Lecture 6 – Google File System (GFS)

CSE 490h – Introduction to Distributed Computing, Winter 2008

Except as otherwise noted, the content of this presentation is licensed under the Creative Commons Attribution 2.5 License.

Previous Classes Why Distribute? MapReduce

Why Implementation

Details of RPC Other Large Systems: Web Services But… how is all this data stored?

Distributed File Systems Trade Offs in Distributed File Systems

PerformanceScalabilityReliabilityAvailability

How do you decide?

Two Core ApproachesSuper Computer?Or many cheap computers?

Motivation Google went the cheap commodity route…

Lots of data on cheap machines! Yikes!

Why not use an existing file system?Unique problemsGFS is designed for Google apps and workloadsGoogle apps are designed for GFS

Assumptions

Failures are the norm Detect errors, recover, tolerate faults, etc Software errors (we’re only human)

Huge files Multi-GB files are common But there aren’t THAT many files

Assumptions (cont.)

Mutations are typically appending new dataRandom writes are rareOnce written, files are only read, and typically

sequentiallyOptimize for this!

Large consecutive reads, small random reads

Want high sustained bandwidth – low latency is not that important

Google is designing apps AND file system

GFS Interface Supports usual commands

Create, delete, open, close, read, write Snapshot

Copies a file or a directory tree Record Append

Allows multiple concurrent appends to same file

GFS Architecture Single master Multiple chunkservers

…Can anyone see a potential weakness in this design?

Architecture Files divided into fixed-sized chunks (64 MB)

Each chunk gets a chunk handle from master Stored as linux files

One Master Maintains all filesystem metadata Talks to each chunksever periodically

Multiple Chunkservers Store chunks on local disks No caching of chunks. (Why not?)

Multiple Clients Clients talk to master for metadata operations Read / write data from chunkservers

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 (64 MB) master delegates authority to primary replicas in data mutations

(chunk leases)

Simple, and good enough!

Master’s responsibilities (1/2) Metadata storage Namespace management/locking Periodic communication with chunkservers

give instructions, collect state, track cluster health

Garbage Collection

Master’s responsibilities (2/2) Chunk creation

Place new replicas on chunkservers with below average disk-space utilization

Limit number of recent creations on each chunk server Spread replicas across racks

Re-Replicate when replicas fall below user goal How do you assign priorities?

Periodic rebalancing Better disk space usage Load balancing

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 Any problems?

Metadata (2/2)

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

Garbage Collection First… how would you do it?

Garbage Collection How to…

Master logs deletion immediately, and renames to hidden file

Lazily garbage collects hidden files via re-scans Also identifies orphaned chunks Why?

Simple and reliable when failures are common Merges storage reclamation into background activities Delay is safety net against background activities

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 fromcontrol flow

Atomic record append

Client specifies data

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

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

consumer queues

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

undefined failed writes leave the region inconsistent

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

Relaxed consistency model (2/2) Simple, efficient

Google apps can live with it what about other apps?

Namespace updates atomic and serializable

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

Deployment in Google

50+ GFS clusters Each with thousands of storage nodes Managing petabytes of data GFS is under BigTable, etc.

Conclusion

GFS demonstrates how to support large-scale processing 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!

Discussion Is GFS useful as a general-purpose commercial

product?