`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`162.250.242.2
`
` @ SMPTE All Rights Reserved
` @ SMPTE All Rights Reserved
`
`Moving Uncompressed Video Faster than Real Time
`Don Deel, Marc Friedmann and Howard Green
`
`1996, 105:751-759.
`SMPTE J(cid:160)
`doi: 10.5594/J06423
`
`The online version of this article, along with updated information and services, is
`located on the World Wide Web at:
` http://journal.smpte.org/content/105/12/751
`
`
`
`Oracle Ex. 1030, pg. 1
`
`
`
`TECHNICAL PAPER
`Moving Uncompressed Video Faster Than Real Time
`
`By Don Deel, Marc Friedmann, and Howard Green
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`puters. IOFS is a file access protocol
`that uses Transporter to provide
`enhanced network access of files over
`Fibre Channel. These facilities will be
`discussed later in this paper.
`The Need to Move Data Faster
`With the advent of nonlinear edit-
`ing, video servers, graphics special
`effects, and compositing, computers
`now play an increasingly important
`role in the movie and video produc-
`tion and post-production environment.
`They reside directly in the production
`and revenue flow, being ever more
`broadly used to automate and acceler-
`ate functions such as painting,
`retouching, rotoscoping, color correc-
`tion, titling, image repair, glass paint-
`ing, and wire removal. As computer
`capabilities advance and the number
`of computer-literate artists grows,
`project size has increased and the
`need for broad-based artistic collabo-
`ration has risen rapidly. Rapid and
`responsive communication between
`computer-based artists, editors, and
`producers is becoming a critical factor
`as the ramp-up in digitally processed
`images accelerates.
`
`real-time digital video data transfers.
`Occupying a single bus slot and pro-
`viding either one or two independent,
`fully functional Fibre Channel ports,
`the hardware for the adapter card has
`been optimized for high data through-
`put.
`Specific emphasis is placed on high-
`performance software. Innovative soft-
`ware facilities called “Transporter”
`and “IOFS” (input/output file system)
`offer throughput performance improve-
`ments for transfers of large video data
`files. Transporter is a protocol for
`making very efficient memory-to-
`memory data transfers between com-
`
`Economic methods are now available to move production and post-pro-
`duction data faster than real time in networked environments. Using
`ANSI-standard Fibre Channel serial interfaces on Silicon Graphics
`workstations and servers with optimized hardware and software, com-
`puter-to-disk and disk-to-disk communications have been demonstrated
`to transfer digital image data at a sustained throughput up to 600
`Mbitshec. Incorporating these inte$aces, production and post-produc-
`tion facilities are achieving order-of-magnitude improvements in
`response time when accessing and transferring large jiles. Video server
`applications can use these inte$aces for both storage access and com-
`munications for transporting up to I00 compressed streams simultane-
`ously through a single port.
`
`F expandable interface standard
`
`ibre Channel is an inexpensive,
`
`defined to achieve faster-than-real-
`time digital video data transfers among
`servers, workstations, disk drives,
`scanners, recorders, and displays. Fibre
`Channel combines the best attributes
`of a channel with those of a network
`through a simple technique: it trans-
`fers data between a buffer at the
`source device (e.g., a video server)
`and another buffer at the destination
`device (e.g., a workstation). Fibre
`Channel ignores the data itself and
`how it is formatted and simply takes
`what is in the sending buffer and
`transports it to the receiving buffer.
`Able to operate with multiple proto-
`cols simultaneously, Fibre Channel is
`an excellent interface for environ-
`ments involving a wide variety of
`computing and video equipment.
`An adapter card provides Fibre
`Channel connectivity for Silicon
`Graphics Indigo2 and Challenge M
`computers, which support the 64-bit
`version of SGI’s Graphical VO (GIO)
`Bus. Performance-enhancing hardware
`and software techniques result in the
`very highly sustainable throughput
`performance necessary for faster-than-
`
`Presented at the 137th SMPTE Technical Conference
`and World Media Expo in New Orleans (paper no.
