10
BESIII computing 王王王

BESIII computing

Embed Size (px)

DESCRIPTION

BESIII computing. 王贻芳. Peak Data volume/year. Peak data rate at 3000 Hz Events/year: 1*10 10 Total data of BESIII is about 2*640 TB. BESIII computing needs. CPU power Storage Network System software. CPU power needed for data reconstruction and simulation. - PowerPoint PPT Presentation

Citation preview

BESIII computing

王贻芳

Peak Data volume/year

• Peak data rate at 3000 Hz

• Events/year: 1*1010

• Total data of BESIII is about 2*640 TB

Event size(KB) Data volume(TB)

RAW 12 120

REC. 24 240

DST 2 20

MC-Rec 24 240

MC-DST 2 20

Total 640

BESIII computing needs

• CPU power

• Storage

• Network

• System software

CPU power needed for data reconstruction and simulation

• Four times reconstructions/year• Equivalent to a farm of 200 P4 1.6G• Analysis needs another farm of 100 P4 1.6 G • Maybe underestimated

MIPS/event Events (1010)

Total CPU (MIPS)

Event reconstruction

20 4 40,000

MC simulation 200 1 100,000

MC reconstruction

20 4 40,000

Total 180,000

Data type and storage media

• Store 3 reconstructions

• Virtual storage library

• Fast and automatic access

Data type Data volume(TB) device

Raw 240 Tape lib.

Rec 1440 Tape lib.

DST 120 Disk array

MC-Rec 1440 Tape lib. ?

MC-DST 120 Disk array

Tape reading/writing speed

• Online data recording(writing):

3000*12KB=36MBytes/s

• Reconstruction(reading/writing):

3000*24KB*5*2 = 760MBytes/s

• MC simulation ?

3000*24*2 = 152MBytes/s

It is almost impossible ! We should design our software framework carefully to minimize the data size !

Disk read/write access speed

• Data Reconstruction

3000*2KB*5 = 30 MB/s

• MC simulation

3000*2KB*5 = 30 MB/s

• Analysis

40*40Mb = 200 MB/s

main building to computer center

Network needs• From Online farm to computer center

3000*12KB = 36 MB/s

A dedicated Gbps network line for safety and stability

• Offline farm

network a bottleneck

• User analysis

> 40 users access disk files

Software system

• Mainly based on free software

• CERN library based

• Following latest development of HEP software

Main bottleneck• Tape reading/writing

use more CPU to reduce data size

• Network

within PC farm, main building CC

DST in main building ?

• Large scale storage library management

• Large scale PC farm:

stability, scalability, management