NETAPP, INC. EXHIBIT 1009
`
`Page 1 of 28
`
`NETAPP, INC. EXHIBIT 1009
`Page 1 of 28
`
`

`
`For close to a decade, Cray Research has been the industry leader in
`large-scale computer systems. Today, about 70percent of all
`supercomputers installed worldwide are Cray systems. They are used in
`advanced scientific and research Eaboratories around the world and have
`gained strong acceptance in diverse industrial environments. No other
`manufacturer has Cray Research's breadth of success and experience in
`supercomputer development.
`
`The company's initial product, the CRAY-I Computer System, was first
`installed in 1976and quickly became the standard for large-scale
`scientific computers -and the first commercially successful vector
`processor. For some time previously, the potential advantages of vector
`processing had been understood, but effective practical implementation
`had eluded computer architects. The CRAY-1 broke that barrier, and today
`vectorization techniques are used commonly by scientists and engineers
`in a wide variety of disciplines.
`
`The field-proven GRAY X-MP Computer Systems now offer significantly
`more power tosolve new and bigger problems while providing better value
`than any other systems available. Large memory size options allow a wider
`range of problems to be solved, while innovative multiprocessor design
`provides practical opportunities to exploit multitasking, the next dimension
`of parallel processing beyond vectorization. Once again, Cray Research,
`has moved supercomputing forward, offering new levels of hardware
`performance and software techniques to serve the needs of scientists and
`engineers today and in the future.
`
`NETAPP, INC. EXHIBIT 1009
`Page 2 of 28
`
`

`
`Introducing the CRAY X-MP
`
`Series of Computer Systems
`
`
`Announcing expanded capabilities to serve the
`needs of a broadening marketplace: the CRAY
`X-MP Series of Computer Systems. The CRAY
`X-MP Series now comprises nine models, ranging
`from a uniprocessor version with one million words
`of central memory to a top-end system with four
`processors and a 16-million-word memory. Today's
`CRAY X-MP line is a field-proven technology; Cray
`Research introduced the dual-processor CRAY
`X-MP in 1982 and expanded the series to include
`one- and four-processor models in 1984.
`
`The flexible CRAY X-MP multiprocessor
`corifigurations allow users to employ
`multiprogramming, multiprocessing and
`multitasking techniques. The multiple-processor
`architecture can be used to process many different
`jobs simultaneously for greater system throughput,
`or it can apply two or more processors to a single
`job for better program turnaround time. The
`combination of multiprocessing and vector
`processing provides a geometric increase in
`computational performance over conventional
`scalar processing techniques.
`
`The CRAY X-MP system design is carefully
`balanced to deliver optimum overall performance.
`Fast long and short vector processing is balanced
`with high-speed scalar processing, and both are
`supported by powerful inputloutput capabilities.
`Cray Research software has been developed to
`ensure easy access to these performance features.
`The result is that users can realize maximum
`throughput for a variety of job mixes and
`programming environments.
`
`NETAPP, INC. EXHIBIT 1009
`Page 3 of 28
`
`