`137-21) on September 7, 1995. Don Deel, Marc
`Friedmann. and Howard Green are with Priaa
`Networks, Inc., San Diego, CA 92121. An unedited
`version of this paper appears in Moving Imugcs:
`Meeting the Challenges, SMPTE 1995. Copyright 0
`1996 by the Society of Motion Picture and Television
`Engineers, Inc.
`
`SMPTE Journal, December 1996
`
`00
`
`5
`
`10
`
`15
`Clip Length (minutes)
`
`20
`
`25
`
`30
`
`Figure 1. Digitized film and digital video file size.
`
`751
`
`Oracle Ex. 1030, pg. 2
`
`
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`uncompressed digital video stored on
`remote servers has not been possible.
`Production and post-production
`requirements are rapidly exceeding the
`limits of traditional network approach-
`es and are driving the need for new
`network solutions. These solutions
`must address both hardware and soft-
`ware performance bottlenecks in com-
`puter-to-computer and computer-to-
`disk communications.
`Traditional Network Approaches
`Faster-than-real-time access to
`uncompressed digital video data stored
`in large disk-based files is becoming a
`key objective in today's studios. Using
`either the traditional computer or
`video server approach, many studios
`have adopted the popular client-server
`architecture, which centralizes storage
`access through a large, high-speed
`processor (Fig. 2). Local workstations
`access files through the server, which
`delivers them to local memory or disk.
`Centralized storage attached to the
`server can be quite large, frequently
`exceeding one terabyte, while local
`storage may be a few gigabytes.
`Interface from the server-to-storage
`must be as rapid as possible and is
`therefore generally in the form of
`striped disks or a redundant array of
`inexpensive drives (RAID). To maxi-
`mize performance, fadwide SCSI is
`used for the server to storage connec-
`tion. This yields transfer rates
`approaching 20 Mbytedsec, which is
`well below the 34 Mbyteshec required
`for uncompressed real-time digital
`video.
`File access across the network is in
`a packetized format using Internet pro-
`tocol (1P)-based file transfer protocol
`(FTP) or network file system (NFS)
`(Fig. 3). While the use of NFS is
`straightforward and offers flexible file
`access, its small packet sizes (typically
`less than 8 kbytes) and the consider-
`able management of each packet by
`the operating system significantly
`reduces data throughput rates. Since
`FTP procedures are disk-to-disk trans-
`fers, they are limited to the throughput
`rate of the slowest disk interface
`involved. Applications seeking perfor-
`mance higher than NFS or FTP may
`use the UNIX remote procedure call
`(RPC) directly, which operates as a
`much faster computer memory-to-
`
`SMPTE Journal, December 1996
`
`When translated to digital format,
`the video images used in studio appli-
`cations are among the most data inten-
`sive. At 1 to 40 Mbytedframe, even a
`clip of a few seconds in length rapidly
`
`grows to more than a gigabyte file
`(Fig. 1). Moving files this size from
`central storage or to a collaborator
`using traditional networks can take 10
`min or more, and real-time viewing of
`
`RAIDor
`
`Computer or
`Vldeo Server
`
`Workstations
`
`Figure 2. Client-sewer architecture.
`
`~
`
`~~
`
`APPLICATIONS
`
`UNIX File System Interface
`
`NFS
`
`IP
`
`I
`
`Network I/F Driver
`
`NETWORK PORT
`
`I
`
`I
`
`I
`
`I
`
`Figure 3. IP-based NFS protocol stack.
`
`Table 1 - Network Line Rate vs. Throughput
`Line Rate
`Network
`Throughput
`Standard
`(Mbitdsec)
`(Mbitdsec)
`80
`30
`1
`
`ATM
`FDDl
`Ethernet
`
`155
`100
`10
`
`752
`
`Oracle Ex. 1030, pg. 3
`
`
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`memory transfer. However, even
`RPC-based transfers are ultimately
`limited by the transfer rate capability
`of the network.
