Big Data and Internet Thinking
Chentao Wu
Associate Professor
Dept. of Computer Science and Engineering [email protected]
Download lectures
• ftp://public.sjtu.edu.cn
• User: wuct
• Password: wuct123456
• http://www.cs.sjtu.edu.cn/~wuct/bdit/
Schedule
• lec1: Introduction on big data, cloud computing & IoT
• Iec2: Parallel processing framework (e.g., MapReduce)
• lec3: Advanced parallel processing techniques (e.g., YARN, Spark)
• lec4: Cloud & Fog/Edge Computing
• lec5: Data reliability & data consistency
• lec6: Distributed file system & objected-based storage
• lec7: Metadata management & NoSQL Database
• lec8: Big Data Analytics
Collaborators
Contents
Object-based Data Access
1
The Block Paradigm
The Object Paradigm
File Access via Inodes
•
Inodes contain file attributesObject Access
•
Metadata: Creation data/time; ownership; size …
•
Attributes – inferred: Access patterns; content; indexes …
•
Attributes – user supplied: Retention; QoS …
Object Autonomy
•
Storage becomes autonomous Capacity planning
Load balancing
Backup
QoS, SLAs
Understand data/object grouping
Aggressive prefetching
Thin provisioning
Search
Compression/Deduplication
Strong security, encryption
Compliance/retention
Availability/replication
Audit
Self healing
Data Sharing
homogeneous/heterogeneous
Data Migration
homogeneous/heterogeneous
Strong Security Additional layer
•
Strong security via external service Authentication
Authorization
…
•
Fine granularity Per object
Contents
Object-based Storage Devices
2
Data Access (Block-based vs. Object- based Device)
•
Objects contain both data and attributes Operations: create/delete/read/write objects, get/set attributes
OSD Standards (1)
•
ANSI INCITS T10 for OSD (the SCSI Specification, www.t10.org) ANSI INCITS 458
OSD-1 is basic functionality
Read, write, create objects and partitions
Security model, Capabilities, manage shared secrets and working keys
OSD-2 adds
Snapshots
Collections of objects
Extended exception handling and recovery
OSD-3 adds
Device to device communication
RAID-[1,5,6] implementation between/among devices
OSD Standards (2)
OSD Forms
• Disk array/server subsystem
Example: custom-built HPC systems predominantly deployed in national labs
• Storage bricks for objects
Example: commercial supercomputing offering
• Object Layer Integrated in Disk
Drive
OSDs: like disks, only different
OSDs: like a file server, only different
OSD Capabilities (1)
• Unlike disks, where access is granted on an all or nothing basis, OSDs grant or deny access to individual objects
based on Capabilities
• A Capability must accompany each request to read or write an object
Capabilities are cryptographically signed by the Security Manager and verified (and enforced) by the OSD
A Capability to access an object is created by the Security
Manager, and given to the client (application server) accessing the object
Capabilities can be revoked by changing an attribute on the object
OSD Capabilities (2)
OSD Security Model
• OSD and File Server know a secret key
Working keys are periodically generated from a master key
• File server authenticates clients and makes access control policy decisions
Access decision is captured in a capability that is signed with the secret key
Capability identifies object, expire time, allowed operations, etc.
• Client signs requests using the capability signature as a signing key
OSD verifies the signature before allowing access
OSD doesn’t know about the users, Access Control Lists (ACLs), or whatever policy mechanism the File Server is using
Contents
Object-based File Systems
3
Why not just OSD = file system?
•
Scaling What if there’s more data than the biggest OSD can hold?
What if too many clients access an OSD at the same time?
What if there’s a file bigger than the biggest OSD can hold?
•
Robustness What happens to data if an OSD fails?
What happens to data if a Metadata Server fails?
•
Performance What if thousands of objects are access concurrently?
What if big objects have to be transferred really fast?
