PA200 – Cloud infrastructurePA200 – Cloud infrastructure
Storage and Data repositoriesStorage and Data repositories
Milan BrožMilan Brož
mbroz@redhat.commbroz@redhat.com
Storage cost
OpenStack storage types
Software-defined storage concepts
Data persistence and redundancy
Virtualization
Distributed storage
Security
Q & A
Agenda
Also see: Information Storage and Management, 2nd Edition
EMC Education Services, ISBN: 978-1-118-09483-9
Storage
●
Capacity
●
Availability, Reliability
●
Data integrity, Redundancy
●
Performance
●
Scalability
●
Security
=> Cost
Storage Cost
Manageability
Storage in OpenStack
as an example
Persistent Storage – Example
Ephemeral storage
●
Disappears when VM is terminated
●
Temporary data ~ computing clusters
●
Visible locally (to node)
Persistent storage
●
Data always available (no dependency on instance)
●
Can be shared among resources / instances
Data persistence
Object store
●
Binary objects of various length
●
REST API
Block (volume) storage
●
Block (sector-level) devices
●
Can be backed by a file image
Shared file-system storage
●
Mounted to a directory
Persistent Storage Types
Object store = SWIFT
Stateless swift-proxy
Block (volume) storage = CINDER
Backend Cinder drivers (LVM, GPFS, EMC, …)
Shared File-system = MANILA
Backend Manila drivers (Ceph, GlusterFS, NFS, …)
Image service = GLANCE
deduplication, clones, ...
Persistent Storage – OpenStack
https://docs.openstack.org/project-deploy-guide/openstack-ansible/newton/overview-storage-arch.html
https://docs.openstack.org/project-deploy-guide/openstack-ansible/newton/overview-storage-arch.html
Generic Storage Concepts
Software Defined Storage (SDS)
●
"Commodity hardware with abstracted storage logic"
●
Policy-based management of storage
●
Virtualization
●
Resource management
●
Similar concept as Software Defined Network (SDN)
Note: distributed storage is mostly about networking!
●
Thin provisioning, deduplication, replication, snapshots,
...
SDS definition differs among vendors!
Software Defined Storage
Hardware and low-level storage protocols
●
Physical storage
●
Rotational drives / hard disk drives (HDD)
●
Flash / SSD drives
●
Persistent Memory (byte-addressable!)
●
Tapes, magneto-optical drives, ...
●
Block-oriented storage access protocols
●
"Small Computer System Interface" (SCSI)
●
Serial Attached SCSI (SAS)
●
Serial ATA (SATA)
●
Fibre channel (FC) (not only fiber-optic)
●
InfiniBand (IB)
Storage layers
Storage connectivity through network
●
Direct-Attached Storage (DAS)
●
Local, host-attached
●
Network-Attached Storage (NAS)
●
Remote storage device
●
Communication protocol
●
Usually over IP-based network
●
High-level: NFS, CIFS, HTTP, ...
●
Low-level: iSCSI (SCSI over IP), FC (point-to-point),
Network Block Device (NBD)
NAS – Storage layers
Storage connectivity through network
●
Storage Area Network (SAN)
●
Private network
●
Switched fabric
●
Communication protocol
●
Fibre Channel
●
InfiniBand
●
FC over Ethernet (FcoE)
●
Multipath
SAN – Storage layers
High availability (HA)
●
Assuring access to resources
●
Service-level agreement (SLA)
●
Common 9s levels
High Availability
Resources access
●
On-demand
●
Active/Passive
●
Active/Active
~ Active/Passive – Mid-Range
~ Active/Active – High-End
Generic Storage Concepts
Data Protection and Redundancy
●
Data integrity protection
●
Random error detection (parity) / correction
●
Erasure codes
●
Forward Error Correction (FEC)
●
Redundancy
●
RAID (Redundant Array of Independent Disks)
●
Erasure coding in distributed storage
●
Backup and disaster recovery
●
"RAID is not a backup!"
