[Closed] Soooo.... High Performance Computing

high performance computing

Is that when you type fast? Or when you make Voderman look like a brick? Or when you have an awesome gaming laptop?

Well as far as I know it could be playing Doom whilst swinging on a trapeze. I don't think it is though

You seriously don't even know what it is?

Supercomputers.

http://www.hpcwales.co.uk/what-is-hpc

Though a lot of the time that means (or is closely related to) clusters.

Ignore that guy, do the trapeze thing.

You going to be writing software - if so which language?

What does the company do ?

Hardware or software type high-performance ?

concurrency/multi-threading/interrupt-driven ?

that HPC stuff is rubbish, high-performance computing requires more than one computer ???

It depends on what you are trying to do versus the resources needed to do it, and the cost you are prepared to pay. You might want a single computer, like some huge Cray or similar.

Hard to know without any real context.

Try reading up on map-reduce, concurrency and parallelisation, possibly stats and linear algebra?

All the kool kidz are doing it on GPUs these days.

Renderscript on your phone, and OpenCL everywhere else. Go and write some code to run on the vector processing core of your Nvidia graphics card.

I even read somewhere about virtualized GPUs for cloud GPU compute. Quite how that can possibly work I really don't know.

Or they might want to talk about interconnect - I used to know some people who worked on an ethernet controller designed for HPC. Very interesting problems to solve. They spent a lot of time worrying about latency.

You really need to narrow it down though, it won't look do great if you go to the interview without a good idea of what you are letting yourself on for, and what you might be able to bluff versus what you should 'fess up to being a bit weak at, which is often a better play.

I even read somewhere about virtualized GPUs for cloud GPU compute. Quite how that can possibly work I really don't know.

It has some powerful buzzwords, which it's often all you need...

and what you might be able to bluff versus what you should 'fess up to being a bit weak at, which is often a better play.

+1

My understanding from herself, who's company is in the process of buying a new Cray, they're all node based nowadays, as you get scalability, maintainability and reasonable cost efficiencies from using modular components rather than building a big computermabob with its own building and white coated technical staff.

As others have said, it kinda depends on what is meant by "High Performance Computing" - some folks reckon it's GPUs, others reckon room-filling supers. In truth it's a bit of both and lots of stuff inbetween: specialist software, networks, storage, power, cooling, visualization, etc.

What I'd suggest is a bit of a look into the kind of machines are around these days and what they're built from (standard Xeon cores, Xeon Phi, GPGPU, ARM...) and what makes them special. Then have a think about how you'd program them and the kind of issues that goes a long with that. If you've got a decent scientific background you'll get the idea.

I've worked in HPC for 15 years and have recruited folks - we generally looked for an appreciation of the subject - people with HPC skills are few-and-far between and there's demand for them - so good news for you! But it's dead easy to spot a blagger, so don't even try! 😀

How's your Fortran? 😉

You mean like this aP 🙂 Bunch of Raspberry Pis and some lego.

[img] [/img]

Inmos [url= http://en.wikipedia.org/wiki/Transputer ]Transputer[/url]. remember them?

Need more info. It could be any of the above, could be low latency enterprise systems, or it could just be the employer putting fancy sounding buzzwords on their job request hoping to attract good candidates, when they really mean they want someone who can write excel macros.

What job was it?

OP at home with different login
Employer is univeristy. other skills include Unix (solaris) - check, wintel and sun server hardware - check, lan/wan networking - super check with cisco ios as a bonus, programming languages mentioned - sort of check as in familiar with oop theory and code reusability, ms-sql - enterprise class check. Job is management, head of IT, so it's more likley that a working knowledge of high performance and mass storage devices is what's needed rather than being able to do day to day management.

EDIT - ok with admitting to being weak in an area, not so with knowing nothing

kcal
Inmos Transputer. remember them?