`
`-
`
`---=
`
`-
`
`Each X-MP CPU offers gatherlscatter and
`compressed index vector instructions. These
`instructions allow for the vectorized processing of
`randomly organized data, which previously was
`performed by scalar processing.
`
`Complementing the power of the X-MP Series is a
`new generation of I10 technology. Cray's DD-39
`and DD-49 disk drives offer 1200-Megabyte
`(Mbyte) capacity and very fast sustained transfer
`rates (9.8 Mbytelsec for a DD-49,5.9 Mbytelsec for
`a DD-39).In addition, Cray's Solid-state Storage
`Device (SSD) provides up to 1024 Mbytes of very
`fast random-access secondary MOS memory.
`When connected to a four-processor CRAY X-MP
`through two 1000-Mbytelsec channels, it provides
`a maximum aggregate transfer rate of 2000
`M bytelsec.
`
`A wide variety of applications programs for solving
`problems in industries such as petroleum,
`aerospace, automotive, nuclear research and
`chemistry are available for operation on CRAY
`X-MP computers. Thus, scientists and engineers
`can use X-MP systems and industry standard
`
`--
`
`codes to solve a wide range of problems.
`Additionally, software developed for the CRAY-1
`can be run on all models of the CRAY X-MP Series,
`thus protecting user software investment.
`
`From both a hardware and software standpoint, the
`CRAY X-MP can be integrated easily into a user's
`existing computer environment. Hardware and
`software front-end interfaces for other
`manufacturers' equipment are available. And the
`CRAY X-MP requires a minimum of floor space,
`occupying just 112 square feet (1 1 square meters)
`in its maximum configuration, including the
`Solid-state Storage Device.
`
`Cray computers offer the most powerful and
`cost-effective computing solutions available today
`for advanced scientific applications - both for
`experienced supercomputer users with the most
`demanding computing requirements and for newer
`users whose research needs now require
`supercomputer power. The CRAY X-MP features
`one or more powerful CPUs, a very large central
`memory, exceptionally fast computing speeds and
`I10 throughput to match. As the supercomputer
`marketplace broadens, the CRAY X-MP Series of
`Computer Systems will evolve to meet users'
`expanding computing requirements.
`
`Structural analysis
`
`Frmte element analysls IS a
`mathematical method for
`calculatrng the effects of
`temperature- and pressure-
`related stress on physical
`structures The aerospace,
`automotive and crvrl engrneer-
`ing rndustrres rely on the
`method to conduct engineermg
`design and analysis Usrng
`
`CRAY X-MP systems, sclentlsts
`can rapidly evaluate structures
`too large or complex to be
`analyzed adequately any other
`way The result is improved
`engineering effrcrency and
`more structurally sound and
`lrghtweight components
`
`Experimental and computa-
`tronal results of the impact of a
`nose cone aga~nst an angled
`solrd surface (Left) Actual re-
`sults and (right) the numerical
`srmulat~on of this highly non-
`linear process The striking
`similarity between the two con-
`firms the validrty of the com-
`putational approach (Cred~tLaw-
`rence Lwerrnoreand Sandla Nat~onal
`Laborator~es)
`
`NETAPP, INC. EXHIBIT 1009
`Page 4 of 28
`
`

`
`CRAY X-MP multiprocessor system organization
`
`L-)
`
`Subsystem
`I \
`
`------ control path
`
`aata path
`
`~llustrat~ngstram energy denw
`
`ty d~str~but~on under f~rst mode
`of vlbrat~on. (Credit "1984,Ford
`MotorCo )
`
`IDeformed powertram assernbl)
`
`NETAPP, INC. EXHIBIT 1009
`Page 5 of 28
`
`

`
`- u7I---
`
`
`Systemoverview
`-- - -- -- -
`
`I
`
`
`NETAPP, INC. EXHIBIT 1009
`Page 6 of 28
`
`

`
`The CRAY X-MP/4 computers
`
`The top-of-the-line CRAY X-MP/4 computer
`systems offer an order of magnitude greater
`performance than the original CRAY-1 .They are
`configured with eight or sixteen million 64-bit words
`of ECL bipolar memory and provide a maximum
`memory bandwidth 16 times that of the CRAY-1.
`Central memory has a bank cycle time of 38
`nanoseconds (nsec) and is shared by four identical
`CPUs with a clock cycle time of 9.5 nsec. The
`X-MP/4 mainframe is the familiar 12-column 270°
`arc chassis with the same electrical requirements
`as the CRAY-1.
`
`Each of the four CRAY X-MPl4 processors has
`scalar and vector processing capability and can
`access all of central memory. The CPUs may
`operate independently on separate jobs or may be
`organized in any combination to operate jointly on a
`single job.
`
`The raw computational power of the CRAY X-MPl4
`systems is augmented by the powerful inputloutput
`and data-handling capabilities of the CRAY I10
`Subsystem (10s). The IOS is integral to all CRAY
`X-MP computers and enables fast, efficient data
`access and processing by the CPUs.
`
`Cray Research's DD-49 disk drive matches the
`power of the X-MPl4 models, offering 1200-Mbyte
`capacity, sustained transfer rates of 9.8 Mbytelsec
`and very fast access times (2 milliseconds).
`
`In addition to high-capacity, fast access disk
`technology, the f~eld-proven SSD offers up to 1024
`Mbytes of very fast random-access secondary
`MOS memory. The SSD connects to the CRAY
`X-MP/4 systems through two very high-speed
`channels with a maximum aggregate transfer rate
`of 2000 Mbytelsec. The SSD, in conjuction with the
`X-MP/4 multiprocessor architecture, enables users
`to fully exploit existing applications and to develop
`new algorithms to solve larger and more
`sophisticated problems in science and engineering
`- problems that could not be attempted before due
`to computational or 110 limitations.
`
`These cross-sectlons repre-
`sent three-d~mens~onal airflows
`as deplcted by the Navier-
`Stokes equations Their com-
`putational requirements make
`the Navrer-Stokes equations
`the most difficult to solve, but
`they are also the most accu-
`rate (Credit W L Hankey,s J Scherr
`J S Shang.Air ForceWrtghtAeronaut~cal
`Laboratones)
`
`NETAPP, INC. EXHIBIT 1009
`Page 7 of 28
`
`