`Today’s computer-based studios
`generally use Ethernet or FDDI for
`network communications. To improve
`throughput, some installations are
`beginning to experiment with ATM.
`Ethernet, FDDI, and ATM local area
`networks have been optimized as
`enterprise-wide networks, which are
`dominated by small message traffic
`rather than the large block transfers
`common with digital video files.
`Although specified line rates may
`range from 10 to 155 Mbitshec, the
`combination of network and file sys-
`tem structures yields an observed actu-
`al throughput of 1 to 80 Mbits/sec
`under moderately loaded conditions
`(Table 1). Transfer of a gigabyte file,
`common in studio environments,
`requires two minutes to two hours at
`these data rates. As the number of edi-
`tors and artists using workstations
`within a facility continues to grow, the
`load is rising and the need for much
`higher speed networks is increasing
`dramatically.
`Faster-Than-Real-Time
`Interfaces
`Of the new generation of interfaces,
`Fibre Channel offers the highest per-
`formance. Approved as an ANSI stan-
`dard, Fibre Channel is a scalable inter-
`face defined to achieve high-speed
`data transfers among workstations,
`
`Table 2 - Fibre Channel Features
`
`Feature
`Line rate
`Data transfer rate
`Frame size
`Protocols
`Topology
`Data integrity
`Distance
`
`Fibre Channel
`266, 531, or 1062.5 Mbitskec
`640-720 Mbitslsec
`21 12-byte payload
`SCSI, IP, ATM, SDI, HIPPI, 802.3, 802.5
`Loop, switch
`10E-12 BER
`Local and campus; up to 10 km
`
`personal and large computers, disk dri-
`ves, peripherals, and display devices.
`Using either loop or switch-based
`topology, it combines attributes of
`SDI-like channels with packetized
`computer networks over a serial inter-
`connect capable of operating across
`campus-wide distances. Having the
`ability to support multiple protocols
`simultaneously, Fibre Channel is a
`hardware-intensive interface for envi-
`ronments involving a wide variety of
`computer, disk, and studio equipment.
`Key technical features of Fibre
`Channel are summarized in Table 2.
`The Fibre Channel structure is
`defined as a multilayered hierarchy of
`functional levels. Five layers define
`the physical media and transmission
`rates, encoding scheme, framing pro-
`tocol and flow control, common ser-
`vices, and the upper layer application
`interfaces (Fig. 4). FC-0, the lowest
`layer, specifies the physical features of
`the media, connectors, transmitters,
`and receivers, including electrical and
`
`optical characteristics, transmission
`rates, and other physical elements of
`the standard (Table 3). Note that video
`coax and the 1300-nm, single-mode
`fiber found in broadcast facilities are
`incorporated in the standard. FC- 1
`defines the 8B/1 OB encoding/decoding
`scheme used to integrate the data with
`the clock information as required by
`serial transmission techniques. FC-2
`defines the rules for framing the data
`to be transferred between ports, a
`look-ahead sliding-window flow con-
`trol scheme, different mechanisms for
`circuit and packet-switched classes of
`service, the error-detection techniques,
`and means of managing the sequenc-
`ing of data transfer. FC-3 provides
`common services required for
`advanced features, such as striping and
`hunt groups. FC-4 provides the seam-
`less integration of existing standards
`by accommodating a number of other
`protocols such as SCSI, TCP/IP,
`FDDI, HIPPI, SDI, ATM, Ethernet,
`and Token Ring.
`
`Channels
`
`Networks
`
`Fc-3
`
`FC-2
`
`FC- 1
`
`Common Services
`
`Framing Protocol / Flow Control
`
`Encode / Decode
`
`SMPTE Journal, December 1996
`
`753
`
`Oracle Ex. 1030, pg. 4
`
`
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`Signal
`
`Long-wave laser
`Short-wave laser
`Long-wave LED
`ECL
`ECL
`ECL
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`Fibre Channel combines the best
`attributes of a channel with those of a
`network through a simple technique: it
`provides a means to transfer data
`between a buffer at the source device
`(e.g., a video server drive) and another
`buffer at the destination device (e.g., a
`workstation or frame buffer). Fibre
`Channel ignores the data itself and
`how is formatted, and simply takes
`what is in the sending buffer and
`transports it to the receiving buffer at
`the full bandwidth of the channel.
