(12) Ulllted States Patent
`Derrico et al.
`
`(10) Patent N0.:
`(45) Date of Patent:
`
`US 6,948,021 B2
`Sep. 20, 2005
`
`US006948021B2
`
`(54) CLUSTER COMPONENT NETWORK
`APPLIANCE SYSTEM AND METHOD FOR
`ENHANCING FAULT TOLERANCE AND
`H()T.sWAp]l[NG
`
`(75)
`
`,
`Inventors: Joel Brian Derrico, Atlanta, GA (US);
`Pa"! Jonathan Free“ Duluth’ GA(US)
`.
`(73) A551g“°°3 Raceml Systems’ Duluth’ GA (US)
`( * ) Notice:
`Subject to any disclaimer, the term of this
`patent is extended Or adjusted under 35
`U-S.C. 154(1)) by 324 days.
`
`(21) App], No; 09/987,917
`_
`(22) Fllcdi
`(65)
`
`N0“ 16’ 2001
`Prior Publication Data
`
`US 2002/0078290 A1 Jun, 20, 2002
`
`(60)
`
`Related U.S. Application Data
`§(r)c(:)\(r)isional application No. 60/248,834, filed on Nov. 16,
`7
`.............................................. .. G06F 13/00
`Int. Cl.
`(51)
`(52) US. CL ...........U
`710/302; 710/301
`(58) Field of Search ............................... .. 710/301, 302,
`710/72, 304, 100; 713/100; 709/222, 227,
`219, 203, 223; 361/695, 720, 752, 683,
`687; 363/123; 439/92; 307/46, 66; 370/910
`
`(56)
`
`References Cited
`U_S_ PATENT DOCUMENTS
`
`7/1991 Bowling et a1.
`5,033,112 A *
`.......... .. 398/110
`
`5,161,097 A " 11/1992 lkeda ................... .. 363/124
`gaigfsgg $1 * 3233;
`:?“9P‘£‘ 9131-t --1---------- 71(7)/$0/3
`,
`.
`*
`1erre- ouise a.
`5,452,797 B1 *
`9/2002 Konstad ................... .. 361/695
`6,535,944 B1 *
`3/2003 J h
`'
`l.
`.
`710/302
`
`6,591,324 B1 *
`7/2003 (:11:-ii-‘jail.
`710/302
`OTHER PUBLICATIONS
`“Dynamic runtime re—schedu1ing allowing multiple imple-
`mentations of a task for p1atfonn—based designs” by Tin-
`Man Lee; Henkel, J.; Wolf,
`(abstract only).*
`“Redundant arrays of IDE drives” by Sanders, D.A.; Cre-
`maldi, L.A.; Eschenburg, V>; Lawrence, C.N.; Riley, C.;
`Summers, DJ Petravick, D..L. (abstract only).*
`* cited by examiner
`
`Primary ExtImi"er—G0Pa1 0 Ray
`(74) Attorney, Agent, or Fir-m—DLA Piper Rudnick Gray
`Cary US LLP
`ABSTRACT
`(57)
`Packaging a hot—swappable server module server blade in
`_
`a computer network appliance with shared, hot-swappable
`power, network, and management modules to provide highly
`available computer capacity. Distributing power between
`hot—swappable modules using single DC input voltage.
`
`36 Claims, 5 Drawing Sheets
`
` §
`
`
`
`
`EJHVOB3NVld)l3VH
`:2
`%
`=
`8
`|....-....-.....----..\~
`cmssus
`5 I “”“”“‘”
`
`
`, 3§ P-‘--‘-=‘-‘---£1,
`*2‘
`
`
`g
`
`MODULE =~1oa ,.
`
`122(1))
`
`EXHIBIT
`
`ACCELERON
`
`2001
`
`CHASSIS
`mom
`“Z”
`115
`
`

`
`U.S. Patent
`
`Sep. 20, 2005
`
`Sheet 1 of 5
`
`US 6,948,021 B2
`
`w—_/\zmamnofi=2_;_o85_=Wmamsaauaaaaaaauaaaaain.H
`
`38m
`
`mD
`W
`HOT SWAP CONNECTORS
`
`..._..
