HP NonStop Gets Bladed
Copyright 2008 Illuminata, Inc.TMHP NonStop Gets BladedProduct InsightGordon Haff25 June 2008Licensed toHewlett-Packard Companyfor web posting. Do notreproduce. All opinions andconclusions hereinrepresent the independentperspective of Illuminataand its analysts.In merging the Integrity NonStop line of fault tolerant servers into its blade serverproducts, HP is bringing together a long-running product success with a relativelynew breakout.The NonStop architecture was developed at Tandem Computers, which wasfounded in 1974 by James Treybig and other Hewlett-Packard engineers. In 1997,Compaq bought, but never really absorbed, the company. When HP subsequentlypurchased Compaq, it acquired a lineup that was an integral component of many ofthe world’s most critical computing infrastructures—which is to sayinfrastructures that move a lot of money around. Like Compaq, HP for a timeseemed uncertain what to do with this marvelous asset other than to leave itmostly alone. However, over the past few years, HP has both modernized NonStopand leveraged its technology for a business intelligence play called HP Neoview.Blades are a more recent development; as a technology approach they took a whileto really hit their stride. In HP’s case, it was the BladeSystem c-Class blade formfactor and chassis design—introduced in 2006—that was the takeoff point; HP nowholds the market share lead over the prior#1 vendor, IBM. HP’s blade approach, likeDell’s and IBM’s, emphasizes theintegration of processing, networking,and (in some cases) storage into a singlechassis—in HP parlance, an “AdaptiveInfrastructure in a Box.”When NonStop moved from its MIPSprocessor-based architecture to Itanium in2005, it made a dramatic shift away froma design and components that werelargely unique to ones that highlyleveraged HP’s other Integrity products.1But this latest marriage of the old and thenew takes NonStop’s mainstreaming tothe next level.1HP Integrity is HP’s Itanium-based lineup that primarily runs HP-UX but, in manycases, also supports OpenVMS, Linux, and Windows. See our The 'I' is for Integrity. Byamusing coincidence, the Integrity brand name came from Tandem which had originallyused it for a separate line of FT systems that ran a Unix variant rather than theNonStop Kernel, or NSK (né Guardian) OS that ran on its primary lineup.*ECF9DA05ADEB7A6Illuminata, Inc. 4 Water Street Nashua, NH 03060 603.598.0099 603.598.0198 www.illuminata.com
2From Tandem to IntegrityWhen NonStop returned to HP by way of Compaq,it brought with it truly unique capabilities.NonStop runs many of the world’s bankingsystems; HP estimates that it powers 75 percent ofthe 100 largest electronic fund transfer networks. Italso handles the majority of all ATM and creditcard transactions at organizations like Bank ofAmerica and Barclay’s Bank. What is perhaps evenmore impressive, about 95 percent of the world’ssecurities transactions take place on NonStop gear,at over 100 stock exchanges including the NewYork Stock Exchange, Chicago MercantileExchange, and Hong Kong Stock Exchange.NonStop also underpins important parts of theworld’s network and telecommunicationsinfrastructure (for example, at AT&T, BritishTelecom, and NTT). Among the wins it contributedat Compaq was the 100 million decision by Sabreto kick out well-entrenched mainframes; Sabre’salso a poster child for a newer style of “hybrid”NonStop deployment in which distributed Linuxfront-ends feed NonStop transactional back-ends.NonStops are also found in healthcare,manufacturing, retail, and government—and theyhandle about half the US 911 emergency calls.NonStop plays these important roles because theapplications that run on it share two keycharacteristics. The first is that they demandabsolute reliability and absolute transactionalintegrity. NonStop provides these guaranteedattributes through a combination of hardware andadvanced software such as its NonStop operatingsystem and NonStop SQL/MX DBMS. It is, in aphrase, fault tolerant (FT). They also requiremassive scale—and not just the sort of scale thathooking together a lot of x86 servers with GigabitEthernet can deliver. They need systems that canhandle update-intensive transactional workloadswhere there’s a lot of writing (requiringcoordination