`
`The CRAY X-MPI2 computers
`
`The field-proven CRAY X-MPl2 models have
`become the established price and performance
`leaders in the supercomputer industry. The new
`X-MP dual-processor systems offer up to four times
`the memory and require only half the electrical
`power of the original CRAY X-MPl2 systems.
`Overall throughput is typically three to five times
`that of a CRAY-1.
`
`The CRAY X-MP/2 systems are available with
`four, eight or sixteen million 64-bit words of shared
`MOS central memory, providing a maximum
`memory bandwidth four times that of the CRAY-1.
`Each CPU has a 9.5 nsec clock cycle time and
`memory bank cycle time of 76 nsec. The CRAY
`X-MP/2 models consist of eight vertical columns
`arranged in a 1 80° arc.
`
`As with the X-MP/4 systems, the CRAY X-MPl2
`CPUs can operate independently on different
`programs or can be harnessed together to operate
`on a single user program.
`
`CRAY X-MP/2 computers incorporate the same 110
`Subsystem and SSD hardware as the X-MP/4
`models. One SSD channel, with a total transfer rate
`of 1000Mbytelsec, connects the optional SSD to
`the mainframe. Typically, the system is configured
`with DD-49 disk drives.
`
`Inducinga shock a the ground
`and recording sound waves n-
`fleeted back to the surface 1s a
`methodscientists use to "see"
`underground structures. The
`methad is called reflection
`%eieamology and can Indicate
`the presence or absence of pe-
`troleum and athw resource
`depasits.However, the amount
`
`af data needed to prafile a
`large volume of earth accurate-
`ly can be Immense, and the re-
`quired analyses are stagger-
`ingly complex. GRAY X-MPsys-
`tems Can perform detailed
`analysea on these large
`amouflT8of data in a trmely
`and cost-effective way, savlnp)
`petroleum companies time
`and maney.
`
`NETAPP, INC. EXHIBIT 1009
`Page 8 of 28
`
`

`
`The CRAY X-MP/1 computers
`
`The CRAY X-MP11 models combine a single CRAY
`X-MP CPU with one, two, four or eight million 64-bit
`words of static MOS memory. Memory bandwidth is
`four times that of the CRAY-1. Single processor
`CRAY X-MP systems typically provide the user with
`1.5 to 2.5 times CRAY-1 power at a comparable
`cost. The CRAY X-MP/1 CPU has a 9.5 nsec clock
`cycle time, and a memory bank cycle time of 76
`nsec. The X-MPI1 mainframe is a six-column, 1350
`arc chassis requiring the same electrical power as
`the X-MPl2.
`
`CRAY X-MP/1 models use the same I10 Subsystem
`and support the same range of Solid-state Storage
`Device models as the CRAY X-MP/2 models.
`Typically, the X-MP/1 is configured with DD-39 disk
`drives.
`
`With the availability of a wide range of applications
`software and its superior price/performance
`characteristics, the entry-level CRAY X-MP is
`particularly appropriate for the first-time
`supercomputer customer.
`
`Golor varlawe aenslty aapiays
`of a seismlcwave propagation
`from afluid through a faulted
`structure. The full elastic egua-
`t~onwas used, demonstrating
`the conversion of P waves
`(red) Into S waves (blue). (credtt
`DanKosloffand Moshe Reshaf. Unlverslty
`of TelAvn)
`
`common source selsmlc
`record after depth migration. It
`contains 96 traces, each six
`seconds long and sampled at
`an interval of four milliseconds,
`produc~nga total of 144.000
`samples. Hundreds of these
`records are requrred to study
`subsurface geology. (ereat ~ e o -
`Quest Internatmal.Inc)
`
`NETAPP, INC. EXHIBIT 1009
`Page 9 of 28
`
`