`After initial handshaking, control of
`the rate of data flow is handled by the
`receiving device indicating the amount
`of available memory buffer available.
`This low-level flow control allows
`Fibre Channel to avoid any data loss
`due to congestion. Simple error cor-
`rection is handled in hardware, much
`like a channel. If a data transfer fails
`due to an error, then a retry occurs
`immediately without consulting sys-
`tem software, thus maintaining above
`real-time performance.
`Fibre Channel has four levels of
`communication across the links (Fig.
`5). Signaling occurs via ordered sets,
`which are sets of four 10-bit characters
`used for such functions as start-of-
`frame, end-of-frame, link start-up, and
`special user-defined commands. A
`frame is the smallest undivided packet
`of data sent over the connection. Each
`frame consists of a start-of-frame
`delimiter, a frame header, an optional
`payload header, a data payload hold-
`ing up to 2048 bytes, a 32-bit CRC,
`and an end-of-frame delimiter. A
`sequence is composed of one or more
`related frames flowing in the same
`direction on a link. Sequences consti-
`tute the key unit of transfer between
`
`SMPTE Journal, December 1996
`
`Table 3 - Fibre Channel Media
`
`Medium
`
`Single mode fiber
`50ym multimode fiber
`67-vrn multimode fiber
`Video coax
`Miniature coax
`Shielded twisted pair
`
`Maximum
`Distance
`
`Data Rate
`(M bitdsec)
`
`10 km
`2 km
`1.5 km
`100 rn
`35 rn
`100 rn
`
`266,531,1062
`266,531,1062
`133,266,531, 1062
`133,266,531,1062
`133,266, 531, 1062
`133,266
`
`EXCHANGE I
`
`Sequence I G2
`...
`
`Sequence 2
`
`FRAME
`
`oldend Set
`
`SIGNAWNG
`
`Ordered Sel
`
`Figure 5. Fibre Channel communications structure.
`
`RAIDDrive
`
`Sewer E l
`n UG
`
`(Circuit
`
`Figure 6. Fibre Channel classes of service.
`
`754
`
`workstations
`
`Oracle Ex. 1030, pg. 5
`
`
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`ports that have negotiated available
`buffers. Each sequence is identified
`uniquely, and every frame within each
`sequence is individually numbered to
`facilitate error detection and reassem-
`bly upon arriving at its destination. An
`exchange consists of one or more non-
`concurrent sequences in a series of
`
`communications between two devices.
`Several exchanges between the same
`two devices may be occurring at the
`same time, with each exchange in a
`different phase of progress (e.g., initial
`handshake, data transfer, close of
`transfer, etc.).
`To accommodate on-line, off-line,
`
`Workstation
`
`
`
`Tape Drive Tape Drive
`
`\
`
`
`
`Server Server
`
`w
`
`RAIL) Drives
`
`I
`
`I
`
`video, and computer needs, Fibre
`Channel defines four different classes
`of service (Fig. 6). Class I , a circuit-
`switched connection, functions much
`in the same way as today’s SDI physi-
`cal channels. No other devices can
`share the engaged link when a Class I
`connection has been established
`between two devices. Class 2 is a con-
`nectionless, frame-switched link that
`provides guaranteed delivery with
`acknowledgment of receipt. As with
`traditional packet-switched networks,
`the path between two ports is not dedi-
`cated, allowing for shared use of the
`link’s bandwidth. Class 3 is a connec-
`tionless “datagram” service that allows
`data to be sent rapidly to multiple
`devices attached to the fabric, but no
`confirmation of receipt is given. By
`not having to wait for confirmation,
`Class 3 service speeds up the time of
`transmission. However, if a single
`user’s link is busy, the hardware will
`not immediately know to retransmit
`the data. Class 4 offers constant avail-
`able minimum bandwidth or guaran-
`teed latency and is useful for isochro-
`nous applications such as single or
`multiple streams of real-time digital
`video.
