Speaker
Description
Jefferson Lab has developed and operated Jasmine, its tape storage system, since 1999. Over the life of the system, design iterations have benefitted from advances in disk and network performance. Now, as large-scale data processing is increasingly performed on distributed infrastructure using community supported software tools, we detail how Jasmine’s software strategy has been refocused. Jasmine’s core function as a high performance Tape IO system has been realigned as a deep storage component of a distributed storage ecosystem. We demonstrate how this redesign has simplified aspects of the system and provided significant performance benefits, including robust support for streaming data and small file handling without tape performance penalty.
Jasmine’s core mission is to provide high performance tape read and write access with an emphasis on data integrity. Support for Jasmine’s internal disk buffering and management is now implemented with XRootD clusters. We describe how the composition of several non-hierarchical disk clusters is used to play to Jasmine’s Tape IO strengths, making tape drive performance more predictable in several example cases with divergent patterns of IO performance and improving the overall user experience.
| Consider for long presentation | No |
|---|
