PRODUCT GUIDESBCJust what is a blade, anyway?By J. Eric GulliksenThere is a tremendous amountof ambiguity surrounding theterm “blade,” particularly inthe embedded space. It has become amarketing buzzword used to describea variety of different product types,which has created confusion rather th<strong>and</strong>ifferentiation in the marketplace. Thisarticle presents VDC’s definition of theterm, the logic behind this definition,<strong>and</strong> the ways in which we differentiatebetween blades <strong>and</strong> other embeddedboard types.BeginningsVDC first encountered the term bladeduring the course of research for ourfirst report on Switch Fabrics <strong>and</strong> High-Speed Serial Interconnects, publishedin November of 2001. There appearedto be an intimate association betweenthese interconnect technologies <strong>and</strong> theterm blade. However, there was no cleardefinition for blades, blade architectures,or blade-based systems <strong>and</strong>, as the termbecame more widely used, this associationwith fabric technology started to becomeblurred. Definitions for blade foundin various glossaries on the Web couldbe applied to Single Board Computers(SBCs) in general. We began to ask engineering<strong>and</strong> marketing professionals inboth the embedded board <strong>and</strong> enterprisecommunities for their definitions in anattempt to arrive at a consensus. Some ofthe divergent responses we received fromthe embedded industry included:■ Just another, sexier word for board■ A board with some sheet metalwrapped around it■ A single board computer that has beenoptimized as a server■ The combination of a carrier board<strong>and</strong> a PMC card■ A PICMG 2.16 SBC■ An expansion card that plugs into amotherboard■ Another name for a 1U pizza boxserverHowever, many of the other responsesincluded an element of commonality, inthat they did cite the use of high-speedserial or switch fabric interconnects as theprimary means of interboard data communication.Discrepancies within thisgroup were primarily related to the use ofshared, parallel buses as additional meansof interboard communication, or to functionaltypes.The Enterprise space showed a greaterdegree of clarity, although most Enterpriseclassblades were of proprietary architecturesor form factors. Here, the datacommunication means between bladeswas limited to high-speed serial or fabricinterconnects, with little differentiationbetween functional types.The VDC definitionKeeping these considerations in mind,VDC developed the following workingdefinition for an embedded componentlevelblade.An embedded component-level blade is acomputer board with the following characteristics:1. It is designed to be inserted, usuallyvertically, in a slot in a card cage orchassis mounted on a rack.2. It connects to a passive backplane,<strong>and</strong> communicates data to other boardlevel components in the immediatehost system only via a switch fabric orother high-speed serial interconnect.Any shared, multidrop, parallel databus that may be present is local tothe blade <strong>and</strong> is not carried to thebackplane.We therefore base our definition on thestructure of the interconnect architecture<strong>and</strong> do not limit it in any way by functionaltype, application, local bus, formfactor, feature set, or any particular highspeedserial or fabric interconnect technologyor topology.VDC presented this definition to severalof the individuals that we had previouslyinterviewed, most of whom agreed thatit made sense <strong>and</strong> provided much-neededclarity <strong>and</strong> differentiation betweenblades <strong>and</strong> boards using parallel busesas an interboard communication means.We then presented the definition verballyat the Bus & Board Conferencein January of 2002, <strong>and</strong> it has now beenaccepted by a majority of the embeddedboard industry.