LHC@home

I suspect that many of us joined LHC@home with
the hope of helping LHC with the actual data
analysis.

The LHC (will) generate pentabytes of data from
its detectors. However, it is stated that the
shear size of these data and network bandwidth
limitations prevent running actual data analysis
for any of the detectors with LHC@home.

I have found no information on the LHC@home web
site that tells me what part of the LHC, or what
level of management, has committed itself to
using LHC@home.

So, *is* there a commitment to LHC@home by
CERN as a whole? Do the director general
or any of the departments have a position
on using LHC@home? We represent a potentially
huge computing resource. Is there a commitment
of resources beyond volunteer programming and use?

I cannot imagine that CERN, with its obviously
huge need for computing, could not find a way
to allow us to participate in a larger way. I
know I, for one, would appreciate it.

Please note, I thank CERN for allowing us to
participate even in the current limited way!

So, *is* there a commitment to LHC@home by
CERN as a whole? Do the director general
or any of the departments have a position
on using LHC@home? We represent a potentially
huge computing resource. Is there a commitment
of resources beyond volunteer programming and use?

The information you ask for, can be found in the Café message board, under the \"whatever happened...\" thread.

Look especcially for the bigmac posts. Happy reading ;)

From my understanding we are helping them \\\"tune\\\" the magnets and such [via simulation] so that the LHC can do more work faster and they have less time \\\"tuning\\\" to make things work well.

We are an operations utility rather than a research adjunct.

The work is important, but not core research.

The fact that the LHC has started to return results so soon after the restart shows that what we do works!

So it seems we really do not have much support for LHC@home.

That is unfortunate. Maybe someone should calculate the
actual computing power available through LHC@home and
let the department heads or the director general (or the
people in the member nations WHO ARE ACTUALLY PAYING
THE BILLS) how much they could save or how much more
could be done if LHC@home were actually being used.

We volunteers could contact our representatives and
suggest that CERN could well be using this resource
more efficiently (and save money/get more work done)?

In the SETI@home Number Crunching section there is a cruncher with 1400 computers connected by Infiniband, that is a supercomputer, running SETI@home to keep his machines working even when he has no customers. I believe this to be an example of Cloud Computing and I believe his grid could be useful to LHC. But he probably knows only SETI@home and never heard of LHC@home.
Tullio

Dear Carolina,

Just to let you know that there *is* a project under development here for running \\\"real physics\\\" jobs for the LHC experiments on LHC@home. Live analysis of big data sets with high I/O rates will NOT be possible as BOINC is not suitable for this. But event simulation and some event reconstruction both look very interesting.

The project has been developed with quite a low profile up to now by a few people who are fully aware of its potential interest, both for CERN and for many of you volunteers, and we are just beginning to inform the experiments and the CERN management about it. It is technically quite sophisticated, depending on virtualization and some fairly recent work done here to support that, so we want to be sure that it works well before going seriously public. As you know, a lot of work is needed to turn a working prototype (which we have) into a viable operation.

Thanks a lot for your interest and support,

Ben

PS: I was the technical coordinator of the original LHC@home effort starting in 2004 - seems a long time ago!

---

I am running SETI@home both on my Linux box and on a virtual machine with OpenSolaris obtained installing the SUN VirtualBox on the Linux host. The virtual machine has a BOINC client and an app by a developer called Dotsch and SETI is the only project for which I could find both a Solaris BOINC client and an application. Although it runs well it is about 4 times slower that the Linux app, perhaps because the guest OS sees my dual core Opteron 1210 only as a PentiumPro+mmx, while it is SSE3 capable. Dotsch says that this is the fault of the virtualization software, not of his BOINC client.
Tullio

Hi Tullio,

This is interesting but it needs a proper technical discussion and that\'d be pretty off-topic, so can we take it offline?

Ben (b.segal AT cern.ch)

I am running SETI@home both on my Linux box and on a virtual machine with OpenSolaris obtained installing the SUN VirtualBox on the Linux host. The virtual machine has a BOINC client and an app by a developer called Dotsch and SETI is the only project for which I could find both a Solaris BOINC client and an application. Although it runs well it is about 4 times slower that the Linux app, perhaps because the guest OS sees my dual core Opteron 1210 only as a PentiumPro+mmx, while it is SSE3 capable. Dotsch says that this is the fault of the virtualization software, not of his BOINC client.
Tullio

---

Of course,yes.

Dear Carolina,

Just to let you know that there *is* a project under development here for running real physics jobs for the LHC experiments on LHC@home. Live analysis of big data sets with high I/O rates will NOT be possible as BOINC is not suitable for this. But event simulation and some event reconstruction both look very interesting.