General Principle
•
Architecture File = one or more groups of objects
Usually on different OSDs
Clients access Metadata Servers to locate data
Clients transfer data directly to/from OSDs
•
Address Capacity
Robustness
Performance
Capacity
•
Add OSDs Increase total system capacity
Support bigger files
Files can span OSDs if necessary or desirable
Robustness
•
Add metadata servers Resilient metadata services
Resilient security services
•
Add OSDs Failed OSD affects small percentage of system resources
Inter-OSD mirroring and RAID
Near-online file system checking
Advantage of Reliability
•
Declustered Reconstruction OSDs only rebuild actual data (not unused space)
Eliminates single-disk rebuild bottleneck
Faster reconstruction to provide high protection
Performance
•
Add metadata servers More concurrent metadata operations
Getattr, Readdir, Create, Open, …
•
Add OSDs More concurrent I/O operations
More bandwidth directly between clients and data
Additional Advantages
•
Optimal data placement Within OSD: proximity of related data
Load balancing across OSDs
•
System-wide storage pooling Across multiple file systems
•
Storage tiering Per-file control over
performance and resiliency
Per-file tiering in OSDs: striping
Per-file tiering in OSDs: RAID-4/5/6
Per-file tiering in OSDs: mirroring(RAID-1)
Flat namespace
Hierarchical File System Vs. Flat Address Space
• Hierarchical file system organizes data in the form of files and directories
• Object-based storage devices store the data in the form of objects
It uses flat address space that enables storage of large number of objects
An object contains user data, related metadata, and other attributes
Each object has a unique object ID, generated using specialized algorithm
Filenames/inodes
Hierarchical File System
Object IDs
Flat Address Space
Object Object
Object Object
Object Object
Data
Attributes Object ID
Metadata Object
Virtual View / Virtual File Systems
Traditional FS Vs. Object-based FS (1)
Traditional FS Vs. Object-based FS (2)
• File system layer in host manages
Human readable namespace
User authentication, permission checking, Access Control Lists (ACLs)
OS interface
• Object Layer in OSD manages
Block allocation and placement
OSD has better knowledge of disk geometry and characteristic so it can do a better job of file
placement/optimization than a host-based file system
Accessing Object-based FS
• Typical Access
SCSI (block), NFS/CIFS (file)
• Needs a client component
Proprietary
Standard
Standard→ NFS v4.1
• A standard file access protocol for OSDs
Scaling Object-based FS (1)
Scaling Object-based FS (2)
• App servers (clients) have direct access to storage to read/write file data securely
Contrast with SAN where security is lacking
Contrast with NAS where server is a bottleneck
• File system includes multiple OSDs
Grow the file system by adding an OSD
Increase bandwidth at the same time
Can include OSDs with different performance characteristics (SSD, SATA, SAS)
• Multiple File Systems share the same OSDs
Real storage pooling
Scaling Object-based FS (3)
• Allocation of blocks to Objects handled within OSDs
Partitioning improves scalability
Compartmentalized managements improves reliability through isolated failure domains
• The File Server piece is called the MDS
Meta-Data Server
Can be clustered for scalability
Why Objects helps Scaling
•
90% of File System cycles are in the read/write path Block allocation is expensive
Data transfer is expensive
OSD offloads both of these from the file server
Security model allows direct access from clients
•
High level interfaces allow optimization The more function behind an API, the less often you have to use the API to get your work done
•
Higher level interfaces provide more semantics User authentication and access control
Namespace and indexing
Object Decomposition
Object-based File Systems
•
Lustre Custom OSS/OST model
Single metadata server
•
PanFS ANSI T10 OSD model
Multiple metadata servers
•
Ceph Custom OSD model
CRUSH metadata distribution
•
pNFS Out-of-band metadata service for NFSv4.1
T10 Objects, Files, Blocks as data services
•
These systems scale 1000’s of disks (i.e., PB’s)
1000’s of clients
100’s GB/sec
All in one file system
Lustre (1)
•
Supercomputing focus emphasizing High I/O throughput
Scalability in the Pbytes of data and billions of files
•
OSDs called OSTs (Object Storage Targets)•
Only RAID-0 supported across Objects Redundancy inside OSTs
•
Runs over many transports IP over ethernet
Infiniband
•
OSD and MDS are Linux based & Client Software supports Linux Other platforms under consideration
•
Used in Telecom/Supercomputing Center/Aerospace/National LabLustre (2) Architecture
Lustre (3) Architecture-MDS
•
Metadata Server (MDS) Node(s) that manage namespace, file creation and layout, and locking.
