mirror of
https://github.com/ceph/ceph
synced 2025-01-01 16:42:29 +00:00
doc: Make overview.rst valid reStructuredText, so I can stop seeing warnings.
It's still wrong, but now it won't clutter the output. Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>
This commit is contained in:
parent
c119074055
commit
1430c8abaa
145
doc/overview.rst
145
doc/overview.rst
@ -1,61 +1,120 @@
|
||||
======================
|
||||
Ceph Product Overview
|
||||
======================
|
||||
=====================
|
||||
Ceph Product Overview
|
||||
=====================
|
||||
|
||||
======================
|
||||
About this Document
|
||||
======================
|
||||
This document describes the features and benefits of using the Ceph Unified Distributed Storage System, and why it is superior to other systems.
|
||||
The audience for this document consists of sales and marketing personnel, new customers, and all persons who need to get a basic overview of the features and functionality of the system.
|
||||
======================
|
||||
Introduction to Ceph
|
||||
======================
|
||||
Ceph is a unified, distributed file system that operates on a large number of hosts connected by a network. Ceph has been designed to accommodate multiple petabytes of storage with ease. Since file sizes and network systems are always increasing, Ceph is perfectly positioned to accommodate these new technologies with its unique, self-healing and self-replicating architecture. Customers that need to move large amounts of metadata, such as media and entertainment companies, can greatly benefit from this product. Ceph is also dynamic; no need to cache data like those old-fashioned client-servers!
|
||||
Benefits of Using Ceph
|
||||
Ceph’s flexible and scalable architecture translates into cost savings for users. Its powerful load balancing technology ensures the highest performance in terms of both speed and
|
||||
reliability. Nodes can be added “on the fly” with no impact to the system. In the case of node failure, the load is re-distributed with no degradation to the system.
|
||||
Failure detection is rapid and immediately remedied by efficiently re-adding nodes that were temporarily cut off from the network.
|
||||
Manageability
|
||||
Ceph is easy to manage, requiring little or no system administrative intervention. Its powerful placement algorithm and intelligent nodes manage data seamlessly across any node
|
||||
configuration. It also features multiple access methods to its object storage, block storage, and file systems. Figure 1 displays this configuration.
|
||||
.. image:: /images/CEPHConfig.jpg
|
||||
======================
|
||||
RADOS
|
||||
======================
|
||||
The Reliable Autonomic Distributed Object Store (RADOS) provides a scalable object storage management platform. RADOS allows the Object Storage Devices (OSD) to operate autonomously
|
||||
when recovering from failures or migrating data to expand clusters. RADOS employs existing node device intelligence to maximized scalability.
|
||||
The RADOS Block Device (RBD) provides a block device interface to a Linux machine, while striping the data across multiple RADOS objects for improved performance.
|
||||
RDB is supported for Linux kernels 2.6.37 and higher. Each RDB device contains a directory with files and information
|
||||
===================
|
||||
|
||||
This document describes the features and benefits of using the Ceph
|
||||
Unified Distributed Storage System, and why it is superior to other
|
||||
systems.
|
||||
|
||||
The audience for this document consists of sales and marketing
|
||||
personnel, new customers, and all persons who need to get a basic
|
||||
overview of the features and functionality of the system.
|
||||
|
||||
Introduction to Ceph
|
||||
====================
|
||||
|
||||
Ceph is a unified, distributed file system that operates on a large
|
||||
number of hosts connected by a network. Ceph has been designed to
|
||||
accommodate multiple petabytes of storage with ease. Since file sizes
|
||||
and network systems are always increasing, Ceph is perfectly
|
||||
positioned to accommodate these new technologies with its unique,
|
||||
self-healing and self-replicating architecture. Customers that need
|
||||
to move large amounts of metadata, such as media and entertainment
|
||||
companies, can greatly benefit from this product. Ceph is also
|
||||
dynamic; no need to cache data like those old-fashioned
|
||||
client-servers!
|
||||
|
||||
Benefits of Using Ceph
|
||||
======================
|
||||
|
||||
Ceph's flexible and scalable architecture translates into cost savings
|
||||
for users. Its powerful load balancing technology ensures the highest
|
||||
performance in terms of both speed and reliability. Nodes can be
|
||||
added "on the fly" with no impact to the system. In the case of node
|
||||
failure, the load is re-distributed with no degradation to the system.
|
||||
|
||||
Failure detection is rapid and immediately remedied by efficiently
|
||||
re-adding nodes that were temporarily cut off from the network.