`Fibre Channel supports a variety of
`fabric topologies. It is a closed system
`
`Figure 7. Fibre Channel arbitrated loop topology.
`
`(FC) Controller --
`
`Fibre Channel
`
`--
`
`Fibre Channel
`(FC) Controller
`
`Port
`
`FC Controller
`Interface Logic
`
`Memory
`
`FCController
`Interface Logic
`
`Local
`I--)
`Memory
`
`BiDi FIFO and
`BiDi Register
`
`I
`
`Figure 8. NetFX hardware block diagram.
`
`SMPTE Journal, December 1996
`
`BiDi FIFO and
`BiDi Register
`structure
`
`Host Bus
`Interface Logic
`
`GI064 or
`HI064 Bus
`
`755
`
`Oracle Ex. 1030, pg. 6
`
`
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
`arbitrated loop, and switched fabric
`Fibre Channel topologies and provides
`Class 1, Class 2, Class 3, and Intermix
`classes of service. Copper or fiber-
`optic media can be used by the ports at
`the standard speeds of 266, 531, and
`1062 Mbitshec.
`The NetFX Fibre Channel ports
`have been designed to support an aver-
`age sustained data transfer rate in
`excess of 80 Mbytedsec (640
`Mbitshec) for multi-megabyte trans-
`fers. Delivered performance is highly
`dependent upon the specific mix of
`large block data transfers and small
`message traffic, which trade off
`against each other. Because of per-
`transfer handling overheads, large
`block data transfers result in higher
`data throughput rates than small mes-
`sage traffic.
`NetFX Hardware
`The hardware for the NetFX Fibre
`Channel adapter includes gigabaud
`link modules (GLMs), Fibre Channel
`(FC) controllers, FC controller inter-
`face logic, local memories, BiDi FIFO
`and BiDi register structures, and a host
`bus interface logic section (Fig. 8).
`GLMs perform high-speed 20: 1 par-
`allel-to-serial and 1 :20 serial-to-paral-
`
`lel functions and allow each port to
`communicate using either copper or
`fiber-optic media. Since different
`GLMs are required to support specific
`Fibre Channel speed and media com-
`binations, each port on the NetFX
`adapter has a GLM socket that allows
`it to be configured independently. This
`permits NetFX adapters to support two
`different types and speeds of Fibre
`Channel media simultaneously.
`The FC controller performs the
`high-speed FC-1 and FC-2 functions
`and assists with some of the FC-4
`functions. Each FC controller connects
`directly to its GLM and handles the
`lower-level communications functions
`in hardware. Most of the FC con-
`troller’s interactions with the host sys-
`tem are via direct memory access
`(DMA) operations, which are used to
`access commands, deliver status, and
`perform data transfer operations.
`Memory structures are used for com-
`munications between the NetFX
`adapter and the host system to mini-
`mize the number of interrupts that
`must be handled by the host operating
`system.
`Each FC controller interface logic
`block manages requests made by its
`FC controller for DMA activities, and
`
` @ SMPTE All Rights Reserved
`
`SocketdSJSRoami
`
`I *...el
`
`Diagnostics
`
`SMPTE Journal, December 1996
`
`1
`
`I
`
`1
`
`I
`
`
`
`SCSI Initintor
`
`I
`
`Ngon Driver
`
`Hudware Driver
`
`rigure 9. NetFX software hierarchy.
`
`that relies on ports logging in with
`each other and the fabric trading infor-
`mation on attributes in order to deter-
`mine if information can be exchanged.
`Possible fabric topologies include
`point-to-point, crosspoint-switched,
`and arbitrated loop. The highest-per-
`formance Fibre Channel fabric is
`based on a point-to-point or a cross-
`point-switched topology. The Fibre
`Channel arbitrated loop topology (Fig.