`
`
`&oEzz83_\I\525%._
`?!‘l“"“““2‘2““‘1
`
`E2:
`
`.\uE58:Eu
`
`E8_
`
`Ewe
`
`\nm5:8::8
`
`GE_
`
`33
`
`
`
`
`
`
`
`

`
`U.S. Patent
`
`Sep. 20, 2005
`
`Sheet 2 of 5
`
`US 6,948,021 B2
`
`35-5:
`
`N3E98228
`
`mamas...
`
`558__Nas
`
`9.e_._.__£
`
`am.
`
`
`
`mesgfi.34:9
`
`
`

`
`U.S. Patent
`
`Sep. 20, 2005
`
`Sheet 3 0f5
`
`US 6,948,021 B2
`
`THUMB
`SCREW
`
`1 10
`/J
`
`
`
`.306
`
`I2C/3.3V/GND
`LAN1/LAN2/LAN3/LAN4/LAN5
`LAN6/LAN7/LAN8
`31 2
`
`ouAo W
`ETHERNET
`303
`PHY
`
`
`
`
` 8—PORT
`
`
`
`
`
`302
`
`ETHERNET
`SWTTCH
`
`
`
`
`
`

`
`U.S. Patent
`
`Sep. 20, 2005
`
`Sheet 4 of5
`
`US 6,948,021 B2
`
`
`
`FAST HHERNEF
`
`FAST ETHERNET
`
`/
`
`606
`
`FlG.6
`
`/
`
`610
`
`700
`x-/
`
`
`
`__
`
`
`REDUNDANT ROUTERS f
`é:'—_1_| /
`
`702
`
`704
`
`CLUSTER
`NETWORK
`
`REDUNDANT swrrcH£s
`
`APPLIANCES
`
`

`
`U.S. Patent
`
`Sep. 20, 2005
`
`Sheet 5 of5
`
`Us 6,948,021 B2
`
`

`
`US 6,948,021 B2
`
`1
`CLUSTER COMPONENT NETWORK
`APPLIANCE SYSTEM AND METHOD FOR
`ENHANCING FAULT TOLERANCE AND
`HOT-SWAPPING
`
`This application claims priority from U.S. Provisional
`Application Ser. No. 60/248,834, filed Nov. 16, 2000. The
`entirety of that provisional application is incorporated herein
`by reference.
`
`BACKGROUND OF THE INVENTION
`
`1. Field of the Invention
`
`This invention generally relates to fault tolerant computer
`systems and, more specifically, to a system and method for
`enhancing fault
`tolerance and hot swapping in computer
`systems.
`2. Related Art
`
`Computer systems such as file servers and storage servers
`in computer networks are relied upon by large numbers of
`users. When a file server or storage server is out of operation,
`many users are inconvenienced. Thus, technology has been
`developed which supports maintenance and service of com-
`puter systems while they remain operational. One part of
`maintenance and service includes the replacement of com-
`ponents in the computer systems. “Hot swap” technology
`allows the replacement of components without turning off
`the power or resetting the computer system as a whole.
`Hot swap enables the insertion and/or removal of com-
`ponents in a computer system while it
`is still active or
`operational. In systems that do not support hot swapping of
`components, each process of component insertion and/or
`removal requires a complete shutdown of the entire system
`to prevent damage to other components or to the system. In
`time critical systems such as communications systems,
`system downtime is both a financial problem as well as a
`service quality problem. That is, any downtime means a
`financial loss and disconnection of service to active lines.
`
`is it requires
`A drawback of hot swapping, however,
`trained personnel to insert and/or remove components from
`a computer system to minimize damages caused by pitting
`connectors of the components against connectors of the
`computer system. Another drawback is electrical noise
`which can adversely affect the performance of the computer
`system. The noise is caused by the change in current at the
`instance when connection is made between power pins of a
`component and corresponding elements of the computer
`system. The result
`is voltage transients in the computer
`system backplane that may cause loss of data, incorrect
`program execution and damage to delicate hardware com-
`ponents.
