lecture #8 giant-scale services cs492 special topics in computer science: distributed algorithms and...

39
Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Upload: shon-johnston

Post on 11-Jan-2016

224 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Lecture #8Giant-Scale Services

CS492 Special Topics in Computer Science:Distributed Algorithms and Systems

Page 2: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Lessons from Giant-Scale Services

Eric BrewerUC Berkeley and Inktomi

IEEE Internet Computing July/August 2001

Page 3: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

3

Trade-Offs

• Memory• CPU speed• Hard Disk• Operating Systems

Page 4: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

“Giant-Scale” Services

• Key real-world challenges– High avaiability– Evolution– Growth

4

Page 5: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Advantages of Giant-Scale Services

• Access anywhere, anytime• Availability via multiple devices• Groupware support• Lower overall cost• Simplified service updates

Page 6: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Basic Model for Giant-Scale Services

Page 7: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Assumptions

• Service provide has limited control over clients/network

• Queries drive the service• Read-only queries greatly outnumber updates

Page 8: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Google Data Centers

8

Craig Mitchelldyer/Getty ImagesBrian Nettles/The Post and Courier

Page 9: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Load Management

• Round-Robin DNS• “Layer 4” switch• “Layer 7” switch

Page 10: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Comparison

Page 11: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Availability Metrics

• Uptime– Uptime = (MTBF – MTTR) / MTBF

• Mean-time-between-failure (MTBF)• Mean-time-to-repair (MTTR)

• Yield = queries completed/queried offered• Harvest = data available/complete data

Page 12: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

DQ Principle

• Data per query x queries per second -> con-stant– Amount of data that has to be moved per sec

Page 13: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

13

Yield vs Harvest

• Replicated– Map faults to reduced capacity– Yield drops 50% when half dies

• Partitioned– Map faults to reduced harvest– Yield remains, but harvest drops 50%

• But both halve in DQ– Replicated: Q– Partitioned: D

Page 14: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

14

Bottom Line Is …

• When you double capacity, make sure your DQ doubles

Page 15: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Graceful Degradation

• Peak-to-avg ratio = 1.6:1 to 6:1• Single-event bursts can generate far above-average

traffic• Some faults are not independent

• Explicit process of managing the effect of saturation– Cost-based access control– Priority or value-based access control– Reduced data freshness

Page 16: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Online Evolution and Growth

• “Internet time” - frequent product releases• Maintenance and upgrades

= controlled failures= “online evolution”

• Fast reboot• Rolling upgrade• Big flip

Page 17: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Lessons

• Get the basics right• Decide on your availability metrics• Focus on MTTR at least as much as MTBF• Understand load redirection during faults• Graceful degradation• Use DQ analysis on all upgrades• Automate upgrades as much as possible

Page 18: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

18

Questions from last class

• How many copies of a chunk?• How large is write delay?

Page 19: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

The Google File System

Sanjay Ghemawat, Howard Gobioff, Shun-Tak Leung

SOSP 2003

Page 20: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

20

What do you remember about FS?

Page 21: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

21

Distributed File System?

• How is it different from a local file system?

Page 22: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Motivation• Google needed a good distributed file system– Redundant storage of massive amounts of data on

cheap and unreliable computers

• Why not use an existing file system?– Google’s problems are different from anyone else’s

• Different workload and design priorities

– GFS is designed for Google apps and workloads– Google apps are designed for GFS

Page 23: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Assumptions

• High component failure rates– Inexpensive commodity components fail all the time

• “Modest” number of HUGE files– Just a few million– Each is 100MB or larger; multi-GB files typical

• Files are write-once, mostly appended to– Perhaps concurrently

• Large streaming reads• High sustained throughput favored over low latency

Page 24: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

GFS Design Decisions• Files stored as chunks

– Fixed size (64MB)• Reliability through replication

– Each chunk replicated across 3+ chunkservers• Single master to coordinate access, keep metadata

– Simple centralized management• No data caching

– Little benefit due to large data sets, streaming reads• Familiar interface, but customize the API

– Simplify the problem; focus on Google apps– Add snapshot and record append operations

Page 25: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

GFS Architecture• Single master• Mutiple chunkservers

…Can anyone see a potential weakness in this design?

Page 26: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Single master

• From distributed systems we know this is a:– Single point of failure– Scalability bottleneck

• GFS solutions:– Shadow masters– Minimize master involvement

• never move data through it, use only for metadata– and cache metadata at clients

• large chunk size• master delegates authority to primary replicas in data mutations

(chunk leases)

• Simple, and good enough!

Page 27: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Metadata (1/2)

• Global metadata is stored on the master– File and chunk namespaces– Mapping from files to chunks– Locations of each chunk’s replicas

• All in memory (64 bytes / chunk)– Fast– Easily accessible

Page 28: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Metadata (2/2)

• Master has an operation log for persistent logging of critical metadata updates– persistent on local disk– replicated– checkpoints for faster recovery

Page 29: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Mutations Mutation = write or append

must be done for all replicas Goal: minimize master involvement Lease mechanism:

master picks one replica asprimary; gives it a “lease” for mutations

primary defines a serial order of mutations

all replicas follow this order Data flow decoupled from

control flow

Page 30: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Atomic record append

• Client specifies data

• GFS appends it to the file atomically at least once– GFS picks the offset– works for concurrent writers

• Used heavily by Google apps– e.g., for files that serve as multiple-producer/single-con-

sumer queues

Page 31: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Relaxed consistency model (1/2)

• “Consistent” = all replicas have the same value• “Defined” = replica reflects the mutation, consistent

• Some properties:– concurrent writes leave region consistent, but possibly un-

defined – failed writes leave the region inconsistent

• Some work has moved into the applications:– e.g., self-validating, self-identifying records

Page 32: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Relaxed consistency model (2/2)

• Simple, efficient– Google apps can live with it– what about other apps?

• Namespace updates atomic and serializable

Page 33: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Master’s Responsibilities (1/2)

• Metadata storage• Namespace management/locking• Periodic communication with chunkservers– give instructions, collect state, track cluster health

• Chunk creation, re-replication, rebalancing– balance space utilization and access speed– spread replicas across racks to reduce correlated failures– re-replicate data if redundancy falls below threshold– rebalance data to smooth out storage and request load

Page 34: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Master’s responsibilities (2/2)

• Garbage Collection– simpler, more reliable than traditional file delete– master logs the deletion, renames the file to a hidden

name– lazily garbage collects hidden files

• Stale replica deletion– detect “stale” replicas using chunk version numbers

Page 35: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Fault Tolerance

• High availability– fast recovery

• master and chunkservers restartable in a few seconds

– chunk replication• default: 3 replicas.

– shadow masters

• Data integrity– checksum every 64KB block in each chunk

Page 36: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Performance

Page 37: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Deployment in Google

• Many GFS clusters• hundreds/thousands of storage nodes each• Managing petabytes of data• GFS is under BigTable, etc.

Page 38: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Conclusion

• GFS demonstrates how to support large-scale pro-cessing workloads on commodity hardware– design to tolerate frequent component failures– optimize for huge files that are mostly appended and read– feel free to relax and extend FS interface as required– go for simple solutions (e.g., single master)

• GFS has met Google’s storage needs… it must be good!

Page 39: Lecture #8 Giant-Scale Services CS492 Special Topics in Computer Science: Distributed Algorithms and Systems

Discussion

• How many sys-admins does it take to run a system like this?– much of management is built in

• Is GFS useful as a general-purpose commercial product?– small write performance not good enough?– relaxed consistency model