RHCA Translation Plan-EX436 Chapter 1: Introduction to Cluster Storage 2

1.11 SAN storage technology:
Network Connections from storage to hosts have a variety of mechanisms, including DAS, SAN, and NAS. For large applications, key applications often adopt the SAN Storage Architecture, which has the advantages of high performance, it is easy to expand, but the disadvantage is that the management is complicated and expensive (especially the FC optical network ).

For example, key services such as web servers and database servers are connected to the optical fiber switch through optical fiber cables. The optical fiber switch is a storage network formed by multiple storage sets. This architecture has powerful performance. In enterprise applications, to prevent single point of failure, two sets of Optical Fiber switches are used for physical redundancy to ensure high availability.
ISCSI is required when it comes to SAN. iSCSI (ip Over SCSI) is also known as ipSAN in the industry. It is used to transmit disk SCSI data through TCP/IP network protocol, with the development of Ethernet technology, Gigabit 10-Gigabit Ethernet will be applied to enterprises. However, many small and medium enterprises prefer to use iSCSI storage for cost effectiveness.
It is very important to distinguish between SCSI devices and FC Optical Fiber devices. FC Optical Fiber devices simulate the abstract devices of one or more SCSI devices at the bottom layer of storage virtualization, and do not have real SCSI devices, it is simulated through the SCSI protocol.
GNBD (Global Network Block Device) Global Network Block devices are similar to iSCSI storage architecture, but data transmission is measured in data blocks, the C/S kernel module provides block-level storage access.
 
1.12 Optical Fiber Channel:
In enterprise-level applications, the best choice for communication is FC optical fiber. fiber-optic communication has good anti-interference and high-speed performance in enterprises. The establishment of fiber-optic channels requires optical fiber cables, HbA cards, and optical fiber switches, A high-performance and high-availability optical fiber network.
Because of the complexity and high price of optical fiber network management, it is generally used only in the key part of the most critical business. A typical application is optical fiber storage network.
 
1.13 HBA Optical Fiber Interface Card:
Speaking of optical fiber communication, optical fiber communication is mainly a process of photoelectric conversion. digital signals are converted into optical signals through optical fiber transfer cards, and optical signals are transmitted at high speed in optical fiber, switch to an electrical signal on the hidden card at the other end. The memory card itself has a processor to process data and does not require CPU processing time. This is also one of the reasons for efficient fiber-optic communication.
A memory card is similar to a scsicard and is recognized as a SCSI device in a computer. The memory card uses the SCSI command to convert the SCSI data to the FC Fiber Channel Protocol for transmission. Some HBA cards provide more than one physical fiber path, just like multi-path.
 
1.14 optical fiber switch:
As an important part of the SAN Optical Fiber storage network, optical fiber switches provide excellent scalability, which makes SAN storage network connections almost independent from distance. The Optical Fiber Channel communication distance can exceed 100 km.
There are three types of Optical Fiber switching topologies: point-to-point, arbitration loop, and switched optical fiber network. Point-to-point is a simple connection between two devices. The arbitration loop means that all devices are arranged in one loop connection. The switched optical fiber network means that all devices are connected to one or more optical fiber switches, forming an optical fiber switching network. Switched Optical fiber networks are widely used in enterprises.
 
1.15 iSCSI storage:
ISCSI can be understood as ip Over SCSI, or internet SCSI, that is, SCSI data is transmitted Over the Ethernet, using the TCP/IP protocol, port 3260 is used by default. Some enterprises prefer to use iSCSI, because unlike FC optical fiber networks, iSCSI can be deployed using the existing exchange network, and iSCSI is known for its cost effectiveness, which is the best choice for small and medium enterprises.
 
1.16 NPS Network Power Controller:
NPS (Network Power Switch) Network Power Controller, that is, Power Fence. In a cluster, the Fence device is very important. In a Windows Cluster, there is no Fence concept, and its service switching is achieved through the arbitration mechanism. In a Linux cluster, there is an arbitration mechanism, there is also a Fence mechanism.
Fence devices can be divided into two categories: Power Fence and cable Fence. Most of the current clusters only use the cable Fence. The following describes the working principle of Fence: taking a dual-host HA cluster as an example, when the idle node is notified of a problem with the service node through the jumper, such as a communication failure, the idle node will use the Fence card on the service node to shut down or restart the node, and "grab" the service resources to the node. Power Fence is to switch the service to another node when the cluster node is detected to be powered off.
That is, NPS provides an Ethernet port for each power interface to monitor the power supply in real time. Currently, it is rarely used in China. In fact, it has a UPS and a sound data center management system, there is no need for NPS.