`there is a need for a system and method for
`Thus,
`enhancing fault
`tolerance and hot swapping in computer
`systems so as to reduce both the downtime of computer
`systems and the use of trained personnel to repair and/or
`maintain computer systems.
`SUMMARY OF THE INVENTION
`
`The present invention is directed to a hot swapping
`computer network appliance operating in mission critical
`applications where any computer downtime can result in
`serious consequences. The computer network appliance
`comprises a hot-swappable CPU module, a hot-swappable
`power module, a hot-swappable ethemet switch module and
`a backplane board having a plurality of hot swap mating
`connectors. Each of the CPU module, power module and
`
`10
`
`15
`
`20
`
`30
`
`40
`
`45
`
`50
`
`55
`
`60
`
`2
`cthcrnet switch module includes a hot swap connector for
`connecting with a specific hot swap mating connector of the
`backplane board. The computer network appliance further
`comprises a chassis providing physical support for the
`modules and the backplane board. The chassis comprises
`caddies providing air flow in the chassis. The chassis further
`comprises bays and slot guides to facilitate mounting and
`removal of the modules and to ensure proper alignment
`between the hot swap connectors of the modules and the hot
`swap mating connectors of the backplane board. The com-
`puter network appliance comprises a power connector and a
`data input/output connector, both of which remain connected
`during mounting or removal of the modules.
`Each of the hot swap connectors of the modules com-
`prises pin connections arranged in a specific pattern. The
`pins include ground pins, power pins and signal pins. The
`ground pins of a hot swap connector are connected first to
`corresponding ground elements of a hot swap mating
`connector, and the signal pins of the hot swap connector are
`connected last to corresponding signal elements of the hot
`swap mating connector so as to reduce brown outs in the
`computer network appliance.
`The CPU module of the invention operates as a stand
`alone computer. The CPU module comprises hardware
`BIOS for configuring the CPU module and instructing a
`network attached storage (NAS) to locate an operating
`system (OS) from which to boot. The CPU module is
`configured to boot remotely from an OS located in the NAS
`without user intervention. This remote booting ability of the
`CPU module allows the CPU module to run different types
`of operating systems without the need for a local hard disk
`drive (HDD), which increases the mean time between failure
`(MTBF) and decreases the mean time to repair (M'I'I'R) of
`the computer network appliance.
`The invention further provides that each of the hot swap
`connectors of the modules includes an ethernet connection
`providing communications to all modules attached to the
`backplane board.
`The power module of the invention comprises dual
`DC—DC converters that perform direct conversion of a
`facility DC voltage to voltages required for normal operation
`in the modules. Features of the DC—DC converters include:
`allowing the modules in the computer network appliance to
`accept DC power directly from a battery backup source
`without requiring power inverters; higher MTBF than a
`typical switched power supply; use less power and generate
`less heat than a typical switched power supply; and provide
`better efficieney than a typical switched power supply in
`converting an input voltage to desired operational voltages
`of the modules.
`
`DESCRIPTION OF THE FIGURES
`
`FIG. 1 is an illustration of a cluster computer network
`appliance arranged on a chassis in accordance with an
`embodiment of the invention;
`FIG. 2 is a block diagram of a CPU module in accordance
`with an embodiment of the invention;
`FIG. 3 illustrates an integrated ethemet switch module in
`accordance with an embodiment of the invention;
`FIG. 4 illustrates a power module in accordance with an
`embodiment of the invention;
`FIG. 5 illustrates a microcontroller module in accordance
`with an embodiment of the invention;
`FIG. 6 illustrates an integration of a cluster computer
`network appliance, data storage device and standard intemet
`access hardware;
`
`

`
`US 6,948,021 B2
`
`3
`FIG. 7 illustrates a computer system utilizing multiple
`network appliances, redundant storage and internet access
`points; and
`FIG. 8 illustrates a computer system providing path
`redundancy and equipment redundancy to achieve high
`availability.