across the system) as well as reading.The net is that there are few, if any, other systemson the planet that can do what NonStop can do:handle multi-terabyte data volumes, real-time datafeeds, and on-the-fly transformations whilesimultaneously guaranteeing that the services willalways be up and available.But engineering the unique capabilities that madeNonStop so attractive to all those organizationslong meant highly bespoke design that littleleveraged the work that went into moremainstream systems. And, as important, didn’tbenefit from the volume economics that came frommainstream ubiquity. And that cast a certain pallover NonStop’s future, given that it filled animportant niche, but a niche nonetheless.By melding NonStop into its Integrity product set,HP sharply reduced the amount of uniquehardware that it needed to develop in order tomaintain the NonStop line. While many of theboard- and module-level components were stillunique to NonStop, they extensively leveraged thesame processors, chipsets, and other parts of thehardware design that went into HP’s mainstreamIntegrity Unix platform. This representedenormous savings—and allowed a faster pace ofinnovation—relative to having to largely buildeverything from scratch specifically for NonStop.The integration with c-Class blades puts NonStopanother step closer to the main current ofcomputing—probably as close as possible for asystem architecture with such uniquecharacteristics.c-Class BladesWhen HP decided to shift its p-Class blades to cClass in 2006, it took a calculated gamble.2 Unlikethe approach that IBM took with BladeCenter H,3the new c7000 chassis was not backwardlycompatible with the components that went into theprior p-Class product. This meant that HP had todesign and manufacture a whole new blade lineupfor the new chassis—while also supporting existingp-Class customers for a reasonable interval.Furthermore, because a blade chassis in the Dell,HP, and IBM vein functions as a sort of integration2HP’s c-Class blades are discussed in our HP Blades Gofrom p to c, as is the evolution of blade servers fromdisaggregated computing to integration point.3See our A Resharpened BladeCenter.*ECF9DA05ADEB7A6
3point for third-party network switches and the like,HP wasn’t even the only company whose productdesigns were affected. Partners like Brocade, Cisco,and Nortel likewise had to come out with newproducts for the new chassis form factor.However, changing form factors also gave HP theopportunity to start with “a clean sheet of paper” ata time when blade adoption was still fairly limitedand the installed base correspondingly small. Thisallowed HP, for example, to revamp the power andcooling design of the chassis—a set of technologiesand capabilities that it collectively refers to asThermal Logic. One such component—which HPunderstandably loves to demo at every opportunity—is its Active Cool fans. Inspired by the highvelocity propellers used in model airplanes, theseHP-designed fans—on which it holds, or hasapplied for, over 20 patents—spin at fantasticRPMs. Equally important is their logic; they use acontrol algorithm to optimize operatingparameters. They speed up and slow down inresponse to changes in load; there’s also thepotential for using extra fans in someconfigurations so that they run slower and, hence,quieter. HP estimates that these fans draw 66percent less power than traditional ones.In any event, c-Class blades—propelled by boththeir own capabilities and a general renaissance ofoverall business under Mark Hurd—have paid offhandsomely for HP. Market share data from all theusual sources shows HP’s having taken the bladelead away from IBM. This has mostly been an x86story and doubtless reflects, in no small part, theoverall strength of the ProLiant server productlineup, as backed by solid sales, marketing, andsupply chain management.However, Itanium is part of the c-Class story, too.Introduced in 2005, the original BL60p blade wasbest thought of as a way of bringing HP-UX toHP’s blade integration point—that is, a way to runHP-UX applications alongside those running onx86 blades in the same chassis.4 The BL860c is the4In other words, the BL60p wasn’t really intended as ageneric heavyweight processing blade for Linux highperformance computing workloads and