 
1.17 ACPI mechanism:
The Advanced Configuration and Power interface (ACPI) mechanism is mainly used for host Power management and is very useful for managing Power consumption. Various manufacturers also have corresponding products, such as Dell's remote access interface card DRAC, HP's iLo, and Intel's IpmI.
Because ACPI is a software-based mechanism, it cannot be fully dependent. For example, if a node in a cluster is evicted due to kernel panic, the node cannot process the necessary power polling.
 
1.18 network cluster nodes:
The cluster heartbeat is also part of the cluster communication transmission. Services on the cluster can be separated by subnets on different network adapters. inter-cluster communication forms a subnet, the NIC of each node in the cluster that provides external services is connected to the public network.
Generally, to prevent single point of failure (spof), more than two NICs must be used to bind each communication interface to achieve redundancy and load balancing.
Network devices must support multicast. multicast is used for communications between nodes in a cluster. addresses are automatically generated when the cluster is created, but can be overwritten by other addresses manually. All network hardware (such as routing, switching, hub, and nic) must support multicast. Some of the default configurations of the vswitch disable multicast. You can ping the multicast address and use netstat-g to view the allocated multicast address.
 
1.19 broadcast and Multicast:
Broadcast is characterized by Single-to-many. Like the DHCP service, broadcast is used to send DHCP identities. multicast is a packet sent to a specified receiver. In an image metaphor, broadcast tells people in the company one thing, and multicast tells only some people you want to tell.
Multicast is more efficient than broadcast, because broadcast is sent once, regardless of whether you need it or not, and multicast is targeted, which is more efficient.
Not all Ethernet cards support multicast.
IPv6 relies heavily on multicast because IPv6 removes broadcast and uses more reasonable multicast.
 
1.20 Ethernet channel binding:
The Ethernet channel binding is to bind multiple physical NICs to a logical device on the operating system to achieve high availability and load balancing. In a cluster, the heartbeat line between two machines is the key to the cluster. However, if the NIC of the heartbeat line fails, the entire cluster will lose its functionality, the heartbeat Nic is often bound with two NICs to achieve redundancy and prevent cluster crash due to physical Nic failure.
 
1.21 configurations bound to the Ethernet channel:
Configuring Ethernet Card binding is quite simple.
The first step is to configure and load the bonding kernel module to support the binding function;
# Vim/etc/modprobe. conf
Alias bond0 bonding
Options bond0 mode = 1 miimon = 100 use_carrier = 0
Write the parameters of the preceding two rows and save and exit. mode = 0/1, 0 indicates fault tolerance and load balancing, and 1 indicates Fault Tolerance only; miimon = 100 indicates the connection monitoring frequency, in milliseconds; use_carrier indicates how to determine the connection status; you can use the primary parameter to specify which Nic is the primary priority.
Step 2: configure the NIC configuration file:
# Cd/etc/sysconfig/network-scripts/
If you want to bind eth0 and eth1 to bond0, make the following modifications:
# Vim ifcfg-eth0.
DEVICE = eth0
MASTER = bond0
SLAVE = yes
ONBOOT = yes
BOOTPROTO = static
# Vim ifcfg-eth1.
DEVICE = eth1
MASTER = bond0
SLAVE = yes
ONBOOT = yes
BOOTPROTO = static
# Vim ifcfg-bond0.
DEVICE = bond0
IPADDR = 192.0.2.1
NETMASK = 255.255.255.0
GATEWAY = 192.0.2.254
ONBOOT = yes
BOOTPROTO = static
Restart the network:
#/Etc/init. d/network restart
 
1.22 multi-path:
Multi-path is used to access shared storage to prevent service unavailability caused by spof.
Device ing multi-path allows nodes to access storage through multiple paths. Multi-path supports storage architectures such as FC, iSCSI, and GNBD.
 
1.23 security:
Multi-node communication between clusters is encrypted by default. OpenAIS uses the cluster name as the key, but this is not a thorough security policy.
The firewall must allow the use of the corresponding ports of the cluster and GFS. The following ports must be open:
5404,505 cman udp
11111 ricci tcp
14567 gnbd tcp
16851 modcluster tcp
21064 dlm tcp
50006,50008 ccsd tcp
50007 ccsd udp
 
Author: ye shaochen"