Pretty much still going as these little beasties:

[url= http://www.xmos.com/ ]XMOS microcontrollers[/url]

Sounds like you might have to do some kind of administration on some specific type of supercomputer system or some such. Maybe try to out what they have with some google detective work?

Wow, let me check I've got this right. You have an interview for head of IT at a university somewhere, you know bugger all about HPC and don't even know what the university has access to? Give us the Uni name and I'll google it for you.

Don't mean to sound harsh but I do hope they find someone vaguely competent to do the job. 8)

Did that, but want to keep it generic - ahh yes, high performace computing, well I'm familiar with the concept but not with your particualr implementation

All the kool kidz are doing it on GPUs these days.

No they are using FPGAs to optimise sections of code.

To the OP:

If you are administrating a HPC look into queing systems like

http://www.clusterresources.com/torquedocs21/commands/qsub.shtml

It is in many ways a co operative scheduler. Users write a scripts that requires a number of cores for a preset time and the que manger optimises these requests. This will give you something to talk about.

Also read about PBS

http://rcc.its.psu.edu/user_guides/system_utilities/pbs/

That Pi rack is cool

ms-sql - enterprise class

and don't even know what the university has access to? Give us the Uni name and I'll google it for you.

Don't mean to sound harsh but I do hope they find someone vaguely competent to do the job.

which university is it ?

Inmos Transputer. remember them?

Now there's a blast from the past. Leant Occam as part of my degree 20+ years ago.....

Do you know what software they are running on the HPC

Is it Lustre, Hadoop or something they have developed in house?

I work for a hardware vendor so I understand the nuts and bolts of how the clusters are put together but I'm guessing in your role the actual hardware is less important than an understanding of what work the cluster is actually carrying out.

Inmos Transputer. remember them?

mate got a job with Inmos after Uni.

They sent him to the states several times.

Eventually he decided to go and work for Micron, with a golden handshake forthcoming.

Eventually he was in charge of 5 fab units, and on his way to early retirement in the house he was building in the mountains in Idaho as a multi-millionaire.

OP at home with different login

Breaking forum rules and lying in interviews!? Won't someone please think of the children!

I am guessing that the [b]ChubbyBlokeInLycra[/b] logon is the original logon, and [b]BigButSlimmerBloke[/b] logon is the newer logon, but he is keeping the [b]ChubbyBlokeInLycra[/b] logon in case he reverts...

Sounds like science-based HPC then using clusters, unless it's head of IT for the Computer Science Department. If you are not coming from the Uni / research sector, look at HeCTOR and places like the Daresbury HPC for info on the national-level systems available to researchers. To me, from a science background, HPC means a bunch of processors linked togther, with a software front-end that queues and prioritises jobs. It may or may not include parellel programming.

If the Faculty have their own cluster, I would expect the manager's job to include thinking about running costs as well as maintenance (I spent 3 years at a research centre that spent a significant chunk of its electricity costs on air conditioning for the server room).

(I spent 3 years at a research centre that spent a significant chunk of its electricity costs on air conditioning for the server room).

That does not surprise me. When I worked in research one of the cluster rooms aircon went down and it killed that section of hte cluster off. Took about a week to fix I seem to recall.

When I worked in research one of the cluster rooms aircon went down and it killed that section of hte cluster off.

Google beowulf cluster, and sharpen your Linux-foo

MS have an HPC edition of Windows, but I've yet to see anyone using it.

I am guessing that the ChubbyBlokeInLycra logon is the original logon, and BigButSlimmerBloke logon is the newer logon, but he is keeping the ChubbyBlokeInLycra logon in case he reverts...

Anyway, I like the look of this
[img] [/img]
[url= http://game-insider.com/2010/12/23/1760-ps3s-used-for-pentago-supercomputer/ ]High performance linky[/url]

Tsk, no air-con redundancy...bloody amateurs.

How long before people are putting data centers on the dark side moon, taking advantage of the ice at the poles and in the deeper craters there?

Latency wouldn't be great.