`
`CRAY X-MP design
`
`The CRAY X-MP Series design combines
`high-speed scalar and vector processing with
`multiple processors,large and fast memories and
`high-performance 110. The result is exceptional
`speed and high overall system throughput.
`Innovative architecture and technologies built into
`the CRAY X-MP make such performance a
`practical reality.
`
`Processors
`
`Each CRAY X-MP processor offers very fast scalar
`processing with high-speed processing of long and
`short vectors. Additionally, multiprocessor models
`enable the user to exploit the extra dimension of
`multitasking.
`
`The scalar performance of each processor is
`attributable to its fast clock cycle, short memory
`access times and large instruction buffers. Vector
`performance is supported by the fast clock, parallel
`memory ports and flexible hardware chaining.
`These features allow simultaneous execution of
`memory fetches, arithmetic operations and memory
`stores in a series of linked vector operations. As a
`result, the processor design provides high-speed
`and balanced vector processing capabilities for
`short and long vectors characterized by heavy
`register-to-register or heavy memory-to-memory
`vector operations.
`
`The overall effective performance of each
`processor executing typical user programs with
`interspersed scalar and vector codes (usually short
`vectors) is ensured through fast data flow between
`
`scalar and vector functional units, short memory
`access time for vector and scalar references and
`short start-up time for both scalar and vector
`operations. As a result, CRAY X-MP computers offer
`high performance using the standard FORTRAN
`compiler, without the need for hand-coding or
`algorithm restructuring.
`
`On all models, a second vector logical unit is used
`to provide twice the execution speed of bit-level
`logical operations in each CPU.
`
`Each X-MP processor also includes instructions for
`the efficient manipulation of randomly distributed
`data elements and conditional vector operations.
`Gatherlscatter instructions allow for the
`vectorization of randomly organized data, and the
`compressed index instruction allows for the
`vectorization of unpredictable conditional
`operations. With these features, CPU performance
`can be improved by a factor of five for program
`segments dependent on the manipulation of sparse
`matrices.
`
`Central memory
`
`Depending on the model, one to sixteen million
`64-bit words of directly addressable memory is
`available with the CRAY X-MP Series. Options for
`field upgrade of memory are available on all
`models. The large memory sizes enable users to
`solve larger problems than before without the need
`for out-of-memory techniques. CRAY X-MP memory
`features single-bit error correction, double-bit error
`detection (SECDED) logic.
`
`Nuclear energy researct
`
`Computer simulation of nuclear
`
`power plants requlres the most
`
`advanced computer systems
`
`available Only supercomput-
`
`ers such as the CRAY X-MP
`
`provlde the computmg power
`
`needed to s~mulate the intrl-
`
`cate flu~d flow, heat transfer
`
`and neutronics phenomena
`
`that character~ze today's nucle-
`
`
`(Far left) A pressurized water
`nuclear reactor. The reactor
`core, primary and secondary
`heat exchange loops and con-
`tainment system are shown.
`(Left) A cylindrical section of a
`nuclear reactor core with a cal-
`culated three-dmensional
`pressure field. The vertical
`color scale indicates the pres-
`sure drop ( AP) in the vertical
`direction.
`
`NETAPP, INC. EXHIBIT 1009
`Page 10 of 28
`
`