`7) offers a lower-cost connection alter-
`native that does not require the use of
`switches. With up to 127 ports con-
`nected on a single loop, each port can
`compete for a connection through a
`defined arbitration process. When
`arbitration is complete, the two suc-
`cessful nodes obtain access to the full
`bandwidth of the link. All classes of
`service can be supported by either the
`switch or arbitrated loop topologies.
`Fibre Channel Adapters for the
`Studio Environment
`The most popular servers and work-
`stations used to work with digitized
`film and digital video are made by
`Silicon Graphics, Inc. On the higher-
`performing end of the spectrum,
`Challenge DM, L, and XL machines
`are used as servers, and Onyx systems
`are used as workstations. On the
`lower-cost end of the spectrum,
`Challenge M machines are used as
`servers and Indigo2 systems are used
`as workstations. All of these machines
`can communicate effectively using
`Fibre Channel.
`To achieve the high data throughput
`rates made possible by Fibre Channel,
`it is necessary to have access to the
`fastest I/O bus present in the host
`machine. In the SGI Onyx and
`Challenge DM, L, and XL systems,
`this is the HI0 bus. In the Indigo2 and
`Challenge M systems, this is the
`GI064 bus. Both of these UO buses
`are capable of providing YO adapters
`with access to host memory at data
`rates in excess of 200 Mbyteshec.
`NetFX Fibre Channel adapters are
`made for both GI064 bus connections
`and HI0 bus connections.
`NetFX Fibre Channel adapters pro-
`vide one or two independent, fully
`capable Fibre Channel ports, allowing
`the SGI system to communicate with
`disk and network traffic simultaneous-
`ly. Each port supports point-to-point,
`
`756
`
`Oracle Ex. 1030, pg. 7
`
`
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`data transfers and sustain high data
`throughput rates by minimizing the
`performance impacts introduced by
`bus arbitration and host memory
`access latency. The BiDi registers are
`used to facilitate host-driven accesses
`of the local memories and the configu-
`ration and status registers in the FC
`controllers.
`The host bus interface logic man-
`ages communications via the connec-
`tion to the host system’s UO bus and
`includes special provisions for
`improving data throughput rates by
`pipelining all DMA write and read
`operations to and from host memory.
`This helps to minimize the perfor-
`mance impact caused host memory
`access latency. Both master and slave
`modes of bus operation are supported
`by the host bus interface logic, which
`also provides diagnostic control and
`status functions that support testing
`and servicing the NetFX adapter.
`NetFX Software
`The NetFX Fibre Channel adapter
`supports a hierarchical software facili-
`ty that includes low-level drivers, pro-
`tocols, network management support
`and diagnostics (Fig. 9). NetFX soft-
`ware has been optimized for very
`highly sustained data throughput while
`supporting high transaction rates for
`small-message traffic. Implemented
`using standard interfaces provided by
`the SGI IRIX operating system, the
`hardware and N-Port drivers allow the
`host operating system software to
`communicate with the NetFX hard-
`ware. Protocol software enables the
`
`handles requests from the host system
`for accessing the port’s local memory,
`FC controller, and GLM. DMA
`requests from the FC controller are
`directed to either the local memory or
`to the host memory, depending upon
`the referenced address. Requests from
`the host system are for accessing data
`structures in the local memory, for
`accessing the control and status regis-
`ters in the FC controller, and for exer-
`cising control and status functions
`over the GLM.
`Local memory is provided for each
`FC port to enhance overall system per-
`formance. Each port’s local memory
`
`can be accessed both by the FC con-
`troller and by the host system. By
`placing most of the memory data
`structures for commands and status in
`local memory, DMA operations to and
`from host memory are largely limited
`to data transfers. This enhances perfor-
`mance by minimizing the time it takes
`for the FC controller to access the
`memory structures it shares with the
`host system.
