This report reviews the tested configuration and summarizes the test procedure used to stress the Storage Tank testbed at CERN. In this data challenge, the performance of parallel sequential read and writes has been tested in conditions as close as possible to the 6th ALICE data challenge1. Storage Tank performance results of the 5-day and 7-day tests are presented and discussed as well as performance results under various failure scenarios. Based on these results, actions for future work are proposed. In the test workload, files of size 2GB are created and written by writer threads. Next, these files are read by reader threads and then deleted from the file system. Twelve writer clients are used with four and six writer threads each for the five-day and sevenday test, respectively. Similarly, 12 reader clients execute four reader threads each. All accesses are sequential with an application I/O request size of 1 MB. The storage consists of 15 iSCSI targets that export 53 iSCSI LUNs used as Storage Tank data volumes. Data volumes from each target are combined to form a Storage Pool such that there are 15 pools with a one-to-one mapping between pools and targets. Pools P1 through P8 are hosted by 200i targets. Pools P9 through P15 are hosted by iSCSI targets composed of a mixture of IBM x-series x345 and x335 machines with RAID controllers. A detailed view of the storage setup is provided in Section 4. 15 filesets are defined in the filesystem name-space. A one-to-one policy maps the 15 filesets to the 15 storage pools. As a result I/Os to a file in a particular fileset are isolated to a single storage pool and a single iSCSI target. The Arlo simulation program (a tool written by IBM Research for the test) is used to drive the workload. Arlo consists of ArloReader, ArloWriter and ArloCop. Upon a request from an ArloWriter, the ArloCop returns the complete path of the next file to be written. The path of the file determines in which of the 15 filesets the file will be created. Upon request from an ArloReader, the ArloCop returns the complete path of the next file to be read. The ArloReader deletes the file once it has read it entirely. Real-time performance monitoring of Storage Tank, generation of graphs and statistical analysis were done using STFS Perfmon (a tool written by IBM Research for the test).
By: R. Ananthanarayanan; P.L. Bradshaw; J. Gomez; Robert Haas; B. Henderson; C. Silvan; C. Tribolet
Published in: RZ3578 in 2005
LIMITED DISTRIBUTION NOTICE:
This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.
Questions about this service can be mailed to reports@us.ibm.com .