headers
Hearns, John | 4 Oct 2010 17:04
Favicon

DIY CAVE - Liquid Galaxy

Yet again another Register article...

http://code.google.com/p/liquid-galaxy/

Might be interesting, as this is a cluster of computers used to make an
immersive visualisation setup.
The 'secret sauce' is a feature in Google Earth which makes it really
easy to slave displays together:

If send == true, sets the IP where the datagrams are sent
; Can be a broadcast address
ViewSync/hostname = SLAVE_IP_GOES_HERE
ViewSync/port = 21567

John Hearns
McLaren Racing

The contents of this email are confidential and for the exclusive use of the intended recipient.  If you
receive this email in error you should not copy it, retransmit it, use it or disclose its contents but
should return it to the sender immediately and delete your copy.
Permalink | Reply | 1 comment |
headers
Gavin W. Burris | 4 Oct 2010 18:40
Favicon

Re: DIY CAVE - Liquid Galaxy

Neat!  I have built tiled display walls in the past.  You can also do a
similar thing for any video that VLC will play.  Or you can use the DMX
project to link all Xorg displays into one giant interactive desktop.
http://viz.aset.psu.edu/ga5in/DisplayWall.html

On 10/04/2010 11:04 AM, Hearns, John wrote:
> Yet again another Register article...
> 
> http://code.google.com/p/liquid-galaxy/
> 
> 
> Might be interesting, as this is a cluster of computers used to make an
> immersive visualisation setup.
> The 'secret sauce' is a feature in Google Earth which makes it really
> easy to slave displays together:
> 
> 
> If send == true, sets the IP where the datagrams are sent
> ; Can be a broadcast address
> ViewSync/hostname = SLAVE_IP_GOES_HERE
> ViewSync/port = 21567
> 
> 
> 
> John Hearns
> McLaren Racing
> 
> The contents of this email are confidential and for the exclusive use of the intended recipient.  If you
receive this email in error you should not copy it, retransmit it, use it or disclose its contents but
should return it to the sender immediately and delete your copy.
(Continue reading)

Permalink | Reply | 0 comments |
headers
Michael Di Domenico | 4 Oct 2010 22:12
Picon

Re: Begginers question # 1

To answer your question directly, the answer is, no the performance is
not the same.  BUT, you've asked a very workload dependent question,
but have not told us anything about what you're trying to do, so
deciding which is the right choice is pretty hard.  It would be unwise
to make this decision solely on price, without understanding the trade
offs in productivity.

On Sat, Sep 25, 2010 at 1:07 PM, gabriel lorenzo <macglobalus <at> yahoo.com> wrote:
> IN CLUSTER COMPUTING, IS THE AMOUNT OF CORE THAT COUNTS?
> If I build a cluster with 8 motherboards with 1 single core each would it be the same as using just one
motherboard but with two quad core processors? I wanna build one of these but wanna save money and space and
if what counts is the amount of cores to process info I think fewer motherboards with dual six-core
processors is definitely cheaper just because I wont be needing that many mothers power supplies etc. thanks
>
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf <at> beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>
Permalink | Reply | 0 comments |
headers
Mark Hahn | 5 Oct 2010 03:44
Picon
Picon
Favicon

Re: Begginers question # 1

> IN CLUSTER COMPUTING, IS THE AMOUNT OF CORE THAT COUNTS?

no.  it's the application that counts.

> If I build a cluster with 8 motherboards with 1 single core each would it
> be the same as using just one motherboard but with two quad core
> processors?

of course not.  communication among cores on a single board
will certainly be faster than inter-board communication. 
it's the application that matters: how frequently do threads/ranks 
of the application communicate?  are messages small or large? 
can the app's communication be formulated as mostly-read sharing of data? 
these are all very much properties of the application, 
and they determine how suitable any particular hardware will be.

