60 likes | 222 Views
Managing Cloud. Jingren Zhou Microsoft Corp. Mission Impossible?. I can manage 16 machines. …. …. …. Database vs. MapReduce. Different in design principles Scale: x100 vs. x1000 or more Hardware: high-end vs. commodity Fault-tolerance: hardware vs. software
E N D
Managing Cloud Jingren Zhou Microsoft Corp.
Mission Impossible? . . . I can manage 16 machines … … … . . .
Database vs. MapReduce • Different in design principles • Scale: x100 vs. x1000 or more • Hardware: high-end vs. commodity • Fault-tolerance: hardware vs. software • AutoAdmin: add-on vs. built-in • Database: first knobs, then auto management • MapReduce: auto recovery/tuning is part of the system
SCOPE / Cosmos • A hybrid of parallel database and map-reduce system • SCOPE [VLDB’08]: • SQL-like declarative language • Relational data model • Database-like query optimization • Cosmos Storage System • Append-only distributed file system for storing petabytes of data • Optimized for sequential I/O • Data is compressed and replicated
Common Roles • System configuration • Resource management: setting quota on disk, CPU, network, etc. • Security implementation • General technical trouble shooting and consultation • Patch system • Plan growth and capacity • …
Different Focus • Database DBA • Focus on database/application performance turning • Database physical design and implementation • System Recovery • Backup implementation • Cloud Administrator • Main job: keep the system alive • Monitor all kinds of events/alerts • Ensure data availability/replication, etc. • Progressive system deployment/upgrade