Directory operations
File open/close
File status
File creation
Map of file object location
Relatively expensive serial atomic transactions to maintain consistency
•
•Metadata Target (MDT) Block device that stores metadata
Lustre (3) Architecture-OSS
• Object Storage Server (OSS)
Multiple nodes that manage network requests for file objects on disk.
• Object Storage Target (OST)
Block device that stores file objects
Lustre (4) Simplest Lustre File System
Lustre (5) File Operation
•
When a compute node needs to create or access a file, it requests the associated storage locations from the MDS and the associated MDT.•
I/O operations then occur directly with the OSSs and OSTs associated with the file bypassing the MDS.•
For read operations, file data flows from the OSTs to the compute node.Lustre (6) File I/Os
•
Single stream•
Single stream through a master•
ParallelLustre (7) File Striping
•
A file is split into segments and consecutive segments are stored on different physical storage devices (OSTs).Lustre (8) Aligned and Unaligned Stripes
•
Aligned stripes is where each segment fits fully onto a single OST.Processes accessing the file do so at corresponding stripe boundaries.
•
Unaligned stripes means some file segments are split across OSTs.Lustre (9) Striping Example
Lustre (10) Advantages/Disadvantages
•
Striping will not benefit ALL applicationsCeph (1)
• What is Ceph?
Ceph is a distributed file system that provides excellent performance, scalability and reliability.
Features
Decoupled data and metadata
Dynamic distributed metadata management
Reliable autonomic distributed object storage
Goals
Easy scalability to peta- byte capacity
Adaptive to varying workloads
Tolerant to node failures
Ceph (2) – Architecture
• Decoupled Data and Metadata
Ceph (3) – Architecture
Ceph (4) – Components
Object Storage
cluster Clients
Metadata Server cluster
Cluster monitor
Metadata I/O
Ceph (5) - Components
Meta Data cluster Clients
Object Storage cluster
Capability Management
CRUSH is used to map Placement Group (PG) to OSD.
Ceph (6) – Components
• Client Synchronization
POSIX Semantics
Relaxed Consistency
Synchronous I/O.
performance killer
Solution: HPC extensions to POSIX
Default: Consistency / correctness
Optionally relax
Extensions for both data and metadata
Ceph (7) – Namespace Operations
Ceph optimizes for most common meta-data
access scenarios
(readdir followed by stat)
But by default “correct”
behavior is provided at some cost.
Stat operation on a file opened by multiple writers
Applications for which coherent behavior is unnecessary use
extensions Namespace
Operations
Ceph (8) – Metadata
Per-MDS journals Eventually pushed to
OSD Sequential
Update More efficient
Reducing re- write workload.
Optimized on- disk storage layout for future
read access
Easier failure recovery. Journal
can be rescanned for
recovery.
• Metadata Storage
• Advantages
Ceph (9) – Metadata
• Dynamic Sub-tree Partitioning
▪ Adaptively distribute cached metadata hierarchically across a set of nodes.
▪ Migration preserves locality.
▪ MDS measures popularity of metadata.
Ceph (10) – Metadata
• Traffic Control for metadata access
• Challenge
• Partitioning can balance workload but can’t deal with hot spots or flash crowds
• Ceph Solution
✓ Heavily read directories are selectively replicated across multiple nodes to distribute load
✓ Directories that are extra large or experiencing heavy write workload have their contents hashed by file name across the cluster
Ceph (11) – Distributed Object Storage
Ceph (11) – CRUSH
• CRUSH(x) → (osd
n1, osd
n2, osd
n3)
• Inputs
• x is the placement group
• Hierarchical cluster map
• Placement rules
• Outputs a list of OSDs
• Advantages
• Anyone can calculate object location
• Cluster map infrequently updated
Ceph (12) – Replication
• Objects are replicated on OSDs within same PG
• Client is oblivious to replication
Ceph (13) – Conclusion
• Strengths:
• Easy scalability to peta-byte capacity
• High performance for varying work loads
• Strong reliability
• Weaknesses:
• MDS and OSD Implemented in user-space
• The primary replicas may become bottleneck to heavy write operation
• N-way replication lacks storage efficiency
• References
• Ceph: A Scalable, High Performance Distributed File System.