> I wanna build one of these but wanna save money and space and
> if what counts is the amount of cores to process info I think fewer
> motherboards with dual six-core processors is definitely cheaper just
> because I wont be needing that many mothers power supplies etc. thanks

power supplies aren't your main concern, since good ones are about 93%
efficient.  but going with more-core systems is, in general, a good idea.
mainly for amortization reasons: probably fewer disks, extraneous sutff
like video interfaces, fewer parts to fail, fewer systems to administer, etc.
there can be disadvantages to more-core systems too, since some of the parts
being shared (amortized) may be performance bottlenecks.

the sweet spots depends on what systems are in volume production - 
right now, 2-socket systems are the right building block in most cases.
(Continue reading)

Permalink | Reply | 1 comment |
headers
Joshua mora acosta | 5 Oct 2010 06:27
Picon

Re: Begginers question # 1

Hello Gabriel

Beginner's questions are usually the harder ones ;)

Without any personal interest, here you have an easy reading that should help
you break the ice

http://www.sun.com/x64/ebooks/hpc_for_dummies.pdf

In my opinion this (ie. HPC) is a very experimental field on both HW and SW so
your best way to learn is by getting on something very affordable, and trying
to use it as much as you can. For that it will be good to get familiarized
with profilers ( performance counter tools ) so you gain confidence in what
you do.
That will force you to learn what is capable the whole thing (specially your
application).
Then once the app can use it all and you find what part of the HW or/and SW is
limiting the performance, start being demanding in that direction (there are
many directions), but one thing at a time or very few at a time if you know
how each thing contributes.

That process is lengthy and the newsgroup could answer much better specific
questions rather than generic ones while you go through it.

At the end, well there is no end, just a continuous refactoring process of
"your own solution" that you impose to yourself while you try to keep up with
the technologies that will allow you to get to the next computational/science
challenge.

Best regards,
(Continue reading)

Permalink | Reply | 0 comments |
headers
John Hearns | 5 Oct 2010 08:05

Re: Begginers question # 1

On 25 September 2010 18:07, gabriel lorenzo <macglobalus <at> yahoo.com> wrote:
> IN CLUSTER COMPUTING, IS THE AMOUNT OF CORE THAT COUNTS?
> If I build a cluster with 8 motherboards with 1 single core each would it be the same as using just one
motherboard but with two quad core processors?

My response is that this is a list about building clusters from
commonly used PCs - a 'Beowulf'.
Gabriel, if you asked this question two years ago this list would be
very clear on giving you advice on building a cluster from 8
motherboards.

However, the era of multicore processors is now upon us, and as
Gabriel has found if you look at a simple metric - price per core -
you hit a sweet spot of 8 cores on one motherboard. As I have said
before - the day of the SMP system has returned.

OK - but now for my answer. The one motherboard system will be
excellent for you to learn parallel programming. Go out and buy it.

Now for the second part of my answer - the one motherboard system will
be inevitably limited in RAM, unless you are very, very rich. So the 8
motherboard cluster is still useful for those problems which need more
memory. It also scales better - you can add more systems. It also
suits applications which need to perform a lot of input/output - take
movie rendering for instance.

The 8 motherboard system will help you learn about cluster install
techniques - how to install the smae image on many systems, or how to
run systems with no disks, and also will teach a lot about networking
- as you fundamentally have to have a network to get it running, and
(Continue reading)

Permalink | Reply | 1 comment |
headers
Douglas Eadline | 5 Oct 2010 14:40
Favicon

Re: Begginers question # 1


> IN CLUSTER COMPUTING, IS THE AMOUNT OF CORE THAT COUNTS?
> If I build a cluster with 8 motherboards with 1 single core each would it
> be the same as using just one motherboard but with two quad core
> processors? I wanna build one of these but wanna save money and space and
> if what counts is the amount of cores to process info I think fewer
> motherboards with dual six-core processors is definitely cheaper just
> because I wont be needing that many mothers power supplies etc. thanks

First the short and easy answer: "It all depends"

Now the longer answer.

A single 8-way system has plenty of advantages and four 2-way or
eight 1-way systems certainly have a more overhead, cables, space etc.
If you want to play with parallel computing and MPI the 8-way
system will work just fine. (And yes, MPI works just fine on
SMP systems.) OpenMP is also an option in this case, but remember,
OpenMP will not "scale beyond the motherboard" (or at least
there are no guarantees) while MPI can. I recently did a whole
bunch of tests using both MPI and OpenMP a 12-way (dual 6-core)
SMP box I will be posting soon.

