Almost a month did not update the blog, most recently in the busy WorldSkills network-based system to manage the project, do not have much time, I am one hundred updates about being busy. There comes a recent tournament title this cluster, lvs this thing (I should also learn this, and remained in the basics, it's time to learn some profound things), emm! Ado, let's begin to explain it.
Let's look at what is a cluster:
A cluster is a group of independent, high-speed networks of interconnected computers, which constitute a group, and managed in a single mode system. When a customer interacts with the cluster, the cluster like a stand-alone server. Cluster configuration is used to improve the availability and scalability.
Role: to consolidate server resources, the consolidation of multiple servers into a single use, can improve the compressive strength of the whole structure
1. Cluster categories:
The system consists of a cluster
Load balancing (Load Balance) clusters, referred to as LB cluster
HA (High Availability) cluster, referred to as HA cluster
HPC (High Perfermance Computing) clusters, referred to as HPC clusters. Mainly used for some scientific computing.
2.LB means of achieving
Hardware implementations: - F5 load balancer
Software implementation: - LVS (4 layer, transport layer) - Nginx (7 layer, application layer)
emm, F5 Why is it not lvs it, it may be the boss, after all, money is free and open source software, hardware F5 boss can not afford too expensive (I say too real)
3.LVS (4 layers) mode
LVS/NAT、LVS/DR、LVS/TUN
4.LVS load balancing packet forwarding three ways
LVS provides three packet forwarding mode: NAT (network address mapping), IP Tunneling (IP tunnel), Direct Routing (direct route). Different forwarding mode determines the network structure of different cluster
The scheduling pattern look lvs
| 1 |
rr Poll: evenly distributed, you a, I a. Hardware devices suitable for the same batch of the same configuration approach |
| 2 |
wrr weight based polling: Who processing ability, whose weight is high. With WRR better (ADM) server hardware configuration gap |
| 3 |
Minimum connection lc: Two small number of connections RS, who is to give a new request processing, if the same two randomly select a number of connections |
| 4 |
wlc weight-based Least Connections: As far as the weight is taken into account the performance of your server hardware, high performance two servers weight value is set higher, the low performance set lower. If the algorithm according to the number of servers connected to at least two connections are the same, then the new request will give a good performance processing server |
| 5 |
lblc least partially based on the connection: in general use on the cache server cdn network acceleration when used, can be improved by caching and |
| 6 |
sh source address hash: performs operations in the source access. When the same source to ensure that in the past the same server response time |
| 7 |
dh destination address hash: calculates the destination address accessed. The purpose calculates ip, dns resolution depends |
| 8 |
sed shortest expected delay law |
| 9 |
nq never queuing |
(I can not read too much, in fact, important to both a front, other scheduling pattern look on it)