In Proc. of OSDI’06
Contents
Object-based Storage in Cloud
4
Web Object Features
•
RESTful API (i.e., web-based)•
Security/Authentication tied to Billing•
Metadata capabilities•
Highly available•
Loosely consistent•
Data Storage Blobs
Tables
Queues
•
Other related APIs (compute, search, etc.) Storage API is relatively simple in comparison
Simple HTTP example
HTTP and objects
•
Request specifies method and object: Operation: GET, POST, PUT, HEAD, COPY
Object ID (/index.html)
•
Parameters use MIME format borrowed from email Content-type: utf8;
Set-Cookie: tracking=1234567;
•
Add a data payload Optional
Separated from parameters with a blank line (like email)
•
Response has identical structure Status line, key-value parameters, optional data payload
This is a method call on an object
These are parameters
This is data
OpenStack REST API for Storage
•
GET v1/account HTTP/1.1 Login to your account
•
HEAD v1/account HTTP/1.1 List account metadata
•
PUT v1/account/container HTTP/1.1 Create container
•
PUT v1/account/container/object HTTP/1.1 Create object
•
GET v1/account/container/object HTTP/1.1 Read object
•
HEAD v1/account/container/object HTTP/1.1 Read object metadata
Create an object
Update metadata
Ali OSS (1)
•
Access URL: http://<bucket>.oss-cn-beijing.aliyuncs.com/<object>Access Layer Restful Protocol
LB LVS
Partition Layer Key-Value Engine
Persistent Layer Pangu FS
Load Balancing
Protocol Manager &
Access Control Partition & Index
Persistent, Redundancy
& Fault-Tolerance
Ali OSS (2) Architecture
•
WS: Web Server PM: Protocol ManagerPersistent Layer
M
M
Paxos M OS
OS
OS
OS
OS
Nuwa LockService
KVServer KVServer KVServer KVMaster
WS+PM WS+PM WS+PM WS+PM
Access Layer（RESTful API）
Partition Layer（LSM Tree）
Request ACK
Ali OSS (3) Partition Layer
•
Append/Dump/MergeMemFile
Block
Cache Block Index Cache Bloomfilter Cache Memory
Pangu
Youchao Files Redo Log File
Log Data Files
Ali OSS (4) Partition Layer
•
Read/Write ProcessMemFile
Redo Log File
Memory Read
Youchao Files
Dump memfile to youchao file
Write
Pangu Merge
Ali OSS (5) Persistent Layer
•
Write Pangu Normal FilePaxos Pangu client
M
M
M
OS OS OS OS
Create Chunk Chunk Location
Append Data ACK
Append Append
ACK ACK
Ali OSS (6) Persistent Layer
•
Write Pangu Log FilePaxos Pangu client
M
M
M
OS OS OS OS
Create Chunk Chunk Location
Flush Data ACK
The Evolution of Data Storage
Contents
Google File System (GFS)
5
Why build GFS?
