared” messages in order of time stamp in a list of prepared messages (RDY MSG) (223). The order in which the messages of the TPs are received in the user SPN (103) is not necessarily in the order of the time stamp, but they are organized in the order TS in the RDY MSG list (223). Just before the next set of transfers, the SSM (219) scans the RDY MSG list (223) starting with the oldest 30 timestamp. The SSM (219) first identifies the oldest TS in the RDY MSG list (223) and generates and sends the corresponding TXR message to the TP (215) of the user SPN (103) storing the corresponding sub-fragment to initiate the pending transfer of said subfragment. The SSM (219) continues to scan the list (223) for each subsequent sub-fragment in TS order generating the TXR messages for each sub-fragment whose source and destination have not already been involved in a pending sub-fragment transfer. For the issuance of each CMD TX to all user SPN (103), each user SPN (103) transmits only one sub-fragment at one time and only receives one sub-fragment at a time, although it can do both simultaneously. For example, if a TXR message is sent to the TP of SPN # 10 to schedule a pending sub-fragment transfer to SPN # 2, then SPN # 10 cannot simultaneously send another sub-fragment. However, SPN # 10 may simultaneously receive a subfragment of another SPN. In addition, SPN # 2 cannot simultaneously receive another sub-fragment while receiving the sub-fragment of SPN # 10, although SPN # 2 can simultaneously transmit to another SPN because of the full duplex nature of each of the switch ports (101 ).

The SSM (219) continues scanning the RDY MSG list (223) until all SPN (103) of 45 user have been taken into account or when the end of the RDY MSG list (223) has been reached. Each of the entries in the RDY MSG list (223) corresponding to a TXR message is eventually removed from the RDY MSG list (223) (when the TXR message is sent or after the transfer is completed). When the last transfer of the previous period has been completed, the SSM (219) sends a TX CMD packet that signals all user SPNs (103) to begin the next 50 round of transmissions. Each transfer takes place synchronously within a period of approximately 4 to 5 ms for the specific configuration shown. During each round of transfer additional MSGs are sent to the SSM (219) and new TXR messages to the user SPN (103) to schedule the next round of transmissions, and the process is repeated. The period between successive TX CMDs is approximately equal to the period necessary to transmit all the bits of a sub-fragment, including packet structure and delay between packets, in addition to a period to eliminate all reservations that may have taken place in the switch during the switch. Subfragment transmission, typically 60 microseconds (μs), plus a period to take into account any oscillation caused by a recognition delay of the CMD TX by an individual SPN, typically less than 100 μs. 60

In one embodiment, a mirror or mirror-symmetric MGMT SPN (not shown) is a mirror image of the primary MGMT SPN (205), so that the SSM (219), the VFS (209) and the shipper are duplicated in a pair of Specialized redundant MGMT SPN. In one embodiment, the TX CMD broadcast

Synchronization acts as a synchronization pulse indicating the "health" of the SPN (205) MGMT. The synchronization pulse is a signal to the secondary MGMT SPN that everything is compliant. In the absence of the synchronization pulse, the secondary MGMT SPN takes over all management functions within a certain period of time, such as, for example, 5 ms.

While the present invention has been described in considerable detail with reference to certain 5 versions thereof, other versions and variations are possible and are contemplated within the invention.

Claims (21)