the like. Seeour HP-UX Gets Bladed.current dual-core “Montvale”-based flavor ofItanium blade that replaced the original BL60p.Building Up a NonStopBefore delving into how NonStop works as part ofa blades architecture, it’s worth reviewing somebasic NonStop concepts and how they’reimplemented in the current generation of IntegrityNonStop systems, the H-Series.The basic computing unit in a NonStop system is a“logical processor.” The hardware components thatmake up a logical processor have varied amongdifferent generations of NonStop, but the commonthread is that it’s the smallest standalone unit ofprocessing that’s doing “real work.”A NonStop server is an aggregation of logicalprocessors connected by ServerNet into a looselycoupled, i.e. distributed memory or shared-nothing,computer system. Tandem introduced ServerNettogether with its NonStop Himalaya S-Series in1997. For a time, Tandem promoted ServerNet as astandard “system area network” in a vein similar toDolphin’s Scalable Interface (SCI) and InfiniBand.But it never saw wide use outside of Tandem (and,less so, Compaq after it acquired Tandem).One also sees the term “node” used within thecontext of NonStop systems. The term isn’t alwaysused consistently, including within HP literature,but the official current meaning is to refer to agrouping of up to 16 logical processors that share asingle system name. It’s primarily a managementconcept that isn’t normally visible to applications.For example, database schemas can span nodes outto the maximum system size—4,080 logicalprocessors. This differs from the more conventionaluse of the “node” term to refer to the SMPbuilding blocks within a cluster—that is, somethingakin to a logical processor in the NonStop case.None of these basic concepts change in movingfrom the first generation of Integrity NonStopservers to the current one. However, at a detailedlevel, the system now implements fault tolerance ina conceptually quite different way.*ECF9DA05ADEB7A6
4Fault ToleranceBefore this move to blades, Integrity NonStopservers were based on what HP calls the NonStopAdvanced Architecture (NSAA). In a DMR (dualmodular redundant) NSAA configuration, twoprocessors on two different physical modules carryout each computation.5 The results are thencompared; if the results aren’t identical, acomparison error between the two boards is noted,and the system will try to determine the source ofthe error. If it can do so, it will disable the defectivehardware and continue processing. Otherwise, itwill disable both halves of the logical processor andtransfer the job transparently to another logicalprocessor using a NonStop software FT techniquecalled “process-pair takeover.” LogicalSynchronization Units (LSU) handle thecomparisons by looking at all the I/O traffic acrossthe ServerNet interconnect. The LSUs compareoutput packets to ensure that an error or corruptionhasn’t occurred, and they convert I/O packets toand from ServerNet’s format.6NonStop has used conceptually similar techniquessince abandoning its own processors in favor ofthose from MIPS in 1991. As you’ll note from theprevious paragraph, the NonStop software doesn’tactually depend on the hardware to recover from afailure. If the hardware can transparently recover,all the better, but it isn’t really necessary in mostcases. Contrast this with the pure hardware FTapproach used by NEC and Stratus. In this case, thewhole idea is to mask any hardware failures fromWindows or Linux (which, under normalcircumstances, could no more recover from a CPUfailure than automagically compose a sonnet).7Thus, while NonStop’s approach is often lumped inwith FT that depends on hardware-basedlockstepping, it’s actually quite different.5A module is a hardware building block containingprocessors, I/O Bridge, zx1 memory and buscontroller, and the memory complex. HP used theterm “NonStop Blade Element” but I’m avoiding thatterminology in the text to clearly differentiate fromthe standard blade chassis in this latest iteration.6See our Itanium Goes NonStop at HP for more on NSAA.7See our Stratus Cuts the Cost of Fault Tolerance (Again), Linux FT, àla NEC, and Stratus ftServers: Windows Fault Tolerance for Verticals.So, why bother with all the comparisons, then? In aword, detection.In the NonStop