RCF Status Smooth operation over the course of last week –No reportable outages of the Mass Storage System (HPSS), the Tape Libraries and the Network PHENIX & RCF experts investigating rate limitation (cap at ~2.4 Gbps) observed between Counting House and RCF –Issue understood and solved (see next 3 slides) –Data received from PHENIX & STAR Counting Houses was safely migrated to Tape 30 min interference Sun/Mon night when other applications flooded Mass Storage System w/ requests –Looking at ways to implement request prioritization
2 Gbps Steady-state rate out of buffer boxes Iperf test 5 Gbps PHENIX CH => RCF rate limit (from last Time Meeting) Back-to-back stores Causing long transfer tails Initial tests indicate that there is no limit imposed by network & HPSS data movers
Magnetic Disk Basics Arm Assembly Arm Head CylinderSpindle Platter Track Read-ahead size is a critical parameter to achieve line-rate performance In particular with sequential I/O Caching in (memory) buffer Data
PHENIX Buffer Boxes (BB) sending Data to RCF Mass Storage System 7 th additional BB does not increase the transfer rate Source: Martin Purschke 2 Gbps Last Week 5 Gbps
3150 TB 1300 TB 1250 TB 3100 TB Raw Data Volume collected & archived 02/01/ TB 240 TB Physics PHENIX Raw Data STAR Raw Data
STAR Counting House RCF Network Hourly Data Migration Rate To Tape (GB/hour) 4 Gbps 2000 GB/h
PHENIX Counting House RCF Hourly Data Migration Rate To Tape (GB/hour) 2 TB/h 5 Gbps 2000 GB/h