• Node failures happen frequently
• Files are huge – multi-GB
• Most files are modified by appending at the end
Random writes (and overwrites) are practically non-existent
• High sustained bandwidth is more important than low latency
Place more priority on processing data in bulk
Typical workloads on GFS
•
Two kinds of reads: large streaming reads & small random reads Large streaming reads usually read 1MB or more
Oftentimes, applications read through contiguous regions in the file
Small random reads are usually only a few KBs at some arbitrary offset
•
Also many large, sequential writes that append data to files Similar operation sizes to reads
Once written, files are seldom modified again
Small writes at arbitrary offsets do not have to be efficient
•
Multiple clients (e.g. ~100) concurrently appending to a single file e.g. producer-consumer queues, many-way merging
Interface
•
Not POSIX-compliant, but supports typical file system operations:create, delete, open, close, read, and write
•
snapshot: creates a copy of a file or a directory tree at low cost•
record append: allow multiple clients to append data to the same file concurrently At least the very first append is guaranteed to be atomic
GFS Architecture (1)
GFS Architecture (2)
•
Very important: data flow is decoupled from control flow Clients interact with the master for metadata operations
Clients interact directly with chunkservers for all files operations
This means performance can be improved by scheduling expensive data flow based on the network topology
•
Neither the clients nor the chunkservers cache file data Working sets are usually too large to be cached, chunkservers can use Linux’s buffer cache
The Master Node (1)
•
Responsible for all system-wide activities managing chunk leases, reclaiming storage space, load-balancing
•
Maintains all file system metadata Namespaces, ACLs, mappings from files to chunks, and current locations of chunks
all kept in memory, namespaces and file-to-chunk mappings are also stored persistently in operation log
•
Periodically communicates with each chunkserver in HeartBeat messages This let’s master determines chunk locations and assesses state of the overall system
Important: The chunkserver has the final word over what chunks it does or does not have on its own disks – not the master
The Master Node (2)
•
For the namespace metadata, master does not use any per-directory data structures – no inodes! (No symlinks or hard links, either.)
Every file and directory is represented as a node in a lookup
table, mapping pathnames to metadata. Stored efficiently using prefix compression (< 64 bytes per namespace entry)
•
Each node in the namespace tree has a corresponding read-write lock to manage concurrency Because all metadata is stored in memory, the master can
efficiently scan the entire state of the system periodically in the background
Master’s memory capacity does not limit the size of the system
The Operation Log
•
Only persistent record of metadata•
Also serves as a logical timeline that defines the serialized order of concurrent operations•
Master recovers its state by replaying the operation log To minimize startup time, the master checkpoints the log periodically
The checkpoint is represented in a B-tree like form, can be directly mapped into memory, but stored on disk
Checkpoints are created without delaying incoming requests to master, can be created in ~1 minute for a cluster with a few million files
Why a Single Master? (1)
•
The master now has global knowledge of the whole system, which drastically simplifies the design•
But the master is (hopefully) never the bottleneck Clients never read and write file data through the master; client only requests from master which chunkservers to talk to
Master can also provide additional information about subsequent chunks to further reduce latency
Further reads of the same chunk don’t involve the master, either
Why a Single Master? (2)
•
Master state is also replicated for reliability on multiple machines, using the operation log and checkpoints If master fails, GFS can start a new master process at any of these replicas and modify DNS alias accordingly
“Shadow” masters also provide read-only access to the file system, even when primary master is down
They read a replica of the operation log and apply the same sequence of changes
Not mirrors of master – they lag primary master by fractions of a second
This means we can still read up-to-date file contents while master is in recovery!
Chunks and Chunkservers
•
Files are divided into fixed-size chunks, which has an immutable, globally unique 64-bit chunk handle By default, each chunk is replicated three times across multiple chunkservers (user can modify amount of replication)
•
Chunkservers store the chunks on local disks as Linux files Metadata per chunk is < 64 bytes (stored in master)
Current replica locations
Reference count (useful for copy-on-write)
Version number (for detecting stale replicas)
Chunk Size
•
64 MB, a key design parameter (Much larger than most file systems.)•
Disadvantages: Wasted space due to internal fragmentation
Small files consist of a few chunks, which then get lots of traffic from concurrent clients
This can be mitigated by increasing the replication factor
•
Advantages: Reduces clients’ need to interact with master (reads/writes on the same chunk only require one request)
Since client is likely to perform many operations on a given chunk, keeping a persistent TCP connection to the chunkserver reduces network overhead
Reduces the size of the metadata stored in master → metadata can be entirely kept in memory
Consistency Model
•
Terminology: consistent: all clients will always see the same data, regardless of which replicas they read from
defined: same as consistent and, furthermore, clients will see what the modification is in its entirety
•
Guarantees:Data Modification in GFS
•
After a sequence of modifications, if successful, then modified file region is guaranteed to be defined and contain data written by last modification•
GFS applies modification to a chunk in the same order on all its replicas•
A chunk is lost irreversibly if and only if all its replicas are lost before the master node can react, typically within minutes even in this case, data is lost, not corrupted
Record Appends
•
A modification operation that guarantees that data (the “record”) will be appended atomically at least once – but at the offset of GFS’s choosing The offset chosen by GFS is returned to the client so that the application is aware
•
GFS may insert padding or record duplicates in between different record append operations•
Preferred that applications use this instead of write Applications should also write self-validating records (e.g.