As I see it, one of the issues with the higher core counts is
memory contention. An 8-way parallel program that hits hard
on the memory may not scale as well as eight 1-way cores.
This is where "it all depends" comes into play because it
is very application dependent.

I have a small script that I run on multi-core systems
(Continue reading)

Permalink | Reply | 0 comments |
headers
Bogdan Costescu | 5 Oct 2010 15:23
Picon

Re: Broadcast - not for HPC - or is it?

On Fri, Sep 24, 2010 at 12:21 PM, Matt Hurd <matthurd <at> acm.org> wrote:
> I'm associated with a somewhat stealthy start-up.  Only teaser product
> with some details out so far is a type of packet replicator.

>From your description as well as from a quick look at the website, it
looks and smells like a hub - I mean a dumb hub, like those which
existed in the '90s before switching hubs (now called switches) took
over. If so, then HPC might not be a good target for you, as it has
long ago adopted switches for good reasons.

> Primarily focused on low-latency
> distribution of market data to multiple users as the port to port

HPC usage is a mixture of point-to-point and collective
communications; most (all?) MPI library use low level point-to-point
communications to achieve collective ones over Ethernet.. Another
important point is that the collective communications can be started
by any of the nodes - it's not one particular node which generates
data and then spreads it to the others; it's also relatively common
that 2 or more nodes reach the point of collective communication at
the same time, leading to a higher load on the interconnect, maybe
congestion.

What might be worth a try is a mixed network config where
point-to-point communications go through one NIC connected to a switch
and the collective communications that can use a broadcast go through
another NIC connected to your packet replicator. However, IMHO it
would only make sense if the packet replicator makes some guarantees
about delivery: f.e. that it would accept a packet from node B even if
a packet from node A is being broadcasted at that time; this packet
(Continue reading)

Permalink | Reply | 3 comments |
headers
John Hearns | 5 Oct 2010 15:40

Re: Broadcast - not for HPC - or is it?

On 5 October 2010 14:23, Bogdan Costescu <bcostescu <at> gmail.com> wrote:
>
> HPC usage is a mixture of point-to-point and collective
> communications; most (all?) MPI library use low level point-to-point
> communications to achieve collective ones over Ethernet.. Another
> important point is that the collective communications can be started
> by any of the nodes - it's not one particular node which generates
> data and then spreads it to the others; it's also relatively common
> that 2 or more nodes reach the point of collective communication at
> the same time, leading to a higher load on the interconnect, maybe
> congestion.

True indeed.
However this device might be very interesting if you redefine your
parallel processing paradigm.
How about problems where you send out identical datasets to (say) a
farm of GPUs.
Permalink | Reply | 1 comment |
headers
Glen Beane | 5 Oct 2010 15:54
Favicon

Re: Begginers question # 1


On Oct 4, 2010, at 9:44 PM, Mark Hahn wrote:

>> IN CLUSTER COMPUTING, IS THE AMOUNT OF CORE THAT COUNTS?
> 
> no.  it's the application that counts.
> 
>> If I build a cluster with 8 motherboards with 1 single core each would it
>> be the same as using just one motherboard but with two quad core
>> processors?
> 
> of course not.  communication among cores on a single board
> will certainly be faster than inter-board communication. 
> it's the application that matters: how frequently do threads/ranks 
> of the application communicate?  are messages small or large? 
> can the app's communication be formulated as mostly-read sharing of data? 
> these are all very much properties of the application, 
> and they determine how suitable any particular hardware will be.
> 
>> I wanna build one of these but wanna save money and space and
>> if what counts is the amount of cores to process info I think fewer
>> motherboards with dual six-core processors is definitely cheaper just
>> because I wont be needing that many mothers power supplies etc. thanks
> 
> power supplies aren't your main concern, since good ones are about 93%
> efficient.  but going with more-core systems is, in general, a good idea.
> mainly for amortization reasons: probably fewer disks, extraneous sutff
> like video interfaces, fewer parts to fail, fewer systems to administer, etc.
> there can be disadvantages to more-core systems too, since some of the parts
> being shared (amortized) may be performance bottlenecks.
(Continue reading)

Permalink | Reply | 0 comments |

Gmane