|
||||
|
||||
Manageability
|
||||
=============
|
||||
|
||||
Ceph is easy to manage, requiring little or no system administrative
|
||||
intervention. Its powerful placement algorithm and intelligent nodes
|
||||
manage data seamlessly across any node configuration. It also
|
||||
features multiple access methods to its object storage, block storage,
|
||||
and file systems. Figure 1 displays this configuration.
|
||||
|
||||
.. image:: /images/CEPHConfig.jpg
|
||||
|
||||
RADOS
|
||||
=====
|
||||
|
||||
The Reliable Autonomic Distributed Object Store (RADOS) provides a
|
||||
scalable object storage management platform. RADOS allows the Object
|
||||
Storage Devices (OSD) to operate autonomously when recovering from
|
||||
failures or migrating data to expand clusters. RADOS employs existing
|
||||
node device intelligence to maximized scalability.
|
||||
|
||||
The RADOS Block Device (RBD) provides a block device interface to a
|
||||
Linux machine, while striping the data across multiple RADOS objects
|
||||
for improved performance. RDB is supported for Linux kernels 2.6.37
|
||||
and higher. Each RDB device contains a directory with files and
|
||||
information
|
||||
|
||||
RADOS GATEWAY
|
||||
======================
|
||||
=============
|
||||
|
||||
``radosgw`` is an S3-compatible RESTful HTTP service for object
|
||||
storage, using RADOS storage.
|
||||
|
||||
The RADOS Block Device (RBD) provides a block device interface to a Linux machine. To the user, RDB is transparent, which means that the entire Ceph system looks like a single,
|
||||
limitless hard drive that is always up and has no size limitations. .
|
||||
The RADOS Block Device (RBD) provides a block device interface to a
|
||||
Linux machine. To the user, RDB is transparent, which means that the
|
||||
entire Ceph system looks like a single, limitless hard drive that is
|
||||
always up and has no size limitations. .
|
||||
|
||||
|
||||
======================
|
||||
Hypervisor Support
|
||||
======================
|
||||
RBD supports the QEMU processor emulator and the Kernel-based Virtual Machine (KVM) virtualization infrastructure for the Linux kernel. Normally, these hypervisors would not be used
|
||||
together in a single configuration.
|
||||
==================
|
||||
|
||||
RBD supports the QEMU processor emulator and the Kernel-based Virtual
|
||||
Machine (KVM) virtualization infrastructure for the Linux kernel.
|
||||
Normally, these hypervisors would not be used together in a single
|
||||
configuration.
|
||||
|
||||
KVM RBD
|
||||
The Linux Kernel-based Virtual Machine (KVM) RBD provides the functionality for striping data across multiple distributed RADOS objects for improved performance.
|
||||
KVM-RDB is supported for Linux kernels 2.6.37 and higher. Each RDB device contains a directory with files and information.
|
||||
KVM employs the XEN hypervisor to manage its virtual machines.
|
||||
-------
|
||||
|
||||
The Linux Kernel-based Virtual Machine (KVM) RBD provides the
|
||||
functionality for striping data across multiple distributed RADOS
|
||||
objects for improved performance.
|
||||
|
||||
KVM-RDB is supported for Linux kernels 2.6.37 and higher. Each RDB
|
||||
device contains a directory with files and information.
|
||||
|
||||
KVM employs the XEN hypervisor to manage its virtual machines.
|
||||
|
||||
QEMU RBD
|
||||
QEMU-RBD facilitates striping a VM block device over objects stored in the Ceph distributed object store. This provides shared block storage to facilitate VM migration between hosts.
|
||||
QEMU has its own hypervisor which interfaces with the librdb user-space library to store its virtual machines
|
||||
--------
|
||||
|
||||
QEMU-RBD facilitates striping a VM block device over objects stored in
|
||||
the Ceph distributed object store. This provides shared block storage
|
||||
to facilitate VM migration between hosts.
|
||||
|
||||
QEMU has its own hypervisor which interfaces with the librdb
|
||||
user-space library to store its virtual machines
|
||||
|
||||
.. _monitor:
|
||||
|
||||
Monitors
|
||||
===============
|
||||
========
|
||||
|
||||
Once you have determined your configuration needs, make sure you have access to the following documents:
|
||||
• Ceph Installation and Configuration Guide
|
||||
• Ceph System Administration Guide
|
||||
• Ceph Troubleshooting Manual
|
||||
Once you have determined your configuration needs, make sure you have
|
||||
access to the following documents:
|
||||
|
||||
- Ceph Installation and Configuration Guide
|
||||
- Ceph System Administration Guide
|
||||
- Ceph Troubleshooting Manual
|
||||
|
Loading…
Reference in New Issue
Block a user