checksumming) with unique IDs to handle padding/duplicates
GFS Write Control and Data Flow (1)
•
If the master receives a modification operation for a particular chunk: Master finds the chunkservers that have the chunk and grants a chunk lease to one of them
This server is called the primary, the other servers are called secondaries
The primary determines the serialization order for all of the chunk’s modifications, and the secondaries follow that order
After the lease expires (~60 seconds), master may grant primary status to a different server for that chunk
The master can, at times, revoke a lease (e.g. to disable modifications when file is being renamed)
As long as chunk is being modified, the primary can request an extension indefinitely
If master loses contact with primary, that’s okay: just grant a new lease after the old one expires
GFS Write Control and Data Flow (2)
• 1. Client asks master for all chunkservers (including all secondaries)
• 2. Master grants a new lease on chunk, increases the chunk version number, tells all replicas to do the same. Replies to client. Client no longer has to talk to master
• 3. Client pushes data to all servers, not necessarily to primary first
• 4. Once data is acked, client sends write request to primary. Primary decides serialization order for all incoming modifications and applies them to the chunk
GFS Write Control and Data Flow (3)
• 5. After finishing the modification, primary forwards write request and serialization order to secondaries, so they can apply modifications in same order. (If primary fails, this step is never reached.)
• 6. All secondaries reply back to the primary once they finish the
modifications
• 7. Primary replies back to the client, either with success or error
If write succeeds at primary but fails at any of the secondaries, then we have inconsistent state
→ error returned to client
Client can retry steps (3) through (7)
Contents
Hadoop File System (HDFS)
6
Hadoop History
•
Dec 2004 – Google GFS paper published•
July 2005 – Nutch uses MapReduce•
Feb 2006 – Starts as a Lucene subproject•
Apr 2007 – Yahoo! on 1000-node cluster•
Jan 2008 – An Apache Top Level Project•
May 2009 – Hadoop sorts Petabyte in 17 hours•
Aug 2010 – World’s Largest Hadoop cluster at Facebook 2900 nodes, 30+ PetaByte
Hadoop Commodity Hardware
•
Typically in 2 level architecture Nodes are commodity PCs
20-40 nodes/rack
Uplink from rack is 4 gigabit
Rack-internal is 1 gigabit
Goals of Hadoop Distributed File System (HDFS)
•
Very Large Distributed File System 10K nodes, 1 billion files, 100 PB
•
Assumes Commodity Hardware Files are replicated to handle hardware failure
Detect failures and recovers from them
•
Optimized for Batch Processing Data locations exposed so that computations can move to where data resides
Provides very high aggregate bandwidth
•
User Space, runs on heterogeneous OSBasic of HDFS
•
Single Namespace for entire cluster•
Data Coherency Write-once-read-many access model
Client can only append to existing files
•
Files are broken up into blocks Typically 128 - 256 MB block size
Each block replicated on multiple DataNodes
•
Intelligent Client Client can find location of blocks
Client accesses data directly from DataNode
HDFS Architecture (1)
HDFS Architecture (2)
Namenode → Metadata
•
Meta-data in Memory The entire metadata is in main memory
No demand paging of meta-data
•
Types of Metadata List of files
List of Blocks for each file & file attributes
•
A Transaction Log Records file creations, file deletions, etc.
