Traditional Database Architecture
•
Cluster Storage Volume → Storage nodes with locally attatche SSDs
•
Continuous backup to Amazon S3
한국 REGION, AZ 4
Scale-out, distributed, multi-tenant storage architecture
•
computing node <> storage node
•
six copies of data, two in each AZ to protect against AZ+1 failure modes
•
Storage volume segmented in 10GiB protected group (PG)
→ 분산저장으로 빠른 속도
•
Each PG contains six 10 Gib Segments, copies of the same data on different storage nodes
고가용성, tolerant of AZ+1 failures
•
4 out of 6 for writes
•
3 out of 6 for read (recovery)
•
기본적으로 3개의 AZ 운영
◦
Auto recovery (Segment node down시에도 auto recover가 수행되어 write도 가능)
•
Gossip protocol
Cluster topology
•
DB Cluster
◦
Writer 1, Reader 2 (autoscale 시 Reader를 ↑)
•
Storage volume shared with readers
Accessing your Aurora DB
•
Cluster End Point
◦
write end point + reader end point
◦
read/write 가능
•
Reader End Point
◦
읽기만 가능
◦
Auto scaling → reader endpoint LB (부하분산), endpoint가 늘어나는 것은 아님
Autoscaling을 통해 writer는 늘지않는다.
Aurora I/O profile compared
RDS
•
EBS Storage Attach
◦
주기적 백업 필요
•
복제 시 I/O issue
Aurora
•
storage customizing → 전용 storage volume이 따로 구성되어있음
◦
storage level fault 시에도 auto coverage 가능
•
분산저장
•
빠른 속도
S3에 데이터를 저장 && 백업 유지