vdas_20121122
by
Marie Anne Bizouard
—
last modified
2012-11-23 09:47
VDAS
1. Announcements
2. Focus session: Online computing upgrade
- MBTA/DAQ needs [Benoit/Loic] Slides: VIR-0459A-12 [slide about VDQ computing needs]
- EGO online upgrade [Stefano]: Online issues
- Cluster virtualization [Giuseppe]: Virtualization
3. AOB
Minutes:
Attendees: MAB [minutes], Elena, Alberto, Giuseppe, Stefano, Benoit, Frederique, Cristiano, Loic, Didier, Alain, Emmanuel
Loic presentation:
- MBTA/DAQ needs from ER2.
Q: 32 cores machine is that enough for ADE?
A: ER2 used theoritical ADE sensitivity curves, but MBTA run with a 30 Hz lower frequency cutoff. Optimal sensivity curves will require more computing power, but we do not know how much now. Yes, needs will correspond to more than a 32 cores machine, but it is difficult to give real requirements now. On the other hands, should not be wrong by 1 order of magnitude.
Q: Is 1 single machine a problem if needs scale up?
A: right now 1 single machine is enough to perform the 3 IFOs processing. In case of bandwidth increase, the first thing to do would be to spread the 3 IFOs processing over 3 machines. Still some margin.
- Didier's Detector Monitoring + VDQ:
C: Input / output information to be completed: how many channels? shared memory, which streams, access to database (IO), access to outside world (LIGO), etc ...
Stefano: need to know how many applications will have similar interfaces? Extract a workflow model.
-> Action item: move Didier's table to a wiki page + complete it with more information + other applications
- Guiseppe presentation:
olserver31 --> 37 are already virtual machines. Applications are running on a physical host. 2 hosts run 6 olservers.
2-3% of memory overhead per virtual machine. Memory is assigned to each VM. 2-3% of 2GB.
Represent advantages in terms of management. Requires testing to be sure it can handle all requirements.
Q (Loic): test of MBTA on a virtualized machines?
A: See later after Stefano's presentation
- Stefano presentation:
Description of the different features of the cascina computing elements + a collection of key issues.
Q (MAB): are we ready to sketch out the AdV Cascina infrastructure in the AdV computing plan document (needs to be ready for march)?
A (Stefano): key elements are not going to change. Implementation are not totally known
Conclusions:
1/ Agreement to purchase (asap):
- 1 many cores (24 cores - 4 GB / core) for farmnxx replacement (long term goal) +
- 1 32 cores + 4 GB / core machine for DAQ/MBTA tests
References and characteristics of the 2 machines to be bought should be circulated for final validation in the VDAS list.
For the MBTA/DAQ machine: DELL PowerEdge M820 with the equipment proposed by Stefano and Giuseppe during the Virgo week in the corridor (!) is ok for mbta/DAQ, but references should be circulated.
2/ Applications needs definition --> on the wiki (MAB will send an email to vdas).
2. Focus session: Online computing upgrade
- MBTA/DAQ needs [Benoit/Loic] Slides: VIR-0459A-12 [slide about VDQ computing needs]
- EGO online upgrade [Stefano]: Online issues
- Cluster virtualization [Giuseppe]: Virtualization
3. AOB
Minutes:
Attendees: MAB [minutes], Elena, Alberto, Giuseppe, Stefano, Benoit, Frederique, Cristiano, Loic, Didier, Alain, Emmanuel
Loic presentation:
- MBTA/DAQ needs from ER2.
Q: 32 cores machine is that enough for ADE?
A: ER2 used theoritical ADE sensitivity curves, but MBTA run with a 30 Hz lower frequency cutoff. Optimal sensivity curves will require more computing power, but we do not know how much now. Yes, needs will correspond to more than a 32 cores machine, but it is difficult to give real requirements now. On the other hands, should not be wrong by 1 order of magnitude.
Q: Is 1 single machine a problem if needs scale up?
A: right now 1 single machine is enough to perform the 3 IFOs processing. In case of bandwidth increase, the first thing to do would be to spread the 3 IFOs processing over 3 machines. Still some margin.
- Didier's Detector Monitoring + VDQ:
C: Input / output information to be completed: how many channels? shared memory, which streams, access to database (IO), access to outside world (LIGO), etc ...
Stefano: need to know how many applications will have similar interfaces? Extract a workflow model.
-> Action item: move Didier's table to a wiki page + complete it with more information + other applications
- Guiseppe presentation:
olserver31 --> 37 are already virtual machines. Applications are running on a physical host. 2 hosts run 6 olservers.
2-3% of memory overhead per virtual machine. Memory is assigned to each VM. 2-3% of 2GB.
Represent advantages in terms of management. Requires testing to be sure it can handle all requirements.
Q (Loic): test of MBTA on a virtualized machines?
A: See later after Stefano's presentation
- Stefano presentation:
Description of the different features of the cascina computing elements + a collection of key issues.
Q (MAB): are we ready to sketch out the AdV Cascina infrastructure in the AdV computing plan document (needs to be ready for march)?
A (Stefano): key elements are not going to change. Implementation are not totally known
Conclusions:
1/ Agreement to purchase (asap):
- 1 many cores (24 cores - 4 GB / core) for farmnxx replacement (long term goal) +
- 1 32 cores + 4 GB / core machine for DAQ/MBTA tests
References and characteristics of the 2 machines to be bought should be circulated for final validation in the VDAS list.
For the MBTA/DAQ machine: DELL PowerEdge M820 with the equipment proposed by Stefano and Giuseppe during the Virgo week in the corridor (!) is ok for mbta/DAQ, but references should be circulated.
2/ Applications needs definition --> on the wiki (MAB will send an email to vdas).