Datanode
•
A Block Server Stores data in the local file system (e.g. ext3)
Stores meta-data of a block (e.g. CRC32)
Serves data and meta-data to Clients
Periodic validation of checksums
•
Block Report Periodically sends a report of all existing blocks to the NameNode (heartbeats)
•
Facilitates Pipelining of Data Forwards data to other specified DataNodes
Block Placement
•
Current Strategy One replica on local node
Second replica on a remote rack
Third replica on same remote rack
Additional replicas are randomly placed
•
Clients read from nearest replica•
Pluggable policy for placing block replicas Co-locate datasets that are often used together
Block Replication
HDFS Read
•
To read a block, the client requests the list of replica locations from the NameNode•
Then pulling data from a replica on one of the DataNodesData Pipelining
•
Client writes block to the first DataNode The first DataNode forwards the data to the next
•
DataNode in the Pipeline, and so on When all replicas are written, the Client moves on to write the next block in file
•
Not good for latency sensitive applicationsHDFS Write
•
To write a block of a file, the client requests a list of candidate DataNodes from the NameNode, and organizes a write pipeline.Namenode failure
•
A Single Point of Failure•
Transaction Log stored in multiple directories A directory on the local file system
A directory on a remote file system (NFS/CIFS)
•
This is a problem with 24 x 7 operations AvatarNode comes to the rescue
NameNode High Availability Challenges
•
DataNodes send block location information to only oneNameNode
•
NameNode needs blocklocations in memory to serve clients
•
The in-memory metadata for 100 million files could be 60 GB, huge!NameNode High Availability AvatarNode
•
Active-Standby Pair Coordinated via zookeeper
Failover in few seconds
Wrapper over NameNode
•
Active AvatarNode Writes transaction log to filer
•
Standby AvatarNode Reads transactions from filer
Latest metadata in memory
Rebalancer
•
Goal: % disk full on DataNodes should be similar Usually run when new DataNodes are added
Cluster is online when Rebalancer is active
Rebalancer is throttled to avoid network congestion
•
Disadvantages Does not rebalance based on access patterns or load
No support for automatic handling of hotspots of data
HDFS RAID
•
Triplicate every data block•
Background encoding Combine third replica of blocks from a single file to create parity block
Remove third replica
•
RaidNode Auto fix of failed replicas
•
Reed Solomon encoding for old filesContents
Microsoft Azure
7
Microsoft Azure Storage
• Blobs – File system in the cloud
• Tables – Massively scalable structured storage
• Queues – Reliable storage and delivery of messages
• Drives – Durable NTFS volumes for Windows Azure applications
Windows Azure Storage Stamps
Storage Stamp LB
Storage Location
Service
Access blob storage via the URL: http://<account>.blob.core.windows.net/
Data access
Partition Layer Front-Ends
Stream Layer
Intra-stamp replication
Storage Stamp LB
Partition Layer Front-Ends
Stream Layer
Intra-stamp replication Inter-stamp (Geo) replication
Storage Stamp Architecture – Stream Layer
• Append-only distributed file system
• All data from the Partition Layer is stored into files (extents) in the Stream layer
• An extent is replicated 3 times across different fault and upgrade domains
• With random selection for where to place replicas for fast MTTR
• Checksum all stored data
• Verified on every client read
• Scrubbed every few days
• Re-replicate on disk/node/rack failure or checksum mismatch
M
Extent Nodes (EN) Paxos
M
Stream M
Layer
(Distributed File System)
Storage Stamp Architecture – Partition Layer
• Provide transaction semantics and strong consistency for Blobs, Tables and Queues
• Stores and reads the objects to/from extents in the Stream layer
• Provides inter-stamp (geo) replication by shipping logs to other stamps
• Scalable object index via partitioning
M
Extent Nodes (EN) Paxos
M M Partition
Server Partition
Server Partition