`DETAILED DESCRIPTION OF THE
`PREFERRED EMBODIMENTS
`
`The following detailed description presents a description
`of certain embodiments of the present invention. However,
`the present invention can be embodied in different ways as
`defined by the claims. In this description, reference is made
`to the drawings wherein like parts are designated with like
`numerals throughout.
`FIG. 1 is an illustration of a cluster computer network
`appliance 100 arranged on a chassis 150 in accordance with
`an embodiment of the invention. The cluster computer
`network appliance 100 includes a plurality of CPU modules
`102(a)—102(e), a passive backplane board 104 with hot swap
`mating connectors 124(a)-124(i), a power module 106, a
`microcontroller module 108, an ethemet switch module 110,
`power/ground connectors 112 and ethemet connectors 114.
`The cluster computer network appliance 100 fits in a 1.75"
`tall (1RU) metal chassis that fits in a standard 19“ rack. The
`chassis 150 includes a fold down front panel 116 and
`supports the modules and backplane board of the invention.
`The chassis has five bays accessed via the front for inserting
`the CPU modules 102(a)—102(e) and three bays accessed via
`the rear 118 for inserting one each of the power module 106,
`the ethemet switch module 110 and the microcontroller
`module 108. Each module resides in a caddy 152 of the
`chassis such that when the module is inserted into the chassis
`the caddy ensures that the hot swap connectors are aligned.
`Each of the hot swap connectors used in the modules is
`specific to corresponding hot swap mating connectors in the
`backplane board. For normal operation, the chassis must be
`equipped with at least one CPU module, the power module
`and the ethernet switch module.
`
`The power/ground connectors 112 provide physical con-
`nection for power to the chassis. The ethernet connectors
`114 provide data input/output (I/O) to and from the chassis.
`Power is connected such that should the power module 106
`fails, it may be replaced without disconnecting the actual
`power cabling inside the computer network appliance,
`which saves time and reduces complexity. Similarly, a failed
`ethernet switch module 110 may be replaced without dis-
`connecting any of the power or data cables. Heat generated
`by active elements in each of the modules is dissipated using
`forced air flow from the front to the rear of the chassis using
`a push—pull method. Fans 120(a)—120(e) are provided for
`each CPU module providing a 1:1 ratio of fan to bay and
`positioned near the front panel 116 of the chassis to push
`outside air through the chassis. In the rear of the chassis,
`multiple fans 122(a)—122(d) are mated to the back of both
`the power module 106 and the ethemet switch module 110
`to draw heated air out of the chassis.
`
`Each module is designed to be hot swapped from the
`chassis such that there is no need for on/ofi switches on
`either the chassis or the modules. The passive backplane
`board 104 is equipped with hot swap mating connectors
`l24(a)—124(i) for each of the modules to be inserted into the
`computer network appliance. The chassis is installed and
`wired for power and data I/O such that power is supplied
`directly to a module as soon as it is inserted.
`In order to avoid chassis power drains (brown outs)
`caused by instantaneous power short
`to ground through
`
`4
`uncharged board capacitance, the hot swap connectors of the
`modules (shown in FIG. 1 mated to corresponding hot swap
`mating connectors 124(a)—124(z')) are designed to make pin
`connections in a specific pattern to avoid power drains. Each
`hot swap connector of a module comprises groups of pins
`(ground pins, pre—charge power pins, power pins and signal
`pins) of dilferent length that allow the pins to make con-
`nections in a prearranged pattern. The first group of pins to
`make contact with corresponding elements in a mating
`connector on the passive backplane board is the ground pins
`(chassis ground and common ground). The next group of
`pins to make contact with corresponding elements in the
`mating connector is the pre—charge power pins. The pre-
`eharge power pins connect to a power plane on a printed
`circuit board (PCB) through resistors to limit the flow of
`current while pre-charging the capacitance on the PCB. The
`next group of pins to make contact with corresponding
`elements in the mating connector is the power pins. The last
`group of pins to make contact with corresponding elements
`in the mating connector is the signal pins. By connecting the
`pins in this fashion, the computer network appliance of the
`invention avoids brown outs, arching across pins and false
`grounds that can damage components in the computer
`network appliance.