`
`fetches, one for result store and one for indepen-
`dent I10 operations. Thus, each processor of a
`CRAY X-MP system has four times the memory
`bandwidth of a CRAY-1. Ensuring high efficiency,
`the multiport memory has built-in conflict resolution
`hardware to minimize delays and maintain the
`integrity of simultaneous memory references to the
`same memory bank.
`
`The interleaved and efficient multiport memory
`design, coupled with the short memory cycle time,
`provides high-performance memory organization
`with sufficient bandwidth to support high-speed
`CPU and I10 operations in parallel.
`
`Multiprocessors and multitasking
`
`The CRAY X-MP multiple-CPU configurations have
`made Cray Research the recognized leader in
`multiprocessing. They continue to offer users the
`opportunity to process jobs faster than with single
`CPUs by using either multiprocessing or
`multitasking techniques.
`
`Multiprocessing allows several programs to be
`executed concurrently on multiple CPUs of a single
`mainframe. Multitasking is a feature that allows two
`or more parts of a program (tasks) to be executed in
`parallel sharing a common memory space, resulting
`in substantial throughput improvements over
`serially executed programs. Performance
`improvements are in proportion to the number of
`tasks that can be constructed for the program and
`the number of CPUs that can be applied to the
`separate tasks.
`
`The CRAY X-MP multiprocessor systems share a
`central memory organized in interleaved memory
`banks that can be accessed independently and in
`parallel during each machine clock period. Each
`X-MP processor has four parallel memory ports
`connected to central memory: two for vector
`
`refinetheir themes fblsbrthen
`would be pesotbleby any mer
`means
`
`Incertain fieldsof phy$~m
`such 8s quantum chrsma-
`dynamicsand condenaelhed
`matter physns,expenrnanta-
`tion 15dlffr~ultif nst im ossible.
`Butby tapping the CW!Y
`X-MP$ extraordinaryprocess-
`ing pawer, phpicists can ax-
`perlmenton mathemati
`madelaof a-tornicand auk-
`
`6;narge aenslty contours tar an
`atomic everlayer af cesium an
`~ung&iI. Usinga Crey sy8tern
`for camgutetion and graphics
`
`se materials and
`obtermed results ~mpo%siblato
`
`NETAPP, INC. EXHIBIT 1009
`Page 11 of 28
`
`

`
`When executing in multitasking mode, all
`processors are identical and symmetrical in their
`programming functions; no CPU is dedicated to any
`one function. Any number of processors (a cluster)
`can be dynamically assigned to perform multiple
`tasks of a single job. In order to provide flexible and
`efficient multitasking capabilities, special hardware
`and software features have been built into the
`systems. These features allow one or more
`processors to access shared memory or
`high-speed registers for rapid communication and
`data transmission between CPUs. All of these
`capabilities are made available through library
`routines which can be accessed from FORTRAN. In
`add~tion,hardware provides built-in detection of
`deadlocks within a cluster of processors.
`
`Experience shows that multitasked applications
`running on CRAY X-MP/2 computers can realize
`speed increases of 1.8to 1.9 times over
`single-processor CRAY X-MP execution times;
`speed increases of 3.5 to 3.8 times have been
`achieved with the CRAY X-MP/4 systems.
`
`Input/output processing
`
`For super-scale problems requiring extensive data
`handling, Cray has developed hardware that
`ensures computing power is not held captive by I10
`limitations. The architecture of the IOS, with its
`parallel data paths and direct access to main
`memory, results in a very high I10 bandwidth with a
`minimum of interference to computation.
`
`probes an3 niedlcal ~ m a g ~ n ~
`
`technologies generate tremen-
`
`k u r @mount$of dab. Haw-
`
`
`jlgital imaging technology are
`
`:Omposed of mtllionsof tiny
`
`
`I
`
`The I10 Subsystem (10s) is an integral part of the
`CRAY X-MP design and acts as a data distribution
`point for the X-MP mainframe. The IOS handles I10
`for a variety of front-end computer systems and
`peripherals such as disk units and plug-compatible
`IBM Series 3420 and 3480 tape subsystems. The
`IOS includes two, three or four interconnected 110
`processors, each with its own local memory, and a
`common buffer memory.
`
`NETAPP, INC. EXHIBIT 1009
`Page 12 of 28
`
`

