Naslov
Upravljanje i optimizacija trajne pohrane podataka u ALICE eksperimentu
(Management and optimization of mass data storage system for the ALICE experiment)

Autori
Vicković, Linda

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, doktorska disertacija

Fakultet
Fakultet elektrotehnike, strojarstva i brodogradnje

Mjesto
Split

Datum
27.06

Godina
2007

Stranica
138

Mentor
Gotovac, Sven

Ključne riječi
spremišna mreža; simulacija s diskretnim događajima; ALICE DAQ; sustav za trajnu pohranu podataka; simulacija diska; simulacija sustava diskova; podešavanje radnog opterećenja; optimizacija rasporeda podataka; veličina bloka u sustavu diskova
(storage area network; discrete event simulation; ALICE DAQ; mass storage system simulation; disk simulation; disk array simulation; workload tuning; data layout optimization; stripe size for disk array)

Sažetak
ALICE (A Large Ion Collider Experiment) is one of the four general-purpose experiments that will study physics at the Large Hadron Collider (LHC) at CERN, the European Organization for Nuclear Research. During data taking, ALICE will produce a dataflow of 15 GB/s that will need further processing and filtering by the Data Acquisition System (DAQ) and High Level Trigger (HLT). As a result of the processing, the bandwidth of the outgoing dataflow from the DAQ will be reduced to 1.25 GB/s for archiving in the Mass Storage System (MSS). Such throughput will result in about 1.5 PB of data per year. To satisfy the requirements for storage capacity and input bandwidth, the MSS is organized on two levels. The first is responsible for the real time data storage delivered by the ALICE DAQ and is called the Transient Data Storage (TDS). It has a capacity of eight hours of data taking. The second level, the so called Permanent Data Storage (PDS) should provide enough storage capacity for all data acquired during the lifetime of the experiment. The scope of this thesis is the modelling of the TDS level of the ALICE MSS. With the goal of exploring how the TDS's performance is affected by the individual system components, a dynamic, discrete event (DE) simulation model of the system has been developed. The model described in this thesis is based on the modular, bottom-up approach. First, the disk drive model is developed and is then used as an elementary module to create a disk array simulation model. The disk array model is then used to create a simulation model of the storage area network (SAN) used as a TDS. Another important aspect of the system simulation, discussed in this dissertation, is the definition of the system workload. In this case it is defined by the requirements of the ALICE TDS. The model of disk drive, disk array and SAN presented here predicts the system behaviour in terms of achieved bandwidth and all models have been validated with real measurements. The differences between the measured and simulated values vary between 1.5 % and 4 % depending on the simulated component and the workload type. The model developed for this thesis is finally used to explore the influence of different parameters (data file size, record size, stripe unit size, disk number in disk array, file system, etc.) on the performance of the system. It is also used to predict the effect of changes to the existing system and to predict the performance of the system under varying sets of circumstances. The methods described in this thesis can be disposed to provide an estimation procedure for the behaviour of any storage system based on disk drives and disk arrays.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA