GFS

Google

Servers are a mix of commodity machines and machines specifically designed for Google Not necessarily the fastest Purchases are based on the best

performance per dollar Uses customized versions of Linux

Google

Servers are put into racks and connected through a 1 GB Ethernet

Racks are organized into clusters connected and connected to a gigabit cluster switch

A data center consists of clusters Clusters run an application/service

Google

Google has numerous data centers scattered around the world (2012) 11 are found in the United States + 3 being

built 1 in Canada 17 international (outside of US and

Canada) No one is sure of how many and where

Motivation Google needs a good distributed file system

Redundant storage of massive amounts of data oncheap 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

Google Workload Characteristics

Most files are mutated by appending new data – large sequential writes

Random writes are very uncommon Files are written once, then they are

only read Reads are sequential Large streaming reads and small

random reads High sustained throughput favoured

over low latency

Google Workload Characteristics

Google applications: Data analysis programs that scan through

data repositories Data streaming applications Archiving Applications producing (intermediate)

search results

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

GFS Architecture Single master for a cluster Multiple chunkservers

Chunks

Files are divided into fixed-size chunks of 64 MB

Each chunk has an identifier, chunk handle, assigned by the master at the time of chunk creation Consists of file name, chunk number

Each chunk is replicated 3 times

GFS Architecture

Chunkservers

Store chunks on local disks as Linux files

Read/write chunk data specified by a chunk handle and byte range

GFS Architecture

Master Stores metadata

The file and chunk namespaces Mapping from files to chunks Locations of each chunk’s replicas (referred

to as chunk locations) Interacts with clients Creates chunk replicas

Master Orchestrates chunk modifications across

multiple replicas

Deletes old files (via garbage collection)

Read

The client application translates the file name and byte offset specified application into a chunk index within the file

The master replies with the corresponding chunk handle (information needed to find a chunk) and locations of the replicas.

Read The client then sends a request to one

of the replicas -- most likely the closest one

Note: Further reads of the same chunk require no more client-master interaction

Updates of Replicated Data Each mutation (modification) is

performed at all the replicas

Modifications are applied in the same order across all replicas If you have three chunks do you want one

chunk to have a b c and other to have a c b ?

For each chunk there is a primary chunk that coordinates updates

Updates of Replicated Data After finding replicas the client pushes

data to the replicas

The replicas do not write

The primary tells the replicas in which order they should apply modifications For each update the primary assigns a

sequence number Replicas update chunks based on the

sequence number

Updates of Replicated Data Let us say there are two updates to be

processed by a chunk from two clients: wa and wb

Their requests are received by the replicas

Different replicas may receive requests in different orders

Updates of Replicated Data Replicas should apply the writes in the

same order

The primary replica assigns a sequence number to each write wb may get sequence number 1 wa may get sequence number 2

The sequence numbers are sent to each replica

Updates of Replicated Data Let’s say that a replica got wa but not wb

It receives a sequence number of 2 for wa and the sequence number of 1 for wb

It delays executing wa until it receives wb

Updates of Replicated Data

1. Client asks master for replica locations

2. Master responds3. Client pushes data to all

replicas; replicas store it in a buffer cache

4. Client sends a write request to the primary (identifying the data that had been pushed)

5. Primary forwards request to the secondaries (identifies the order)

6. The secondaries respond to the primary

7. The primary responds to the client

Failure Handling During Updates

If a write fails at the primary: The primary may report failure to the client

– the client will retry If the primary does not respond, the client

retries from Step 1 by contacting the master If a write succeeds at the primary, but

fails at several replicas The client retries several times (Step 3-7)

Metadata On Master File names and chunk mappings are

also kept persistent in an operation log Chunk locations are kept in memory

only It will be lost during the crash The master asks chunk servers about their

chunks at startup – builds a table of chunk locations

Why Keep Metadata In Memory?

To keep master operations fast Master can periodically scan its internal

state in the background, in order to implement: Re-replication (in case of chunk server

failures) Chunk migration (for load balancing)

Why Not Keep Chunk Locations Persistent?

Chunk location – which chunk server has a replica of a given chunk

Master polls chunk servers for that information on startup

Thereafter, master keeps itself up-to-date: It controls all initial chunk placement,

migration and re-replication It monitors chunkserver status with regular

HeartBeat messages

Why Not Keep Chunk Locations Persistent?

Motivation: simplicity Eliminates the need to keep master and

chunkservers synchronized Synchronization would be needed when

chunkservers: Join and leave the cluster Change names Fail and restart

Operation Log Historical record of metadata changes Maintains logical order of concurrent

operations Log is used for recovery – the master

replays it in the event of failures Master periodically checkpoints the log

Chunk Size

Chunk size is 64 MB Larger than typical file system block sizes

Advantages Reduces a client’s need to interact with the

master Reduce network overhead by keeping a

persistent TCP connection to the chunkserver over a period of time

Reduces the size of the metadata stored on the master

Chunk Size

Disadvantages A small file consists of a small number of

chunks The chunkservers storing these chunks may

become hot spots if many clients are accessing the same file• Does not occur very much in practice

Deployment in Google

Many GFS clusters Hundreds/thousands of storage nodes

each Managing petabytes of data

Summary

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