`
`Buffer memory is solid-state secondary storage,
`accessible by all of the I10 processors in the IOS.
`With its 8,32 or 64 Mbytes of static MQS memory, it
`provides I10 buffering of data to and from the
`peripheral devices. It can also be used to store user
`datasets, thus contributing to faster and more
`efficient data access by the CPUs.
`
`Complementing and balancing CRAY X-MP
`computing speeds are the DD-39 and DD-49 disk
`drives, high density (1 200-Mbyte) magnetic
`storage devices. The DD-39can sustain a data
`transfer rate of 5.9 Mbytelsec with an average
`access time of 18 milliseconds (msec); the DD-49
`can sustain a rate of 9.8 Mbytelsec with an average
`access time of 16 msec. These disks are the fastest
`available, and when combined with the data
`handling and buffering capability of the IOS, they
`provide unsurpassed 110 performance. From 2 to
`32 disk drives can be connected to an I10
`Subsystem for up to 38 gigabytes of total disk
`storage. Typically, DD-49 disks are configured on
`the CRAY X-MPl4 and CRAY X-MP12 and DD-39
`disks are configured on the GRAY X-MP11 .
`
`Effective disk transfer rates can be increased
`further by the use of optlonal disk striping
`techniques. When specified, striping causes
`system software to distribute a single user dataset
`across two to five disk drives, depending on the
`device type, Successive disk blocks are allocated
`
`cyclically across the drives and consecutive blocks
`can thus be accessed in parallel. The resultant I10
`performance improvements are in proportion to the
`number of disk drives used. DD-49disks may be
`striped two or three wide; DD-39 disks may be
`striped two to five wide.
`
`The CRAY X-MP supports three channel types,
`identifiedby their maximum transfer rates: 6
`Mbytelsec, 100 Mbytelsec and 1000 Mbytelsec.
`Depending on the X-MP model, two or four 6-Mbyte
`channels and one to four 100-Mbyte channels are
`connected to each system. The 100-Mbyte
`channels are available for transferring data
`between the I10 Subsystem and central memory
`andlor to the SSD.
`
`Solid-state Storage Device
`
`The optional Solid-state Storage Device (SSD) is a
`very fast random-access device suited for use with
`the CRAY X-MP. The SSD in conjunction with
`multiprocessor architecture allows the
`development of algorithms to solve larger and more
`sophisticated problems in science and engineering.
`
`The SSD is used as a fast-access device for large
`prestaged or intermediate files generated and
`manipulated repetitively by user programs.
`Datasets may be assigned to the SSD by a single
`Cray Operating System (COS) control statement
`without modification of the user program.
`
`Impper
`I I I I G WI ~ I W I I I ~ L I C ;
`Imagesof regions in the M d
`west. (Left) An area west of
`KansasCity, Kansas. bR1ghZ1A
`close-up of the Minnea olislSt,
`Paularea.md(tar rig&
`area southwest of ReTwin
`Citl&.
`
`P
`
`NETAPP, INC. EXHIBIT 1009
`Page 13 of 28
`
`