Server Partition Server Partition
Master Lock
Service
Partition Layer
Stream Layer
Storage Stamp Architecture – Front End Layer
• Stateless Servers
• Authentication + authorization
• Request routing
M
Extent Nodes (EN) Paxos
Front End
Layer FE
M M Partition
Server Partition
Server Partition
Server Partition Server Partition
Master
FE FE FE FE
Lock Service
Partition Layer
Stream Layer
Storage Stamp Architecture – Request
M
Extent Nodes (EN) Paxos
Front End
Layer FE
Incoming Write Request
M M Partition
Server Partition
Server Partition
Server Partition Server Partition
Master
FE FE FE FE
Lock Service
Ack
Partition Layer
Stream Layer
Partition Layer – Scalable Object Index
• 100s of Billions of blobs, entities, messages across all accounts can be stored in a single stamp
• Need to efficiently enumerate, query, get, and update them
• Traffic pattern can be highly dynamic
• Hot objects, peak load, traffic bursts, etc
• Need a scalable index for the objects that can
• Spread the index across 100s of servers
• Dynamically load balance
• Dynamically change what servers are serving each part of the index based on load
Scalable Object Index via Partitioning
• Partition Layer maintains an internal Object Index Table for each data abstraction
• Blob Index: contains all blob objects for all accounts in a stamp
• Table Entity Index: contains all entities for all accounts in a stamp
• Queue Message Index: contains all messages for all accounts in a stamp
• Scalability is provided for each Object Index
• Monitor load to each part of the index to determine hot spots
• Index is dynamically split into thousands of Index RangePartitions based on load
• Index RangePartitions are automatically load balanced across servers to quickly adapt to changes in load
Partition Layer – Index Range Partitioning
Account Name
Container Name
Blob Name
aaaa aaaa aaaaa
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
…….. …….. ……..
zzzz zzzz zzzzz
• Split index into
RangePartitions based on load
• Split at PartitionKey boundaries
• PartitionMap tracks Index RangePartition assignment to partition servers
• Front-End caches the
PartitionMap to route user requests
• Each part of the index is assigned to only one Partition Server at a time
Storage Stamp
Partition Server Partition
Server
Account Name
Container Name
Blob Name richard videos tennis
……… ……… ………
……… ……… ………
zzzz zzzz zzzzz
Account Name
Container Name
Blob Name harry pictures sunset
……… ……… ………
……… ……… ………
richard videos soccer
Partition Server
Partition Master
Front-End Server
PS 2 PS 3
PS 1
A-H: PS1 H’-R: PS2 R’-Z: PS3
A-H: PS1 H’-R: PS2 R’-Z: PS3
Partition Map Blob Index
Partition Map Account
Name
Container Name
Blob Name
aaaa aaaa aaaaa
……… ……… ………
……… ……… ………
harry pictures sunrise
A-H
H’-R R’-Z
Each RangePartition – Log Structured Merge Tree
Checkpoint File Table
Checkpoint File Table
Checkpoint File Table
Blob Data Blob Data Blob Data Commit Log Stream
Metadata log Stream
Writes Read/Query
Stream Layer
• Append-Only Distributed File System
• Streams are very large files
• Has file system like directory namespace
• Stream Operations
• Open, Close, Delete Streams
• Rename Streams
• Concatenate Streams together
• Append for writing
• Random reads
Stream Layer Concepts
Extent E2 Extent E3
Block Block Block Block Block Block Block Block
Block
• Min unit of write/read
• Checksum
• Up to N bytes (e.g. 4MB)
Extent
• Unit of replication
• Sequence of blocks
• Size limit (e.g. 1GB)
• Sealed/unsealed
Stream
• Hierarchical namespace
• Ordered list of pointers to extents
• Append/Concatenate
Block Block Block Block Block Block Block
Extent E4
Stream //foo/myfile.data Ptr E1 Ptr E2 Ptr E3 Ptr E4
Extent E1
Creating an Extent
SMStream SM Master
Paxos
Partition Layer
EN 1 EN 2 EN 3 EN
Create Stream/Extent
Allocate Extent replica set
Primary Secondary A Secondary B
EN1 Primary
EN2, EN3 Secondary
Replication Flow
SMSMSM
Paxos
Partition Layer
EN 1 EN 2 EN 3 EN
Append
Primary Secondary A Secondary B
Ack
EN1 Primary
EN2, EN3 Secondary