1. Synchronized data transfer system (100) comprising: a series of processor nodes (103); a network core switch (101) coupled to said series of processor nodes to enable communication between said series of processor nodes; 5 a series of storage devices (111) distributed on said series of processor nodes and storing a series of titles, each title being divided into a series of sub-fragments (113) that are distributed on said series of storage devices; a series of transfer processes (215), each executed on a corresponding processor node of a series of processor nodes and being operative to send a message (MSG) to a synchronous switch management process (219) for each sub-fragment that has to be transmitted from a local storage device of a source processor node (203) to a destination processor node, each message including a source node identifier (SRC) that identifies one of a series of processor nodes as said source processor node and a destination node identifier (DST) that identifies one of said series of processor nodes as said destination processor node; Y said synchronous switching management process, executed in at least one of said series of processor nodes (205), which periodically sends a transmission order (TX CMD) to said series of processor nodes to start each of the periods of a series of sequential transmission periods that receives a series of messages (MSG), and before 20 of each transmission period, which selects from said series of messages to ensure that each process node sends to a sub-fragment and receives up to a sub-fragment during a following period, and that sends a series of transmission requests (TXR) corresponding to selected messages; Y wherein each transfer process that has sent at least one message and that has received a transmission request (TXR) of said synchronous switching management process, identifying a corresponding sub-fragment, sends said corresponding sub-fragment (SC) during the following period of transmission initiated by a transmission order sent. 2. Synchronized data transfer system according to claim 1, wherein each of a series of messages comprises a time stamp (TS) and wherein said synchronous switching management process prioritizes said series of messages based on in the order of the time stamp and sends said plurality of transmission requests in the order of the time stamp. 3. Synchronized data transfer system according to claim 2, further comprising: a series of user processes (207), each executed on a node of a series of processor and operational nodes to send a series of time stamp read requests (TSRR); Y in which each transfer process incorporates a time stamp (TS) of a corresponding read request with time stamp in a corresponding message (MSGA). 4. Synchronized data transfer system according to claim 3, wherein said synchronous switching management process organizes said series of messages in a list of prepared messages (223) in order of time stamp, scanning said list of messages prepared in the order of the time stamp just before each of said series of sequential transmission periods and selects messages based on the priority of the time stamp. 5. Synchronized data transfer system according to claim 4, wherein said synchronous switching management process selects a message if an identified source processor node has not already been selected to transmit a sub-fragment during a subsequent transmission period and if an identified destination processor node has not already been selected to receive a sub-fragment during said next transmission period. 6. Synchronized data transfer system according to claim 1, further comprising: each of said series of transfer processes stores requests for reading subfragments received in a queue for read requests (217), each request for subfragment reading indicating a locally stored subfragment (SCA); each of said series of storage devices reading subfragments identified in a local read request queue by physical order; each of said series of processor nodes makes a list of subfragments read successfully by a corresponding storage device in a read queue successfully (218); and 5 each of said series of transfer processes sends a message (MSG) for each enrollment in a corresponding satisfactory read queue (218) to said synchronous switching management process. 7. Synchronized data transfer system according to claim 6, wherein each of said subfragment read requests comprises a read request with time stamp 10 (TSRR), so that the inscriptions in each of said queues read successfully are listed in the order of the time stamp, and in which each transfer process (215) sends a message for each inscription in a corresponding queue of successful reading in the order of the time stamp. 8. Synchronized data transfer system according to claim 6, further comprising: each of said series of transfer processes (215) removes an inscription from a corresponding satisfactory request queue in which said inscription is associated with a sub-fragment identified by a corresponding transmission request; Y a series of network transfer processes (221), each executed on a corresponding node of a series of processor nodes (103), and each of them being operative to constitute network packets used to transfer an identified sub-fragment to a destination processor node in response to a transmission order. 9. Synchronized data transfer system according to claim 1, wherein said network switch (101) comprises a Gigabit Ethernet switch with a series of ports, and wherein each of said processor nodes is coupled to a corresponding body of said network switch 25. 10. Synchronized data transfer system according to claim 1, wherein said series of processor nodes comprises a management node ((205) executing said synchronous manager process (219). 11. Synchronized data transfer system according to claim 1, wherein said series of processor nodes comprises a first management node (205) executing said synchronous switching manager process and a second management node executing a Symmetric mirror synchronous switching management process. 12.- Procedure for synchronous transfer of distributed sub-fragments of data between a series of processor nodes (103) coupled (105) to a network switch (101), with a series of 35 storage devices (111) distributed on said series of processing nodes and storing a series of titles, each title being divided into a series of sub-fragments (113) that are distributed on said series of storage devices, the procedure of which comprises: periodically issue, by a management process (219) executed in at least one of the processing nodes, a transmission order (TX CMD) to said series of process nodes 40 to start each of said series of sequential transmission periods; send to the management process, for each processor node (203) that has at least one sub-fragment (SC) to send, a message (MSG) for each sub-fragment (SCA) to be sent, from a node (203) of the local storage device to one destination node, each message identifying a node in the series of process nodes as the source processor node (SRC) and one in the series of processor nodes as the destination processor node (DST); select, before each transmission period, by the management process, messages received from the processing nodes to ensure that each processor node that has been identified as the source processor node sends up to a sub-fragment during a subsequent transmission period and that each processor node which has been identified as the destination processor node 50 receives up to a sub-fragment during the following transmission period; send, by the management process, a series of transmission requests (TXR), each transmission request being sent to a process node (203) that has sent a corresponding message that has been selected; Y transmit, for each processor node (203) that receives a transmission request (TXR), an identified sub-fragment (MSGID) by the transmission request received to a destination processor node in response to the next transmission order. 13. Method according to claim 12, further comprising: before said sending of a message for each sub-fragment to be sent, the time stamp 5 (TS) of each message; said selection comprising prioritization based on the order of time stamping (223); Y said sending comprising a series of transmission requests sending transmission requests in the order of time stamping. 10 14. Method according to claim 13, further comprising: the sending, at least by a processor node (201), of a series of read requests with time stamp (TSRR); Y wherein said time stamp of each message comprises the incorporation of a time stamp (TS) of a read request with time stamp received (TSRR) in a corresponding message 15 (MSG1). 15. Method according to claim 14, further comprising: sorting by the management process (219) of the messages received in a list of prepared messages (223) in the order of time stamp; Y scan through the management process (219) the list of prepared messages (223) in order 20 of time stamping just before each transmission period. 16. A method according to claim 15, wherein said scanning comprises the selection of a message if the identified source processor node has not already selected a sub-fragment for transmission during the following transmission period and if the identified destination processor node It has not already been selected to receive a sub-fragment during the next transmission period. 25 17. Method according to claim 16, wherein said scanning is completed when the complete list of prepared messages has been scanned or if all of the processor nodes have been selected to transmit a sub-fragment or if all of the nodes Processor have been selected to receive a subfragment. 18. Method according to claim 12, further comprising: storing subfragment read requests received in a read request queue (217), each subfragment read request indicating a request for a locally stored subfragment (SCA); reading by a local disk drive (111) of subfragments identified in the read request queue in physical order; 35 make the list of subfragment inscriptions read successfully in a successful read queue (218); Y said message sending for each sub-fragment to be sent comprises sending a message for each inscription in said satisfactory reading queue. 19. A method according to claim 18, wherein each subfragment read request 40 comprises a time stamp read request (TSRR), such that said list of subfragment inscriptions read successfully in a satisfactory read queue it comprises the list of inscriptions in order of time stamping, and in which said sending of a message for each inscription in said satisfactory reading queue comprises sending messages in the order of time stamping. 20. Method according to claim 18, further comprising: remove an inscription from the satisfactory request queue that is associated with a sub-fragment identified by a corresponding transmission request (TXR); Y construct (221) network packets (SC) used to transfer the identified sub-fragment to a destination processor node (DST) in response to a transmission order (TX CMD). 21. Method according to claim 12, further comprising the execution by the management processor (219) in a first management node (205) and the execution of a mirror symmetric management process in a symmetric management node a mirror, which is the mirror symmetry of the first management node. 5