`FIG. 2 is a block diagram of a CPU module 102 in
`accordance with an embodiment of the invention. The CPU
`modules 102(a)—102(e) do not have moving parts and com-
`ponents defining a direct user interface. Each CPU module
`comprises a microprocessor 202, memory module 204, bus
`management chipset including a Northbridge chip 206(a)
`and a Southbridge chip 206(b), an ethemet interface chip
`208, hardware BIOS 210 and a hot swap connector 212
`mounted on a PCB. A PCI bus header is included for
`development and debugging purposes. Each CPU module
`functions as a stand alone computer.
`The hardware BIOS 210 configures the CPU module for
`normal operation and instructs the ethemet interface chip
`208 where to look on an NAS for the OS from which to boot.
`This remote boot capability of the CPU module enables the
`system administrator to direct the module to boot from a
`specific OS stored in a predetermined location in the NAS.
`This, in turn, enables the CPU modules in a network to run
`dilferent types of OS (e.g., Unix, BSD, Linux, and Solaris)
`and removes the necessity for a local hard disk drive (HDD).
`Under management software control, a CPU module may be
`booted with an OS along with an “image” including several
`pre—installed applications (user defined quantity) stored in an
`NAS or a storage area network (SAN). This diskless booting
`of the CPU module allows the CPU to run difierent OS’s and
`applications at dilIerent times. For example, a CPU module
`may be booted with a first OS and a first set of applications
`at one time and with a second OS and a second set of
`applications at another time. In another embodiment of the
`invention, ditferent CPU modules operating in the same
`chassis may be booted with dilferent OS’s and diiferent
`applications. In yet another embodiment of the invention,
`the same OS, applications and user data of one CPU module
`may be installed in another CPU module so as to provide for
`hot swapping of a failed CPU module or for installation of
`a redundant CPU module. Removal of the local I IDD is a
`feature of the invention that allows hot swapping of the CPU
`modules Without rebooting the system.
`Once the OS is loaded on the CPU module and is
`
`operational, the health of the CPU module can be monitored
`using an 12C bus 214 that provides status information about
`the CPU module to the optional microcontroller module 108
`as shown in FIG. 1. Along with information such as CPU
`
`10
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`60
`
`

`
`US 6,948,021 B2
`
`5
`temperature, fan RPM and voltage levels, a watchdog timer
`is provided in the hardware design to provide a way of
`determining if the OS is unstable or has crashed. If the OS
`is unstable or has crashed, then the microcontroller module
`108 has the ability to remotely reset the CPU module 102
`and log the failure. Such a reset can be configured to take
`place automatically or manually under the control of the
`administrator.
`
`The CPU module 102 is configured to remotely boot
`without user intervention to allow for the removal of unnec-
`essary user interface hardware such as video and standard
`I/O chipsets. The removal of this hardware and the HDD as
`described above reduces the complexity of the design and
`increases the mean time between failure (MTBF) of the
`hardware while simultaneously lowering the pan count
`(cost) and power consumption of the module. In addition,
`the network mean time to repair (MTTR) is lowered through
`the use of the hot swap design and remote OS boot capability
`of the module because a failed unit can be removed and
`replaced rather easily and no user interaction is necessitated
`once a CPU module has been inserted into the chassis. A
`CPU module can be inserted in any of the bays in the front
`of the chassis.
`
`Communications to and from each module is made using
`a standard fast ethernet connection rather than a complicated
`external bus structure. That is, a single ethernet connection
`via the hot swap connection of each module allows the
`module to communicate with other modules connected in
`the computer network appliance. The pinout of the hot swap
`connection is limited to ethemet signal path pins, dedicated
`power and ground pins, and an I2C bus for out-of—band
`monitoring of the health of the CPU module and remote
`rebooting of the microprocessor if the OS is determined to
`be unstable or have crashed. The process of out-of-band
`monitoring and control of the CPU module is mediated by
`the microcontroller module 108. In-band monitoring pro-
`cesses are used to load applications and data and are
`controlled by direct communications between the manage-
`ment software and the CPU module microprocessor 202.
