NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/sys/hp/15206

Text File | 1993-01-22 | 10.8 KB | 266 lines

Newsgroups: comp.sys.hp
Path: sparky!uunet!usc!sdd.hp.com!apollo.hp.com!netnews
From: [e-mail] (#David McSavaney)
Subject: Re: SW for snake farms
Sender: [e-mail] (Usenet News)
Message-ID: <[e-mail]>
Date: Fri, 22 Jan 1993 16:45:37 GMT
References: <[e-mail]>
Nntp-Posting-Host: macsav.ch.apollo.hp.com
Organization: Hewlett-Packard Corporation, Chelmsford, MA
Keywords: cluster of snakes, load-balancing
Lines: 252

In article <[e-mail]> [e-mail] writes:
>Snake-farm: a more or less loosely coupled bunge of snakes where one
>            can share resources (e.g. balance the load over more
>            machines,...)
>          [much stuff deleted]
>--
>Christoph F. Strnadl              | The term 'politically correct', co-
>Institute for Theoretical Physics | opted by the white power elite as a
>TU Vienna, Austria                | tool for attacking multiculturalism,
>email: [e-mail] | is no longer politically correct.
>

I've included a long piece of background information from July 92 about the
Snakes Farm installed at CERN, though I am not sure if the contacts are
still valid. You be pleased to know that further development
of the Snakes Farm concept at CERN with Convex resulted in the
announcement of the Cluster Computing program in October 92. Your local
HP or Convex office will have more details.

Software available for Cluster Computing includes:
Linda from SCA
NQS from Sterling
Express from Parasoft
ISIS from ISIS
Load Balancer from Freedman Sharp
PVM/HeNCE from Oak Ridge NL (Public Domain)
Taskboker from HP
ConvexNQS, ConvexPVM and ConvexMLIB
Many networked sys admin tools from HP

Best wishes
Dave McSavaney, Product Manager,
MP Workstations, Advanced Systems Division,
Hewlett-Packard Co.

Still not an official statement of Hewlett-Packard Company
-----------------------------------------------------------