(Note that certain blade servers, whichinclude 1U, 2U, <strong>and</strong> 4U devices, do notcomply with provision 1 of this definition<strong>and</strong> may communicate with other bladeservers via fiber or cable, without a backplane.We consider these to be systemlevel,not component-level, devices.)SpecificationsOf the various open st<strong>and</strong>ards in existenceto date, only <strong>AdvancedTCA</strong> (PICMG 3.x)comprises a true blade specification.CompactTCA will also be a blade specification.Other st<strong>and</strong>ards such as PICMG2.16 <strong>and</strong> 2.17 have provisions that allow,but do not m<strong>and</strong>ate, blade configurationsunder the VDC definition. A <strong>CompactPCI</strong>single board computer example may helpto clarify this gray area. Note that we willuse the term switch fabric generically toinclude both fabrics <strong>and</strong> other high-speedserial interconnect means that may not befabrics. In addition, the VDC definitionallows for other functional types (such asI/O <strong>and</strong> mass storage) to be substitutedfor SBCs, <strong>and</strong> allows for PICMG 2.16<strong>and</strong> Ethernet Fabric to be replaced byother specifications <strong>and</strong> technologies, asappropriate. Thus, 2.16 may be replacedby 2.18 <strong>and</strong> Ethernet Fabric by RapidIO,for example.”Traditional” <strong>CompactPCI</strong> SBCs do notinclude switch fabric access. Therefore,these cannot be blades under the VDCdefinition, <strong>and</strong> there is no ambiguity.PICMG 2.16-compliant SBCs do includeEthernet Fabric access via the P0 connector.In other words, these are fabricenabled.Fabric-enabled SBCs may ormay not be blades, depending on theirconfiguration:■ If these SBCs carry both the sharedPCI bus <strong>and</strong> the Ethernet Fabric to thebackplane as is permitted by the specification,these are not blades. We callthese nonblade fabric-enabled SBCs.These configurations allow backward48 / <strong>CompactPCI</strong> <strong>and</strong> <strong>AdvancedTCA</strong> <strong>Systems</strong> / June 2005
compatibility with legacy backplanes<strong>and</strong> systems.■ If, as is also allowed under thePICMG 2.16 specification, the sharedPCI bus is not carried to the backplane,these SBCs are blades underthe VDC definition.What if a fabric-enabled PICMG 2.16SBC is used with a backplane that doesnot have the capability of connecting toor carrying the PCI bus between boards?Does this make the SBC a blade? No. Inthis case, the board is still a fabric-enabledSBC, but it is being used as a blade.Recent study findings on<strong>CompactPCI</strong> SBCs <strong>and</strong>CPU bladesVDC’s newly published report, MerchantComputer Boards for Embedded/RealTime Applications Market IntelligenceProgram, 2004: Volume V: Overview,indicates that, to date <strong>and</strong> other than<strong>AdvancedTCA</strong>, the only st<strong>and</strong>ards-basedblades available are of the <strong>CompactPCI</strong>local bus architecture <strong>and</strong> form factor.Table 1 shows the dollar volume shipmentshares of <strong>CompactPCI</strong> SBCs, in 2004,Configuration“Traditional” <strong>CompactPCI</strong>Fabric-enabled nonblade SBCs<strong>CompactPCI</strong> CPU bladessegmented into the three configurationsmentioned earlier.As a whole, shipment shares of fabricenabledSBCs, including blades, areexpected to continually increase. Therelationship between shares of blade <strong>and</strong>nonblade fabric-enabled configurationsis, however, projected to remain relativelyconstant until the CompactTCA specificationbecomes finalized. Ultimately,shipments of <strong>CompactPCI</strong> CPU bladesare expected to overtake those of nonbladefabric-enabled SBCs, with thelatter becoming relegated to a transitionarchitecture.J. Eric Gulliksen been with VDCsince October of 1999 <strong>and</strong> is currentlypractice <strong>and</strong> project director for theEmbedded Hardware discipline, whichShipment shares by percent of $ volumeTable 143 percent44 percent13 percentincludes Merchant Computer Boards <strong>and</strong>Integrated <strong>Systems</strong> for Embedded <strong>and</strong>Real-Time Applications. He holds BSEE<strong>and</strong> MMgS&E degrees from WPI, <strong>and</strong>an MBA from Clark University. Eric hasbeen awarded 17 US Patents, <strong>and</strong> hasinternational field experience in22 countries.For further information, contactEric at:Venture Development CorporationOne Apple Hill DriveSte. 206, Box 8190Natick, MA 01760Tel: 508-653-9000E-mail: ericg@vdc-corp.comWebsite: www.vdc-corp.comRSC# 49 @www.compactpci-systems.com/rsc<strong>CompactPCI</strong> <strong>and</strong> <strong>AdvancedTCA</strong> <strong>Systems</strong> / June 2005 / 49