`As stated above, each CPU module includes a PCI bus
`header that is provided for debugging and test purposes only.
`If a CPU module is suspected of being faulty, then it can be
`removed and plugged into a test fixture that provides video,
`keyboard, mouse, and HDD access through a cable connec-
`tion to the PCI bus header. Power and ethernet I/O are
`accessed through the hot swap connector 212.
`In this
`fashion, the CPU module combined with the test fixture
`emulates a desktop computer and the CPU module can be
`debugged accordingly.
`Since only a limited number of modules make up the
`configuration of the computer network appliance, an end
`user’s spare parts inventory is greatly reduced and configu-
`ration variability is low. Each module can be easily replaced
`and does not require a skilled person, and no spare parts need
`to be inventoried on—site and can be shipped overnight from
`the supplier. As a result, the computer network appliance
`MTTR is greatly reduced through the ease of module
`replacement and the MTBF is high through the simplified
`design of the CPU module.
`Abyproduct of using standard fast ethemet as the method
`of signal I/O for network communications is that heteroge-
`neous CPU modules having dilferent CPU speeds, memory
`space and bus chipsets may be mounted in the same chassis
`without affecting the operation of any other CPU module.
`Specifically, dilferent generations of CPU modules may
`operate in the same chassis without requiring an update of
`existing modules.
`
`10
`
`15
`
`20
`
`25
`
`30
`
`40
`
`50
`
`55
`
`60
`
`6
`FIG. 3 illustrates an integrated cthernct switch module
`110 in accordance with an embodiment of the invention. The
`ethernet switch module 110 comprises an ethemet switch
`302, EEPROM 304, buffer memory 306, ethemet transceiv-
`ers 308 and a hot swap connector 312 all mounted on a
`single PCB. In the preferred embodiment of the invention,
`the ethemet switch 302 is an unmanaged 8-port ethernet
`switch. The ethemet switch module 110 operates as a traflic
`cop for data communications in the computer network
`appliance, allowing each CPU module to communicate with
`other CPU modules in the same chassis. The ethernet switch
`module 110 further includes cooling fans 314 mounted to the
`rear of the PCB; the cooling fans 314 operate to draw heated
`air out of the chassis.
`Once the ethemet switch module 110 is inserted into the
`rear of the chassis 150, it connects to the passive backplane
`board 104 via the hot swap connector 1240;) to derive
`power, establish ground and establish all ethemet oonnec-
`tions within the computer network appliance. The ethemet
`switch module 110 is secured to the chassis using thumb
`screws 316 mounted on the 1RU panel. The ethernet switch
`module 110 is designed such that if a failure occurs, then the
`module can be quickly replaced without disconnecting any
`signal or power cables, thus attaining a low MTTR and
`allowing the use of less skilled maintenance personnel.
`A function of the ethernet switch module 110 is to filter
`out
`inappropriate signal
`traffic so as to limit collisions
`caused by signal traflic in the computer network appliance.
`Communications between CPU modules in the same chassis
`occur without disruption to signal traific between other CPU
`modules and the network and does not add to the overall
`level of network traflic. As a result, the efliciency of the
`signal bandwidth is increased without sacrificing perfor-
`mance or network cost.
`
`the application servers are not signal I/0
`Moreover,
`limited and this allows all network traflic with the computer
`network appliance to be multiplexed over a switched fast
`ethernet (up to three connections) and does not require a
`direct ethemet connection between each CPU module and
`other modules in the computer network appliance.
`Consequently, the amount of external wiring required to
`connect the CPU modules to the computer network appli-
`ance is greatly reduced by integrating the switch into the
`design of the network server appliance. The use of multiple
`switched ethernet connections permits the computer net-
`work appliance to operate with different topologies or soft-
`ware configurations without additional hardware. Since five
`of the eight switched ethernet ports are dedicated to the five
`CPU module connections, a typical network connection
`would dedicate the remaining three ports to a mixture of
`NAS, network data I/O and an in-band appliance manage-
`ment channel.