`
`System performance is significantly enhanced by
`the SSD's exceptionally high transfer rates and
`short data access times: Up to 1024 Mbytes of
`rapid-access MOS memory may be configured on
`an SSD. Transfer rates of 10040 1000 Mbytelsec
`per channel and access times of less than 25
`microseconds are achievable between the SSD
`and an X-MP mainframe. The SSD offers significant
`potential for performance improvement on
`110-bound applications, and thus allows users to
`develop new algorithms that would not otherwise
`be practical with traditional disk 110.
`
`An SSD can also be connected to the I10
`Subsystem. This connection enables data to be
`transferred between the IOS and the SSD directly,
`without passing through central memory.
`
`On the CRAY X-MPl4, support is provided to link the
`SSD to the mainframe via two 1000-Mbyte
`channels. For linkage to the X-MP11 and X-MPl2
`models, one 1000-Mbyte channel is used.
`
`Physical characteristics
`
`The CRAY X-MP is extremely compact; keeping
`wire lengths short minimizes signal propagation
`times. The elegant and compact CRAY X-MP11
`mainframe consists of six vertical columns
`arranged in a 135O arc that occupies 32 square feet
`(3 square meters) of floor space. A CRAY X-MPl2
`model consists of eight vertical columns arranged
`in a 1 80° arc that occupies 43 square feet (4 square
`meters) of floor space. And a CRAY X-MPl4 system
`is composed of 12 vertical columns arranged in a
`270°arc and requires just 64 square feet (6 square
`meters) of floor space.
`
`The accompanying I10 Subsystem is composed of
`four vertical columns in a 90° arc and occupies 24
`square feet (2.3 square meters) of floor space. The
`IOS can be positioned up to 19 feet (5.8 meters)
`from the mainframe.
`
`Scenefrom The LastStar-
`
`clouas. (Credtt Dlg~taI Scenes
`tlonSmby DigitalProductlona,Lo8
`Angeles. Caltforn~a U S A 1985 All
`rightsreservedf
`
`I
`
`NETAPP, INC. EXHIBIT 1009
`Page 14 of 28
`
`

`
`The optional SSD consists of four columns arranged
`in a 900 arc occupying 24 square feet (2.3 square
`meters) and is connected to the mainframe through
`one or two short aerial bridgeways, depending on
`model.
`
`High-speed 16-gate array integrated logic circuits
`are used in the CRAY X-MP CPUs. These logic
`circuits, with typical 300 to 400 picosecond
`propagation delays, are faster and denser than the
`circuitry used in the CRAY-1. CRAY X-MPl4
`
`memory is composed of ECL bipolar circuits; CRAY
`X-MP11 and CRAY X-MPl2 memory is composed of
`static MOS components.
`
`The dense concentration of components requires
`special cooling techniques to overcome the
`accompanying problems of heat dissipation. A
`proven, patented cooling system using liquid
`refrigerant cooling maintains the necessary internal
`system temperature which contributes to high
`system reliability and minimizes the requirement for
`expensive room cooling equipment.
`
`A terrain mapping of the San
`FranciscoBay area developed
`for real-time ernemsncvss-
`sessrngnt.A 7qlgibyt6
`database and a ray-tracing ai-
`gonthrn were used to prepare
`
`'
`1 Characters trorn 'The Adven-
`tures of Andre and Wally 8"
`generated on a CRAY X-MP.
`(Credit @ r $84, Lucasf~lm
`
`NETAPP, INC. EXHIBIT 1009
`Page 15 of 28
`
`

`
`CRAY X-MP software
`
`
`A full range of system and applications software
`compatible with that provided on the CRAY-1
`computer systems is available for the CRAY X-MP
`systems. This software includes the efficient Cray
`Operating System (COS), an auto-vectoring ANSI
`78 Cray FORTRAN compiler, extensive FORTRAN
`and scientific library routines, program and dataset
`management utilities, debug aids, a selection of
`compilers, a powerful Cray assembler (CAL) and a
`wealth of third-party and public-domain application
`codes.
`
`The operating system, the FORTRAN compiler and
`library programs are designed to allow users to take
`advantage of the vectorizing, multiprocessing and
`multitasking features of the CRAY X-MP systems.
`Multitasking is a technique whereby an application
`program can be partitioned into independent tasks
`that can execute in parallel on a multiprocessor
`CRAY X-MP system. Two methods can be used:
`FORTRAN callable subroutines to explicitly define
`and synchronize tasks at the subroutine level, or a
`FORTRAN preprocessor to identify DO loops whose
`independent iterations may be dispatched to
`separate processors. The first method
`(macrotasking) is best surted to programs with
`large tasks running with dedicated processors. The
`second method (microtasking) is beneficial for
`programs with any size tasks running in either a
`dedicated or a production environment.
`
`NETAPP, INC. EXHIBIT 1009
`Page 16 of 28
`
`