>                   HP 'SNAKES FARM' AT CERN
>
>
>
> About CERN and the HP 'Snakes Farm'
> -----------------------------------
> Located near Geneva, Switzerland, CERN (European Laboratory
> for Particle Physics) is one of the world's largest
> scientific laboratories. Funded mainly by 17 Member States,
> it's annual budget for 1992 was 945.47 million Swiss francs.
> CERN has over 3000 employees and provides support for over
> 6,000 visiting scientists per year.
>
> HP's 'Snakes Farm' consists of up to 50 HP 9000/700 servers
> linked together and accessed via an HP 9000/720 workstation
> centralizing job requests. The 'Snakes Farm' can be
> configured to provide from 57 to 2,850 MIPs using 9000/720's.
>
> In December 1991, CERN installed 16 HP 9000/720 systems. The
> system cluster is linked together and accessed via an HP
> 9000/720 workstation that centralizes job requests.
>
>
> What CERN Does
> --------------
> CERN's business is particle physics; studying the
> innermost constituents of matter to find out how the Universe
> was created and how it works.
>
> It is thought that the Universe was created when a huge
> quantity of energy was suddenly transformed into billions of
> particles which, after many changes and transformations,
> finally stabilized and created the Universe as we know it.
>
> At CERN, this process can be artificially induced on a very
> minute scale so that the process can be studied. To do this,
> physicists inject particles with a large amount of energy by
> accelerating them to almost the speed of light. When two
> particles collide, energy is freed and transforms into
> matter. Using detectors and computers, physicists observe
> what these new particles do.
>
>
> Scientific Equipment at CERN
> ----------------------------
> CERN has several types of particle accelerators and
> colliders. Much of the research today is performed on the
> Large Electron-Proton collider, or LEP. LEP is built in a 27-
> Kilometer underground ring and is currently the world's
> largest scientific machine. There are four large experiments
> being conducted on LEP, which provide research material for
> over 2,000 physicists worldwide.
>
> LEP works as follows: Electrons and positrons (the antimatter-
> counterparts of electrons) are whirled around in the ring and
> smashed together. Each collision produces a spray of up to
> several hundred fragments (secondary particles). Four
> detectors spaced round the ring intercept the emerging
> fragments of matter. Sophisticated electronics equipment
> within the detectors record and monitor the mass of incoming
> data. Each detector has its own data acquisition system,
> processing raw information before passing it on to the
> experiment's main computing system for further processing and
> recording onto cassettes.
>
> A new machine, the Large Hadron Collider (LHC) is planned for
> 1998. LHC will bring protons into head-on collisions at
> higher energies than ever achieved before. It will reveal
> behavior of the fundamental particles of matter that has
> never been studied before.
>
>
> CERN's Computing Requirements
> -----------------------------
> CERN's data storage requirements are immense. The information
> from a single collision could fill a telephone directory. All
> of CERN's experimental data is stored on 60,000 cassettes,
> representing over 50 Terabytes (1 Terabyte = 1,000
> Gigabytes). The large size of the data sets presents
> challenges in networking, data handling and data storage.
>
> As for processing power, a typical job is so large that it
> runs for one CPU day. Each of the four current experiments on
> LEP requires about 1000 MIPs for offline data processing and
> event simulation. (Offline processing is performed some hours
> or days after an experiment has been completed, typically
> using data that has been recorded on cassettes.)
>
> The CERN computing center also has a fast-growing requirement
> for CPU time for event simulation for LHC and LEP
> experiments. For example, a large simulation effort is
> underway by accelerator physicists to design the magnets and
> beam transport specifications of the LHC.
>
> CERN needs experience with very large computing facilities in
> preparation for LHC. LHC experiments will require between 100
> and 1000 times the computer resources of the current
> generation of LEP experiments. An investigation is being
> undertaken to determine how to build and manage a system that
> can meet future requirements.
>
>
> Starting with the 'Snakes Farm' at CERN
> ---------------------------------------
> In 1991, computer specialists at the CERN decided to install
> a 'Snakes Farm.' This decision was based on mainly on two
> factors: price/performance and scalability.
>
> Computing services based on the 'Snakes Farm' were introduced
> in December 1991. The service is called the Central
> Simulation Facility (CSF) and is based on a cluster of 16 HP
> 9000/720 workstations and servers connected to the CERN disk
> and tape service developed for a centrally operated RISC*
> environment (CORE). CSF is a joint CERN/HP project.
>
> The goal of CSF is to provide physicists an integrated system
> capable of handling the full range of batch work typical of a
> large High Energy Physics lab. For 1993, CSF plans to provide
> an aggregate system throughput of 500 CERN Units*
> (approximately 1 GFlop).
>
>
> CSF Hardware Installation
> -------------------------
> At CSF, the 16 HP 9000/720 systems are rack mounted and
> linked together. One of these 16 centralizes job requests and
> manages the job queue.
>
> The 'Farm' is connected to an Ethernet segment. In this same
> segment there are three Apollo DN 10000 workstations. The HP
> 9000/720 servers work together with and augment the
> capability of these workstations.
>
> Also on the same Ethernet segment are two servers that act as
> tape-staging devices for jobs running anywhere in CERN's
> centrally operated RISC environment. Over 50 tape drives are
> operated round-the-clock by a dedicated tape robots and a
> tape-handling team. CERN uses an IBM 3090/600J mainframe
> computer to manage the data on these cassettes and the tape
> robots. (The IBM mainframe was originally purchased as a
> batch machine, but today it is used primarily for tape
> handling and interactive online tasks such as electronic mail
> and file editing.)
>
> The Ethernet segment accesses the sitewide FDDI backbone.
> From the FDDI backbone, the cluster accesses the rest of the
> site's Ethernet LAN through an FDDI-to-Ethernet DEC 600
> bridge.
>
> Each HP 9000/720 system is configured with a 400 Mbyte local disk,
> 16 MByte RAM, and HP-UX.
> Most of the HP servers in the 'Farm' run 'diskless,' that is,
> they are booted over the LAN from an HP-UX cluster server.
> This frees up over 80 Mbytes per disk while reducing the
> administrative load of maintaining separate file systems.
>
> CSF Batch Environment, File System and Usage
> Users submit jobs to the HP 9000/720 workstation via NQS
> (Network Queueing System). NQS manages the job queue and
> distributes jobs evenly to the destination servers.
>
> At the start of a job, the executable program and any data
> card files the job might need are copied to a temporary
> workspace on the local disk. Job output is also written to
> the local disk.
>
> When the job finishes, the results are recorded onto a
> cartridge or sent over NFS (Network File System) to the user.
> NQS then cleans up the temporary work area, and the system is
> ready to accept the next task.
>
> People are currently in the various stages of porting their
> large programs to the HP-UX Fortran compiler. Existing code
> running on the HP-UX platform does not need to be recompiled
> for the 'Farm'.
>
> Conclusions and Benchmark Results
> ---------------------------------
> Today, the 'Farm' handles a significant amount of the scalar
> computation for simulation, analysis, data processing, and
> visualization at CERN. The 'Farm' has doubled the CPU
> capacity of CERN's central mainframes.
>
> CERN benchmark testing on the HP 9000/720 resulted in the
> following:
>
> - One HP 9000/720: 10.5 CERN Units (together the 16 9000/720s
>   provide 168 CERN Units)
>
>
> - IBM 3090/600J: 9.5 CERN Units per CPU (for a total of almost
>   60 CERN units)
>
> - CRAY Y-MP (6NS): 11 CERN Units
>
> In addition, CERN's benchmark tests showed that the
> processing power of the 'Farm' was comparable to that of an
> IBM 9000/900.
>
> For more information
> --------------------
> Anyone interested in visiting the 'Snakes Farm' at CERN
> should contact Les Robertson in Geneva at (41) 22-767-4916. More
> detailed information about the HP 'Snakes Farm' can be
> obtained from Michel Benard (HP) at (41) 22-780-8165 (Switzerland).

 ****Not sure if these are still valid*****
>
> A CERN Unit is a combination of floating-point, integer, and
> double-precision tests
>
> .......................................................................