Title: Converting ASGARD into a MC-Farm for Particle Physics
1Converting ASGARD into a MC-Farm for Particle
Physics
- Beowulf-Day 17.01.05
- A.Biland IPP/ETHZ
2Beowulf Concept
Three Main Components
3Beowulf Concept
Three Main Components
CPU Nodes
4Beowulf Concept
Three Main Components
CPU Nodes
Network
5Beowulf Concept
Three Main Components
CPU Nodes
Network
Fileserver
6Beowulf Concept
Three Main Components
CPU Nodes
Network
Fileserver
? ?
? How much of the (limited)
money to spend for what ??
7Beowulf Concept
Intended (main) usage
Eierlegende Woll-Milch-Sau (one size fits
everything) Put equal amount of money into each
component gt ok for (almost) any possible use,
but waste of money for most applications
8Beowulf Concept
Intended (main) usage
CPU-bound jobs with limited I/O and interCPU
communication
80 10
10
ASGARD , HREIDAR-I
9Beowulf Concept
Intended (main) usage
Jobs with high interCPU communication needs
(Parallel Proc.)
50 40
10
HREIDAR-II
10Beowulf Concept
Intended (main) usage
Jobs with high I/O needs or large datasets
(Data Analysis)
50 10
40
11Fileserver Problems
a) Speed (parallel access)
Inexpensive Fileservers reach disk-I/O 50
MB/s 500 single-CPU jobs gt 50 MB/s /500 jobs
100kB/s /job (as an upper limit typical values
reached much smaller) Using several Fileservers
in parallel -- difficult data management
(where is which file ?) use
parallel filesystems ? -- hot spots (all jobs
want to access same dataset ) data
replication gt
12Fileserver Problems
a) Speed (parallel access)
How (not) to read/write the data Bad NFS
(constant transfer of small chunks of data)
gt always disk repositioning gt disk-I/O --gt
0 (somewhat improved with large cache
(gtgt100MB) in memory if write-cache
full long time to flush to disk gt server
blocks) ok rcp (transfer of large blocks
from/to local /scratch ) /scratch
rather small on ASGARD if many jobs
want to transfer at same time ??? Best
fileserver initiates rpc transfers on request
user discipline, not very transparent,
13Fileserver Problems
b) Capacity .
500 jobs producing data Each writes 100kB/s gt
50 MB/s to Fileserver gt 4.2 TB
/ day !
14Particle Physics MC
- Need huge amount of statistically independent
events - events gtgt CPUs gt embarassingly parallel
problem - gt 5x500 MIPS as
good as 1x2500 MIPS -
- Usually two sets of programs
- Simulation produce huge, very detailed MC-files
- (adapted standard packages GEANT,
CORSKA, - b) Reconstruction read MC-files, write smaller
reco-files - selected events, physics data
- (special SW developed by each
experiment) - Mass-Production only reco-files needed
- gt combine both tasks in one job, use /scratch
15ASGARD Status
10 frames
/home
24 nodes/ frame
/work
/arch
Local disk per node 1GB / 1GB swap 4GB /scratch
16ASGARD Status
10 frames
/home
24 nodes/ frame
/work
/arch
Local disk per node 1GB / 1GB swap 4GB /scratch
Needed
Fileserver bandwidth, capacity /scratch
guaranteed space/job
17ASGARD Upgrade
4x 400GB SATA, RAID-10 (800GB usable) / frame
10 frames
/home
24 nodes/ frame
/work
/arch
Local disk per node 0.2GB / 0.3GB swap 2.5GB
/scr1 2.5GB /scr2
18ASGARD Upgrade
4x 400GB SATA, RAID-10 (800GB usable) / frame
10 frames
/home
24 nodes/ frame
/work
/arch
Local disk per node 0.2GB / 0.3GB swap 2.5GB
/scr1 2.5GB /scr2
Adding 10 Fileservers (65kFr), ASGARD can serve
2years as MC Farm and GRID Testbed