`
`FIG. 4 illustrates the power module 106 in accordance
`with an embodiment of the invention. The power module
`106 comprises dual DC—DC converters 402 mounted on a
`1RU panel of a printed circuit board, a hot swap connector
`404, cooling fans 406 and thumb screws 410. The DC—DC
`converters 402 perform direct conversion of a facility DC
`voltage (48V) to voltages required for normal operation of
`the modules that make up the computer network appliance.
`The hot swap connector 404 operates to draw facility
`voltage and supply operational voltages to the passive
`backplane board 104. The cooling fans 406 operate to draw
`air out of the chassis across the cooling fins of the heat sinks
`on the DC—DC converters 402.
`
`Once the power module 106 is inserted into the rear of the
`chassis, it connects to the passive backplane board 104 via
`
`

`
`US 6,948,021 B2
`
`7
`the hot swap connector 1240‘) to derive facility DC power,
`establish ground and generate all operational voltages used
`by the other modules in the computer network appliance.
`The hot swap connector 404 includes an I2C bus 408 that is
`used to monitor the health of the power module. The power
`module is secured to the chassis using thumb screws 410.
`The power module is designed such that if a failure occurs,
`then the power module can be quickly replaced without
`disconnecting any power cables in the computer network
`appliance. As a result, the MTTR is lowered and less skilled
`maintenance personnel may be used.
`In typical commercial electronics designs, the portion of
`the design having the lowest MTBF is the svwtched power
`supply. A common practice to overcome this drawback to
`increase the MTBF is to include redundant power supplies
`in a design so that if one power supply fails,
`then the
`redundant unit automatically backs it up. Because typical
`commercial electronics power supplies run on alternating
`current (AC) power, the battery backup system must convert
`power from its normal direct current (DC) state to AC using
`power inverters. Power inverters, however, are ineflicient
`because of their
`fundamental operation (generator
`hysteresis) in converting power from DC to AC. Power
`inverters are also expensive and do not scale well should
`additional power capacity is required. Thus, designing an
`appliance using DC—DC converters instead of a switched
`AC power supply would allow an appliance to accept DC
`power directly from the battery backup source and negate
`the need for power inverters. That is, an appliance using
`DC—DC converters alone would not require use of experi-
`sive power inverters and increase the overall efliciency of
`the battery backup system. Another compelling reason to use
`DC DC converters in a commercial electronics design is
`the MTBF of a DC—DC converter is much greater than that
`of a switched power supply. A DC—DC converter is also
`more eflieient than a power supply in converting the input
`voltage to the desired operational voltages, which means
`that the appliance will use less power and generate less heat
`than a power supply.
`FIG. 5 illustrates a microcontroller module 108 in accor-
`dance with an embodiment of the invention. The microcon-
`troller module 108 is optional and is not required for normal
`operation; the microcontroller module 108 is employed for
`monitoring out-of-band communications and for controlling
`the computer network appliance modules. The microcon-
`troller module 108 comprises a stand-alone microprocessor
`502 running an embedded OS, flash RAM 504 including the
`OS and application software, a dedicated ethemet chip 506
`providing connection to the network, an I2C bus chipset
`508, a hot swap connector 510 and thumb screws 512.
`Once the microcontroller module is inserted into the rear
`of the chassis, it connects to the passive backplane board 104
`via the hot swap connector 124(g) to derive power, establish
`ground and establish ethemet connection with the computer
`network appliance. The microcontroller module 108 is
`secured to the chassis using thumb screws 512 mounted on
`a 1RU at the rear of the module. The microcontroller module
`is designed such that if a failure occurs, then the microcon-
`troller module can be quickly replaced without disconnect-
`ing any signal or power cables so as to attain a low MTTR
`and to use less skilled maintenance personnel.
