David Malan teaching CS75 lecture 9, Scalability

https://youtu.be/-W9F__D3oY4

 

Storage

　　PATA, SATA, SAS (15,000 rpm), SSD,

　　RAID0 : striping, double throughput / size, no redundancy; RAID1 : mirror; RAID2 : bit-level striping with parity, not used today; RAID3 : bit-level striping with dedicated parity drive, not used today; RAID4 : block-level striping with dedicated parity ; RAID5 : block-level striping with dedicated parity, single drive failure tolerance; RAID6 : block-level striping with double parity, two drive failure tolerance

　　NAS, a file server, accessed through NFS, SMB etc.

　　SAN, a virtual disk, accessed through iSCSI, Fiber Channel etc.

 

Vertical scaling

 

Horizontal scaling

　　DNS round-robin load balancing, heavy vs. light requests

　　load balancer

　　　　Software-based LB: ELB, HAProxy, LVS etc.

　　　　Hardware-based : F5, Cisco, Citrix etc.

　　cookie/session affinity (sticky session)

　　　　separate session server (e.g. a redis or MySQL server for hash table)

　　　　LB inserts cookie content to remember backend-id

　　　　this is standard approach actually nowadays, AWS ELB (Elastic Load Balancer) provides this feature as service, http://docs.aws.amazon.com/elasticloadbalancing/latest/classic/elb-sticky-sessions.html, for example, you can specify a cookie name which will drive the distribution of requests.

Caching

　　DNS cache

　　MySQL Query Cache

　　memcached

 

Database Replication

　　master / slave : and you can make master write-only, and slaves obviously read-only

　　master / master - slave set up

 

Database Partitioning

　　mit.facebook.com.

　　LB, cross data center, geolocation distribution