architecture, the issue isn’t somuch recovering from a known error—theNonStop software knows how to do that—butdetecting that such an error has occurred in thefirst place. If an undetected error does occur, theresult can be worse than an outright failure becauseit can result in silent data corruption and otherbehaviors that just aren’t acceptable for the sort ofenvironments where NonStop plays. We’re nottalking high frequency events of course, but whenyou’re aiming for “runs for years withoutinterruption” even low probability glitches have tobe eliminated.In fact, explicit synchronization between processorswas an approach that Tandem only introducedwhen it moved away from its own processors thathad far more internal data checking mechanismsthan their MIPS successor did. In other words,when the processor and associated hardware couldno longer be depended upon to police its ownactions, Tandem introduced a sort of “buddysystem” to mitigate any deficiencies.Detecting Errors in HardwareThis latest NonStop generation, in a sense, comesfull circle. It eliminates processor synchronizationhardware in favor of a purely software-basedapproach. Programs running on each logicalprocessor are “backed up” on another logicalprocessor in the system. In this context, “backedup” means there’s a copy of the running programand that context changes and other changes in theprogram status get copied to the backup overServerNet. There’s obviously some overhead inthis, but the backup program isn’t actually runningand consuming CPU resources; it’s just sitting inmemory. HP estimates there’s about five percentoverhead associated with maintaining this copy.What’s made this possible is the substantialincrease in error detecting and correcting featuresbuilt into even relatively mainstream, if not exactlymass market, Unix servers. Thus, the zx2 chipset,developed by HP for its Itanium blade and other*ECF9DA05ADEB7A6
5entry-level Itanium-based servers, includes anynumber of reliability and availability features.These include memory-related features such aspage de-allocation, chip sparing, and DIMM addressparity protection.Intel has likewise been augmenting the Itaniumprocessor itself with features such as Cache Safe(detects and shuts off malfunctioning areas ofcache),8 Enhanced Machine Check Architecture(enables multi-level error handling acrosshardware, firmware, and operating system),memory scrubbing, and ECC on the system bus.More features of this type are planned for“Tukwila,” the next generation of Itaniumprocessor planned for late-2008 or 2009. Reliability,Availability, and Serviceability (RAS) features thatgo above and beyond the norm are increasingly abig part of Intel’s pitch for Itanium as a point ofdifferentiation from 64-bit x86 processors fromitself and AMD. For many classes of servers, onecan reasonably ask how important theseincremental functions are. NonStop, however, hasan application environment and user base thatgives even small reliability gains outsizedsignificance. When “it must not go down!” is ineffect, “detect and correct everything you possiblycan!” becomes much more the rule of the road.Bringing It All TogetherThe heart of the new NB50000c is a c7000BladeSystem, a 10U-high chassis that houses fromtwo to eight BL860c Integrity blades. Each bladeincorporates one Itanium 9100 series dual-coreprocessor (“Montvale”) running at 1.66 GHz.9Blades can be configured with between 8 GB and 48GB of DDR2 memory. Each blade functions as alogical processor within a NonStop system; itconnects the other logical processors in the systemusing a ServerNet mezzanine card on each blade, inconcert with redundant ServerNet switch modulesthat plug into the back of each blade chassis.108Originally codenamed “Pellston.”9The BL860c has two processor sockets and cannormally be configured with either one or twoItaniums. In the NonStop application, only a singledual-core processor per blade is used.10Although the ServerNet mezzanine card andOne significant change that’s enabled by the new JSeries NonStop Kernel is that the logical processorscan make use of both physical cores of the Itaniumprocessor. The NonStop system as a whole is stillshared nothing/distributed memory/MPP—chooseyour term—but the individual logical processor cannow be an SMP in which