`The BiDi FIFO and BiDi register
`structures are used by the FC con-
`trollers and the host bus interface logic
`as communications buffers. The BiDi
`FIFOs are used to “pipeline” DMA
`
`APPLICATIONS
`
`UNIX File System Interface
`
`NFS
`
`L
`
`I
`I
`
`I
`
`Network I/F Driver
`
`FIBRE CHANNEL PORT
`
`Figure 70. IOFS and NFS protocol stacks.
`
`Figure 7 7. Throughput test setup.
`
`SMPTE Journal, December 1996
`
`757
`
`Oracle Ex. 1030, pg. 8
`
`
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`Fibre Channel has the ability to
`interconnect multiple systems, as do
`conventional networks. It also allows
`large data transfers to proceed at near-
`ly the full speed of the communica-
`tions media and incorporates hard-
`ware-level flow control and error-
`detection capabilities. These character-
`istics are typical of conventional YO
`channels. Transporter takes advantage
`of these capabilities by implementing
`network-style functions in an YO-style
`fashion.
`The Transporter software supports a
`new file access protocol called
`“IOFS.” Offering a client-server func-
`tionality very similar to NFS, IOFS
`uses Transporter to send large blocks
`of data between machines at very high
`data transfer rates. This allows it to
`bypass many of the software-level
`flow control and error-detection proto-
`cols used in conventional networks
`(Fig. lo). IOFS does not require
`changes to either application files or
`programs because it allows files to be
`manipulated in the same ways as NFS
`does. Additionally, IOFS can coexist
`with NFS to provide parallel access
`paths to the same files.
`IOFS makes files accessible to
`remote systems in a manner similar to
`NFS. IOFS server software “exports,”
`or makes available, any of the file sys-
`tems that are local to the server sys-
`tem. IOFS client software “mounts,”
`or locally attaches, any remote file
`system that it is authorized to use.
`Once an IOFS client has mounted a
`remote file system, it accepts file
`access requests from its local applica-
`tions programs. The UNIX system
`directs requests to the IOFS client
`based upon the name of the file being
`accessed. Requests are for any of the
`standard kinds of file-related opera-
`tions, such as reading, writing, creat-
`ing, renaming, deleting, or inquiring
`about files.
`IOFS clients communicate requests
`to the server. IOFS servers process
`each request by referring it to the local
`file system; in the case of IRIX, this
`would typically be an EFS or XFS file
`system. The local file system is
`responsible for actually carrying out
`the request. When the local file system
`completes a request, the IOFS server
`returns a reply to the client. The reply
`not only provides the client - and
`
`SMPTE Journal, December 1996
`
`0
`
`F’DDI
`
`SDI
`
`FC Adapter
`(prototype)
`
`FC Adapter
`(Expected)
`
`Figure 12. Relative throughputs.
`
`operation system to communicate with
`attached Fibre Channel devices.
`Facilities for SNMP-based network
`managers are supported, as are diag-
`nostics, to maintain the network and
`verify the correct operation of the
`hardware.
`Hardware driver software generates
`commands and responds to interrupts
`and status from the Fibre Channel
`adapter hardware. N-Port driver soft-
`ware coordinates interactions with the
`hardware to implement Fibre Channel
`FC-2 functions including link manage-
`ment, fabric and N-Port
`login,
`exchange resource allocation, sequence
`sending and receiving, acknowledg-
`ment handling, and input stream
`demultiplexing. These functions per-
`form essential services and handshake
`functions for FC-4 protocols.
`NetFX I/O and network software
`facilities are made available to the
`operating system and to user-level
`applications by the FC-4 protocols
`above the N-Port driver. YO-oriented
`FC-4 protocols for encapsulating SCSI
`traffic are provided for communicating
`with attached Fibre Channel I/O
`devices, such as disk drives. Both
`SCSI initiator and SCSI target mode
`operations enable software to initiate
`or receive SCSI commands, perform
`data transfers, and receive or return
`responses. These FC-4 protocols com-
`ply with the interoperability require-
`ments set forth in the Fibre Channel
`Systems Initiative (FCSI) SCSI Profile
`document.