`The microcontroller module uses a dedicated ethernet
`path separate from the network data 1/0 to remotely poll the
`health of the power module 106, the ethemet switch module
`108 and the CPU modules 102(a)—102(e). The microcon-
`troller module communicates with other modules using an
`I2C bus that gathers status information, logs the results and
`
`8
`provides the log to the management software either actively
`(should a failure is detected) or as part of a routine poll. The
`microcontroller module 108 also gathers information relat-
`ing to the voltage levels, CPU temperatures, fan RPMs and
`CPU module OS stability. In addition, the microcontroller
`module has the ability to perform a remote reset of a CPU
`module if the OS of the module is determined to be unstable
`or have crashed. If the integrated ethernet switch fails, then
`the dedicated ethemet path may still be able to pinpoint the
`failure and dilferentiate the failure of the switch from an
`
`overall failure of the chassis. The dedicated ethemet path
`further informs the system administrator of the failure so as
`to facilitate a timely fix of the switch or a module on the
`computer network appliance.
`
`FIG. 6 illustrates a system 600 integrating a computer
`network appliance, a data storage device and standard inter-
`net access hardware. The system 600 comprises a router 602,
`a computer network appliance 604 connected to the router
`602 via a fast ethemet connection 606, network database 608
`connected to the computer network appliance 604 via a fast
`ethernet connection 610, and intemet backbone 612. Data
`switching is performed in the computer network appliance
`604. This simplistic representation provides a framework for
`more sophisticated forms of clustering configurations based
`upon specific design criteria, such as availability and fault
`tolerance.
`
`FIG. 7 illustrates a system 700 integrating multiple com-
`puter network applianoes, a storage device and redundant
`internet access hardware. The system 700 comprises a
`plurality of routers 702, a plurality of redundant switches
`704, a plurality or cluster of computer network appliances
`706, NAS 708 and intemet backbone 710. Routers 702,
`computer network appliances 706 and NAS 708 are con-
`nected to redundant switches 704 by fast ethernet connection
`712. Afeature of system 700 is the system layer remains flat
`in that access to the routers, network appliances and NAS
`are all wired through the redundant switches 704. The
`system 700 provides a simple and easy to install/maintain
`framework for redundant network cabling by minimizing the
`amount of equipment external to the cluster of computer
`network appliances. In order to handle the increased traflic
`associated with the large number of servers in the cluster of
`computer network appliances, redundant gigabit ethemet
`paths 714 are introduced to connect intemet backbone 710
`and redundant switches 704 to redundant routers 702 as
`illustrated in FIG. 7.
`
`Alternate forms of configurations can be generated to add
`other requirements to the system such as high availability
`and database security. FIG. 8 illustrates a system 800 pro-
`viding path redundancy and equipment
`redundancy to
`achieve high availability. The system 800 comprises 800 a
`plurality of routers 802, a plurality of redundant switches
`804, a plurality of cluster computer network appliances 806,
`a firewall 808, NAS 810 and internet backbone 812. Redun-
`dant switches 804 and firewall 808 are connected to cluster
`computer network appliances 806 by fast ethernet connec-
`tion 814. Firewall 808 secures NAS 810 from direct access
`of internet connection by accepting only secure connections.
`The increased traffic associated with the large number of
`servers in the cluster of computer network appliances is
`addressed by introducing redundant gigabit ethernet paths
`816 as the front-end connection between internet backbone
`812 and redundant routers 802 and between redundant
`routers 802 and redundant switches 804, and as the back-end
`connection between firewall 808 and NAS 810.
`
`10
`
`15
`
`20
`
`25
`
`30
`
`40
`
`45
`
`50
`
`60
`
`65
`
`

`
`US 6,948,021 B2
`
`9
`
`What is claimed is:
`1. A computer network appliance, comprising:
`a plurality of hot-swappable CPU modules, wherein each
`CPU module is a stand-alone independently-
`functioning computer;
`a hot-swappable power module;
`a hot-swappable ethernet switch module; and
`a backplane board having a plurality of hot swap mating
`connectors,
`wherein the at least one backplane board interconnects
`each of the CPU modules with the at least one power
`module and the at least one ethernet switch module,
`such that the at least one power module and the at least
`one ethernet switch module can be used as a shared
`resource by the plurality of CPU module