the two cores share theblade’s memory. While this may seemstraightforward, it’s actually a significant changefrom historical NonStop logical processors in whichonly a single CPU actively executed code in eachlogical processor.11This change is important because individualprocessor cores aren’t getting much faster. Chiphorsepower increasingly comes from having moreengines, not more powerful individual ones.Therefore, benefitting from the increasingtransistor densities predicted by Moore’s Lawpretty much requires making effective use of coresin the aggregate. To do otherwise is a recipe forstagnant performance.The one downside of this approach is that, at leastcurrently, it’s incompatible with implementing anerror-checking regimen that involves three copiesof the program rather than two—a true belt-andsuspenders level of paranoia that goes by theacronym TMR (triple modular redundant). Fornow, this remains a feature of H-Series IntegrityNonStops but not the new bladed flavor. HPestimates that TMR is of interest to about 10 to 15percent of its base; it does expect to have a followon implementation of TMR that leverages bladesby trading off some performance for theincremental availability.12Storage and local area networks connect to theserver through dedicated interfaces. In theNB50000c, the interface is provided by aswitches are specific to NonStop, they plug in thesame way that Ethernet or Fibre Channel devices do.11That the two CPUs in this case reside on the sameprocessor die doesn’t change the fact that they’rearchitecturally independent CPUs as far as thesoftware is concerned.12The difference here is somewhere in the five-ninesversus seven-nines range; the difference between99.999 percent uptime and 99.99999 percent uptimetranslates to about 4 minutes a year.*ECF9DA05ADEB7A6
6ServerNet-connected Cluster I/O Module (CLIM)that is just software running atop a standardProLiant rackmount server (a DL385 G5); the priorgeneration used a purpose-built device called an I/O Adapter Module Enclosure (IOAME) togetherwith other specialized hardware.13As with other NonStop hardware, CLIMs aretypically configured redundantly. Thus, if an entireIP CLIM (i.e., the CLIM used for networking) fails,all the Ethernet interfaces on that CLIM are failedover to its preconfigured failover destination on adifferent CLIM. When individual links to anexternal network fail, interface resources can beswitched within a CLIM using a feature calledbonding interfaces.The Storage CLIM connects to SAS disks within anMSA70 enclosure. This, too, is standard HPhardware, except that it’s configured with a secondI/O Module for increased redundancy. TheNB50000 can also connect to various FibreChannelbased HP StorageWorks XP arrays.The rack is rounded out with an in-cabinet UPSand a slide-out management console. The newsystem makes use of standard HP managementtools. HP NonStop Cluster Essentials integratewith HP Systems Insight Manager (SIM), whichcan also monitor and manage an entire bladedinfrastructure: the HP SIM Blade plug-in. TheOnboard Administrator provides managementspecifically for the blade infrastructure.14Alternatively, NonStop management tools andapplications remain available for deep-dive tasksthat are specific to the NonStop architecture oroperating environment.ConclusionEbullient with the success of its c-Class blades, HPhas, of late, been loudly touting a strategy that goesby “Blade Everything” or “BladesUnbound” (depending upon who is doing thetalking). In practice, the strategy is a more practical13IOAMEs are also supported for compatibility.14The HP SIM Integrated Blades plug-in manages thechassis and server blades through the OnboardAdministrator.