`
`758
`
`A network-oriented FC-4 protocol
`for encapsulating Internet protocol
`(IP) is also made available to the oper-
`ating system. This software allows the
`operating system to send and receive
`IP traffic over Fibre Channel for stan-
`dard data communications protocols,
`such as TCP, UDP, NFS, SNMP, and
`Telnet. This FC-4 protocol supports
`functionality specified in the FCSI ZP
`Profile document.
`NetFX Transporter and IOFS
`Software
`A proprietary FC-4 called the
`“Transporter” protocol has been devel-
`oped that optimizes transfers of large
`blocks of data between computers that
`are connected by Fibre Channel. This
`FC-4 protocol is ideal for moving digi-
`tized film and uncompressed digital
`video data between different comput-
`ers very quickly; in the case of digital
`video data, it can make these transfers
`happen faster than real time.
`Transporter treats data transfers
`between computer systems as memory-
`to-memory UO operations, rather than
`as the more traditional data communi-
`cations networking operations used in
`conventional local area networks, such
`as Ethernet and FDDI. It avoids many
`of the software overheads and ineffi-
`ciencies associated with standard IP-
`based data communications protocols
`by utilizing the “native” capabilities of
`Fibre Channel, most of which are
`implemented in hardware for maxi-
`mum throughput efficiency.
`
`Oracle Ex. 1030, pg. 9
`
`
`
`Downloaded by guest on 2014-07-10 from IP
`
`162.250.242.2
`
` @ SMPTE All Rights Reserved
`
`MOVING UNCOMPRESSED VIDEO FASTER THAN REAL TIME
`
`ultimately the application - with the
`requested information, but also pro-
`vides confirmation that the request
`was correctly communicated.
`IOFS is not concerned with actually
`managing stored information on disks
`or other media. For that purpose, IOFS
`makes use of the storage management
`services provided by the local file sys-
`tems. The main purpose of IOFS is to
`provide throughput-effective access to
`files. Its function is essentially one of
`communication, and in this respect it
`is identical to NFS. The difference
`between IOFS and NFS is how data is
`moved from one system to another.
`While NFS uses traditional IP-based
`data communications protocols and
`techniques, IOFS uses Transporter,
`which takes advantage of Fibre
`Channel’s hardware-based high data
`throughput capabilities.
`NetFX Throughput
`Two SGI Indigo2 workstations were
`connected together with prototype
`NetFX Fibre Channel adapters to run
`throughput tests. Application-level
`
`programs were used to repeatedly send
`a multi-megabyte data buffer from one
`Indigo2’s memory to the other
`Indigo2’s memory (Fig. 1 1), and the
`actual throughput was measured.
`Running this test with prototype
`boards, 60 Mbyteshec (480 Mbitdsec)
`was measured for the end-to-end,
`application-to-application sustained
`throughput. This transfer rate is
`approximately twice that needed for
`real-time digital video.
`It is important to note, however, that
`the prototype NetFX Fibre Channel
`adapters are running with an early ver-
`sion of the FC controller device that
`does not support the full speed that the
`hardware was designed for. The new
`version of the FC controller device is
`now becoming available, and when it
`is incorporated into the NetFX hard-
`ware, the resulting throughput is
`expected to go up to 80 Mbytedsec
`(640 Mbitdsec).
`The differences between the sus-
`tainable throughput rates that are pos-
`sible using FDDI, SDI, and the NetFX
`Fibre Channel adapters are dramatic
`
`(Fig. 12). The sustainable throughput
`rate shown for FDDI is a typical num-
`ber for moderately loaded conditions,
`and the rate shown for SDI is the line
`rate.
`Conclusion
`Using Fibre Channel, it is possible
`to move uncompressed digital video
`around in the studio environment at
`faster-than-real-time speeds. Fibre
`Channel adapters have been created
`for the computer systems most com-
`monly used in studios, and early
`throughput measurements show data
`transfer rates of 480 Mbits/sec
`between these systems, with the
`transfer rate expected to go