Basic linux cluster knowledge

Cluster
Generally speaking, a cluster is an architecture that combines several electric brain cells in a certain way to complete each specific task.
Load balancing and LB can achieve load balancing to a certain extent to achieve high availability.
HA, High Availability, High Availability, real-time online, able to respond to client requests in a timely manner, enterprise applications require 7x24 hours and 99.999% online, so that organizations can provide real-time services online.
HP, High Performance High-Performance clusters provide clusters with a large number of super computing capabilities
--- Provides a large number of complex operations, and completes examples in a short time, such as ibm's dna graph cluster,
Scake on can be expanded up to replace poor technical machines with computers with better performance, or use high availability. High Cost
Scale out, for example, cluster
 
Server Load balancer
Easy scalability Load Balancing lvs for load forwarding
Disadvantage: a single point of failure (spof) causes all servers to go down.
Server Load balancer can also provide high availability and convenient scalability.
 
High-availability cluster
Function to ensure real-time online and high availability of services, but to achieve failover
Rsync synchronization drbd can also achieve synchronization of block-level networks that achieve shared storage san Region network output
 
High-performance Clusters
The forwarding node at the front of the forwarding node will dissect n small tasks for the complex tasks, and then each node will calculate the n small tasks. After the calculation, the forwarding node at the front of the node will be returned.
 
LB complex balanced Cluster
Lvs: (linux virtual server) a linux virtual server that receives user requests. It does not provide requests, realservers lvs is the one that actually improves the provision of nodes, and only needs to be installed on the scheduling node. This forwarding node is also called a virtual server. The forwarding mechanism of quad-route is transparent. lvs is only a node.
Working Mode: Three
Lvs provides advantages: for higher throughuput high throughput
For redundancy High Availability
For adaptability
Lvs id address name conventions: the IP address naming mechanism indicates different IP addresses.
1) virtual id (vip) address: the virtual IP address is forwarded to other nodes instead of providing services.
2) read ip (rip) address: the ip address on the node where the backend actually improves the service
3) director's ip (dip) address: distributor
4) Clinet computer's ip (cip) address: ip address obtained from the client

 
Lvs cluster type
1) The lvs-nat cluster node must be in the same CIDR block, which can be a sub-website or vlan and cannot span the ip CIDR block. The dip and rip must not span the ip CIDR Block in the same subnet, rip can usually be a private address, and all rip must be a drp gateway, this nat mechanism can be used for port ing and port translation. Any operating system can use lvs-nat, single lvs-nat, which may be used as a system bottleneck.

 
2) lvs-dr direct routing
1. cluster nodes
2. rip can use a public IP address.
3. director only processes user-sent requests. realservers does not respond. director only processes requests and does not respond.
4. The Gateway cannot point to dip
5. Port conversion is not supported.
6. Most Operating Systems Support realservers.
 
 
3) lvs-TUN 1,
Haproxy:

 
HA
* Heartbeat indicates that the heartbeat project is split.
Corosync + openais: RHCS
Ultramokey
* Keepalive
HP
Bowerful
 
Algorithm used for distribution
Fixed scheduling algorithm: a fixed algorithm does not consider the current connection status.
Round robin (RR): select a real server from the service list based on the current request arrival time, and redirect the request to this read server.
Destination hashing (DH) Target hash: redirects requests from the same server to the same read server.
Source hashing (SH) Source Address hash: forward requests from the same user through the same router and Firewall
Dynamic algorithm: checks the activity status of the current connection on the server to determine how to implement the next scheduling method.
Minimum lease connection (LC) connections: when the number of connections on the real server is small, the next connection request is directed to the real serever.
[Number of algorithm connections = number of active connections x 256 + number of inactive connections]
Weight least-Connection (WLC) weighted least connections:
[Algorithm: connections = (active connections x 256 + inactive connections)/weight]
Shorterst expected delay (SED) Shorterst expected latency: do not consider the number of inactive connections to add 1 [algorithm: connections = (number of active connections + 1) * 256/weight]
Nerver queue (NQ) Never queues algorithms. When a new request comes in, it depends not only on the value obtained by the SED algorithm, but also on whether there is an active link on the real server.
Locality-Based Least-connection (LBLC) is Based on the minimum local connection. Based on the DH algorithm, we also need to consider the number of active connections on the server. Weight supported
Locality-Bassed least-connection with requliction sscheduling (LBLCR) improved local least-connected LBLC algorithm with replication.
Author: "Birds of the earth"