“Blade Everything It Makes Sense To.” HP willhappily continue to ship rackmount, and eventower, servers to those who prefer them—whetherin SMB, emerging markets, or even enterprises thathaven’t quite seen the light yet. Furthermore, whilesome degree of modular design is doubtlessappropriate for even large SMP servers, that’s notthe same as saying that everything must beshoehorned into a single standardized blade chassis.That said, the NB50000c provides a data point thatdemonstrates that HP is serious about goingfurther in this direction, more quickly than anyother major vendor. Perhaps SGI comes closest, butwith a far more specialized product lineup.Certainly, none of the other companies typicallyconsidered Tier 1 OEMs are currently pushingblades as broadly as HP despite past leanings in thisdirection—such as IBM’s (largely x86-centric)modular computing push of a few years back.None of this should be taken as a suggestion thatNonStop is Just Another Blade. NotwithstandingServerNet I/O cards and switches, the NB50000c isbest thought of as largely co-opting hardwaredeveloped for other purposes, rather thangracefully co-existing on the same set of hardware.Thus, although future HP plans call for “AdaptiveInfrastructure in a Box” blade solutions in whichNonStop works alongside Linux, Windows, or HPUX blades, for now we’re talking about dedicatedracks of NonStop gear that just—by no meansincidentally—take standard gear, add ServerNet,and sprinkle a liberal dose of NonStop softwarepixie dust. Think of these as NonStop systems thathappen to leverage the BladeSystem design, ratherthan standalone blade servers.But that’s OK. HP has (if belatedly) recognized theincredible intellectual property it has in NonStop,and started to apply it to problems and customersets that fall outside its traditional market.NonStop will never be a truly mainstream productif, by mainstream, you mean something that’sapplied to pedestrian problems. Rather, theopportunity here is when “good enough” isn’tenough even when it doesn’t truly reach the mostinsane levels of the most ultimate scale andultimate reliability.*ECF9DA05ADEB7A6
Compaq bought, but never really absorb ed, the company. When HP subsequently purchased Compaq, it acquired a lineup that was an integral component of many of the world s most critical computing infrastructures which is to say infrastructures that move a lot of money around. Like Compaq, HP
HP OpenView NonStop support in most BTO Centers Many customers using BTO products in NonStop environment Looking for customer input Supported today for NonStop Consolidated infrastructure monitoring tools allow you to spend more t
IBM MQ for HPE NonStop V8 is similar to the pr oduct on Linux and UNIX platforms. Most of the tools and options available on Linux and UNIX platforms work in exactly the same way on HPE NonStop. This documentation covers only the dif fer ences between IBM MQ for HPE NonStop V8 compar ed to Linux and UNIX platforms.
HP NonStop Enscribe to MS SQL Server Page 1 HP NonStop Enscribe to MS SQL Server Objective Upon completion of this lesson, you will be able to keep two heterogeneous databases synchronized, in this case NonStop audited Enscribe to Microsoft SQL Server. During this lesson, you will learn how to:
9 13 January 2010 Operations Agent for NonStop - OVNM Provides object monitoring and event processing Automatic discovery of NonStop objects Processes subsystem and application events Provides customized threshold setting Sends messages upon violations Supports ACI Base24, Telco (HLR, INS), MQ Series Easy i
Sets/gets the join delay on Rx window 2. AT RX1DL. Sets/gets the delay of the Rx window 1. AT RX2DL. Sets/gets the delay of the Rx window 2. AT RX2DR [ datarate] where X [0:7] Sets/gets data rate of the Rx window 2. AT RX2FQ [ freq] where freq in Hz Sets/gets the frequency of the Rx window 2. AT TXP [ txpow] where txpow [0:7] Sets/gets the .
boat wind turbines and make them facing the wind [3]. The number of blades of boat wind turbines is often 3. Three-bladed boat wind turbines can produce power at low wind speed and can be self-started by the wind. This paper is focused on three-bladed boat wind turbines with passive yaw motion.
Plán plavby Nonstop osamělá plavba kolem světa – projekt Jarmily a Petra Ondráčkových Pokus o rekord v LOA 28 – 32 ft - zatím Robin Knox- Johnston – 313 dní (31 ft, prům. rychl. 3,69 kt z roku 1968 Trasa – E směrem – Falmouth (Lizzard Point), mys Dobré naděje, Austrálie, Antipode, mys
geomagnetic field Magnetic “Operative” physical property Method Measured parameter. Further reading Keary, P. & Brooks, M. (1991) An Introduction to Geophysical Exploration. Blackwell Scientific Publications. Mussett, A.E. & Khan, M. (2000) Looking into the Earth – An Introduction to Geological Geophysics. Cambridge University Press. McQuillin, R., Bacon, M. & Barclay, W .