`
`Cray Research provides support for the ongoing
`process of converting and maintaining applications
`software on the CRAY X-MP Series. A compre-
`hensive directory of available programs is
`published by the Cray Applications Software
`Library Service.
`
`The above-mentioned software teamed with an IS0
`Level 1 Pascal compiler, a sort package, a C
`compiler and many other software tools and
`products, provides users with the software they
`need to use the CRAY X-MP to its fullest capabilities.
`
`The Cray Operating System efficiently delivers the
`full power of the hardware to both batch and
`interactive users. The operating system, which is
`distributed between central memory and the IOS,
`effectively manages high-speed data transfers
`between the CRAY X-MP and peripherals such as
`disks, SSD and on-line magnetic tapes. Standard
`system software is also offered for interfacing the
`CRAY X-MP Computer System with other vendor's
`operating systems and with networks. This is
`described further under "System Integration". COS
`also includes a variety of utility programs that assist
`in program development and maintenance.
`
`Cray's FORTRAN compiler fully meets the ANSI 78
`standards while offering a high degree of automatic
`scalar and vector optimization within these
`standards. The Cray compiler permits maximum
`portability of programs between different Cray
`systems and accepts many nonstandard constructs
`written for other vendor's compilers. There is no
`need for using nonstandard vector syntax to
`produce vectorized object code. The compiler is
`fully supported by highly optimized FORTRAN and
`scientific library routines for maximum performance
`from the CRAY X-MP Series computers.
`
`The success of the CRAY-I stimulated the
`development of a wide variety of third-party and
`public domain application programs, which are now
`available on CRAY X-MP computers. Major
`applications codes are offered for the CRAY X-MP
`in fields such as computational fluid dynamics,
`mechanical engineering, nuclear safety, circuit
`design, seismic processing, image processing,
`molecular modeling and artificial intelligence.
`
`Two-dimensional flow around
`8. fast-back automabite.The
`Imagewas predueed by solv-
`lnrilthe two-dimensional
`~av~er-stokeseqLatlons.
`(Credt Dorn~erGmbH. West Oermany )
`
`NETAPP, INC. EXHIBIT 1009
`Page 17 of 28
`
`

`
`System integration
`v
`
`
`-&
`
`CRAY X-MP Series computers are designed to be
`connected easily to one or more front-end
`computer systems. Thus, a CRAY X-MP computer
`can be added into an existing configuration so that
`the end user continues to work in a familiar
`computer environment but now has access to a
`considerably greater computational resource. Jobs
`can be submitted from a front-end to the CRAY
`X-MP for processing and results returned to the
`user on the originating front-end or optionally to a
`different front-end. Data can be transferred readily
`between any front-end system and the X-MP, with
`data conversion and reformatting handled
`automatically by software.
`
`Cray Research offers hardware interfaces that
`connect the CRAY X-MP I10 Subsystem to a wide
`variety of front-end equipment, ikluding IBM, CDC,
`DEC, Data General, Sperry and Honeywell.
`Additionally, the I10 Subsystem may be connected
`to one or more Network Systems Corporation
`HYPERchannelTMadapters for those installations
`wishing to configure their CRAY X-MP in a
`high-speed local area network.
`
`Cray Research provides software interface support
`for a variety of front-end systems. Station software
`runs on the front-end system and provides the
`logical connection between other vendors'
`equipment and CRAY X-MP computers. Standard
`Cray software is available for the following: IBM
`MVS and VM, CDC NOS and NOSIBE, DEC
`VAXIVMS, Data General RDOS and AT&T UNIXTM.
`Station software for Sperry and Honeywell
`operating systems is currently available from
`third-party sources.
`
`NETAPP, INC. EXHIBIT 1009
`Page 18 of 28
`
`

`
`Support and maintenance
`
`Customer support
`
`Cray Research has developed a comprehensive
`array of support services to meet customer needs.
`From pre-installation site planning through the life
`of the installation, ongoing on-site engineering and
`system software support is provided. Additional
`assistance is available from technical centers
`throughout the company.
`
`Cray Research provides comprehensive
`documentation and offers customer training on-site
`or at Cray training facilities. Cray Research's
`responsive customer support program results from
`extensive accumulated experience in the
`supercomputer business and from a strong
`customer orientation.
`
`CRAY X-MP reliability and
`maintenance
`
`Cray Research recognizes the n