The project has been developed with quite a low profile up to now by a few people who are fully aware of its potential interest, both for CERN and for many of you volunteers, and we are just beginning to inform the experiments and the CERN management about it. It is technically quite sophisticated, depending on virtualization and some fairly recent work done here to support that, so we want to be sure that it works well before going seriously public. As you know, a lot of work is needed to turn a working prototype (which we have) into a viable operation.

Thanks a lot for your interest and support,

Ben

PS: I was the technical coordinator of the original LHC@home effort starting in 2004 - seems a long time ago!

WOW! I am sure that there are many lurking around who will be anxiously waiting to hear more about this effort and looking forward to crunching numbers for the LHC project.

Thanks to all those involved for their efforts.

Just to let you know that there *is* a project under development here for running real physics jobs for the LHC experiments on LHC@home. Live analysis of big data sets with high I/O rates will NOT be possible

...

Hey all,

Could I suggest a collaborative effort under (e.g.) .Net? The Windows/Intel platform is established but those lucky few geniuses (genii?) with Mono running on Linux boxes could contribute to the overall build.

Whilst such an app will not run as fast as native code, it will speed the development; any good optimisations can be left to the compiler, CUDA/Stream/OpenCL capability added and all peer-reviewed through Sourceforge/SubVersion.

Straight-line speed aside, is wider uptake with support for multi-core and/or capable graphics not attractive? Most of my BOINC installs are on laptops with WinXP on Core 2 Duo hardware with variable GPU ability; most desktops are CPU-heavy with little need for GPU support.

Attempting to rally popular support, my name is Spartacus! Are you with me?!

Ben

---

About number crunching and GPU/CUDA.

In my opinion, there is a careful approach that needs to be taken with tasks targeted for GPU crunching. I an doing tasks for both SETI and Einstein and, while both projects will crunch on my GPU, I only allow SETI tasks to be run on the GPU. The main reason is that Einstein tasks are very long while the vast majority of the SETI tasks enabled to run on the GPU are relatively short, e.g., 15 minutes to 2.5 hours. The long Einstein tasks just simply grab the GPU and hold on to it for the total length of the task (which in the case of Einstein can be 14 or 15 hours). In other words, the Einstein tasks do not play well with others when running on the GPU. My experience shows that by limiting the Einstein tasks to only run on the CPU(s) (an Intel Core 2 Quad) I realize a higher overall throughput(or so it seems).

I do not know how SETI creates the shorter tasks. Also, it appears that Einstein allows all of their tasks to run on either the CPU or the GPU while SETI designates which tasks are targeted to the CPU(s) and which ones are targeted to run on the GPU.

To Ben (the other Ben...) and Tom Mehrens:

Please note that the new LHC project I have referred to is not designed to run on GPUs but uses virtualization.

Thanks anyway for your interest,

Ben (Segal)

Just to let you know that there *is* a project under development here for running real physics jobs for the LHC experiments on LHC@home. Live analysis of big data sets with high I/O rates will NOT be possible

...

Hey all,

Could I suggest a collaborative effort under (e.g.) .Net? The Windows/Intel platform is established but those lucky few geniuses (genii?) with Mono running on Linux boxes could contribute to the overall build.

Whilst such an app will not run as fast as native code, it will speed the development; any good optimisations can be left to the compiler, CUDA/Stream/OpenCL capability added and all peer-reviewed through Sourceforge/SubVersion.

Straight-line speed aside, is wider uptake with support for multi-core and/or capable graphics not attractive? Most of my BOINC installs are on laptops with WinXP on Core 2 Duo hardware with variable GPU ability; most desktops are CPU-heavy with little need for GPU support.

Attempting to rally popular support, my name is Spartacus! Are you with me?!

Ben

---

Dear Carolina,

Just to let you know that there *is* a project under development here for running \\\\\\\\\\\\\\\"real physics\\\\\\\\\\\\\\\" jobs for the LHC experiments on LHC@home. Live analysis of big data sets with high I/O rates will NOT be possible as BOINC is not suitable for this. But event simulation and some event reconstruction both look very interesting.

The project has been developed with quite a low profile up to now by a few people who are fully aware of its potential interest, both for CERN and for many of you volunteers, and we are just beginning to inform the experiments and the CERN management about it. It is technically quite sophisticated, depending on virtualization and some fairly recent work done here to support that, so we want to be sure that it works well before going seriously public. As you know, a lot of work is needed to turn a working prototype (which we have) into a viable operation.

Thanks a lot for your interest and support,

Ben

PS: I was the technical coordinator of the original LHC@home effort starting in 2004 - seems a long time ago!

Thank you, Ben! It seems there *is* something for
which to look forward! Do you think there will be
a time when LHC@home projects might have active support
from your administration?

Thank you for all your efforts!

-- Carolina Calling