The Problem

Customer processing of network management data take 23+ hours to complete for a 24 hour daya of collection. Any processing disruption results in run times exceeding 24 hours and results in full data sets being lost.

Situation Assessment

• Long processing times caused by numerous serial CPU-intensive processes.
• Occasional data errors not properly handled, causing processing to ‘hang’ and causing processing to exceed the critical 24-hour window.
• No capacity to add additional network management data sets of interest.

Our Solution

• Re-engineer serial processes into 43 parallel routines that run independently of each other.
• Partitioning means that, in the now-unlikely event of a stalled process, only that small data set is affected.
• Add fault detection and handling to significantly reduce the possibility of stalled processes.
• Implementation uses low-cost, open-source, high-throughput processing GRID running on hardened LINUX OS and X86 blade hardware.

Benefit

• Problem of dropped data sets completely eliminated.
• Solution processes 15x the amount of data in 1/12 the time – an almost 200x efficiency improvement.
• Solution is economically scalable and inherently less expensive to maintain.
• Data collection can be expanded to additional network elements with existing solution.