●
File corruption, bugs (disk, controller, OS, application, …)
●
Admin error, malware
●
Catastrophic failure (datacentre fire)
●
Offline and off-site backup replica
Data protection
Common non-RAID and RAID disk configurations
●
JBOD – "Just a Bunch of Disks" (collection of disks, no redundacy)
●
RAID-0 – striping (for performance, no redundancy, no parity)
●
RAID-1 – mirroring (no parity)
●
RAID-5 – block-level striping + distributed parity (XOR)
●
RAID-6 – block-level striping + double distributed parity
●
RAID-10 – nested RAID example (1+0: striping over mirrored drives)
●
RAIDZ (in ZFS) – similar to RAID-5, dynamic stripes, self-healing
●
MAID (Massive Array of Idle Disks) – "Write once, read occasionally"
...
●
Degraded mode
●
RAID-5 (RAID-6 soon): large reconstruction time, fail during rebuild
●
Hardware RAID vs software RAID vs "fake RAID" (in fw/driver)
RAID – Data protection
●
Data protection is trade-off
●
Storage overhead
●
Reconstruction cost
●
Reliability
●
Still active research ...
Erasure coding – Data protection
Generic Storage Concepts
Virtualization
Storage pool
Set of disks, blocks, ... allocatable area for data
Pre-allocated
●
Partition table, logical volume in Logical Volume Manager
On-demand allocated
●
Thin provisioning (only blocks in use are allocated)
●
Flexible allocation
●
Used in snapshots
●
Possible over-allocation (sharing "unallocated" space)
Storage Pool
Volume Group
Logical Volume Manager (LVM)
Storage Pool Example
●
Snapshot of storage in specific time
●
Allows quick revert to older state (recovery)
●
Copy on (first) Write (COW/COFW) principle
●
Delayed copy to snapshot (before origin write)
●
Write to origin => need to copy the changed block first
Snapshots
Template
●
Application of deduplication + snapshots (+ thin provisioning)
●
Virtual machine template
●
Base operating system
●
Common configuration (networking, firewall, ...)
●
Common applications (webservers, user packages, ...)
●
One base image, only changes are stored
●
Application containers + template
●
Used in Docker
Templates
Deduplication
●
Avoid to store repeated data
●
File or block level
●
Space-efficient, stateless mode
●
Deduplication performance
●
Data corruption amplification
Deduplication / Compression
Compression
●
More generic algorithms
●
Special case: zeroed blocks
Tiered storage
●
Several layers of storage in one chain
●
Different performace, availability, recovery requirements
●
Cache (REST API)
Virtualization of drivers
●
virtio, pass-through device
Performance
Generic Storage Concepts
Distributed Storage
Clustered
● Cooperating nodes
Distributed
● Storage + network
Distributed storage transparency
●
Access (same as local)
●
Location (any node)
●
Failure (self-healing)
Distributed storage
Distributed storage examples
●
Ceph, GlusterFS (Red Hat)
●
General Parallel File System – GPFS (IBM)
●
Hadoop File-System HDFS (Apache)
●
Windows Distributed File-System (Microsoft)
●
GoogleFS / GFS (Google)
●
Isilon (EMC2
)
...
Distributed storage
CEPH – Distributed storage
Example of access of GlusterFS resources
GlusterFS – Distributed storage
Generic Storage Concepts
Security
●
Security policies
●
Confidentiality
●
Storage encryption (at-rest)
●
Data connection encryption (in-transit)
●
Key management
●
Authentication
●
Integrity (authenticated encryption)
●
Access control, permissions
●
Secure data disposal / destruction
●
Audit
...
Security
Cloud Storage Encryption
Encryption on client side
●
"End-to-End" encryption
●
Lost Efficiency for deduplication/compression
Encryption on server side
●
Partially lost confidentiality for clients
(server has access to decrypted data)
Data at-rest – combination of ...
●
Full disk encryption
●
Filesystem encryption
●
Object store encryption
Questions?