A, LVS Introduction
LVS (Linux Virtual Server), the Linux Virtual Server, is led by Dr. Zhang Wen-song project open source load balancing, currently LVS has been integrated into the Linux kernel module. The project implements IP-based data requested load balancing scheduling scheme, which architecture shown in Figure 1, the user terminal from the external Internet access the company's external server load balancing, the end user requests sent to a Web LVS Linux kernel scheduler , a scheduler decides to send the request to a Web server in accordance with their rear end of a preset algorithm, for example, the polling algorithm may request the external server to the average of all of the rear end, while the end user access scheduler is LVS forwarded to the real back-end servers, but if the real server connection is the same memory, the same service is also provided services to end users regardless of which real server access, services get are the same, the entire cluster to the user For all transparent. Finally, depending on the transmission mode of the LVS, the real servers will choose a different way to the end user data required by the user, the operation mode is divided into LVS NAT mode, TUN mode and DR mode.
Second, resolve three operating modes.
1, based on LVS load balancing NAT mode
NAT (Network Address Translation) i.e., network address translation, which role is to modify the data packet header, private IP address is such that within the enterprise can access the Internet, as well as with external users can access the private IP hosts located within the company. VS / NAT operation mode topology shown in Figure 2, LVS load scheduler may use two different IP addresses NIC configuration, eth0 set private IP network are interconnected through the internal switching device, eth1 equipment and the external network IP external network connectivity.
The first step the user through the Internet DNS server to resolve balancing device outside the network load above the company address, as opposed to a real server, LVS external network IP, also known as VIP (Virtual IP Address), through user access VIP, you can connect the end real server (real server), and all this is transparent to the user, the user thought he visit is a real server, but he did not know their own VIP access is just a scheduler, it is unclear the real back-end where server, the number of real servers.
A second step, the user sends the request to 124.126.147.168, the rear end of this time LVS selected according to a preset algorithm for a real server (192.168.0.1 ~ 192.168.0.3), a data packet forwards the request to the real server, and before forwarding the LVS will modify the destination address of the packet and the destination port, destination address and destination port will be modified to selected real server IP address and the corresponding port.
After the third step, the real server response data packet back to the LVS scheduler, the scheduler is VIP and modify the corresponding port in the scheduler in response to the data packet to obtain the source address and source port will modification is completed, the scheduler the response packet back to the end user, Further, since there is a connection scheduler LVS Hash table, which will be recorded and forwarding the connection request message, when the next data packet with a connection to a dispatcher, from the Hash table can be connected directly to the previous record, and elected the same real server and port information according to the recording information.
2, the LVS load balancing based TUN
In a cluster environment LVS (the NAT) mode, since all data requests and data packets in response to the need to go through forward scheduler LVS, if the number is greater than the back-end server 10, the scheduler will become the bottleneck of the entire cluster environment . We know that the data request packet is always much smaller size of the response packet. Because the response packet contains data specific customer needs, the LVS (TUN) The idea is to separate request and response data, so that only the data request process scheduler, and let the real server response data packet back to the client directly. VS / TUN mode of topology as shown in Figure 3. Wherein, the IP tunnel (IP tunning) is a data packet encapsulation technique, it can be encapsulated original packet and adds a new header (including the new source address and port, destination address and port), so as to achieve a goal packet encapsulation VIP address scheduler, forwarded through the tunnel to the backend of the real server (real server), through the client sent to the scheduler of the original packet encapsulation, and add a new header on its basis (modified target IP address and the corresponding port address is selected from the dispatcher real server), LVS (TUN) mode requires real server can be directly connected to the external network, the real server to the client host in response to receipt of the request data directly to a data packet.
3, load balancing based on the DR LVS
In LVS (TUN) mode, due to the need for a tunnel between the dispatcher and LVS real server, which also will increase the burden on the server. And LVS (TUN) Similarly, the DR mode is also called direct routing mode, the architecture shown in Figure 4, the data mode remains liable LVS inbound requests, and only the selected algorithm according to reasonable real servers, and ultimately by the rear end real server is responsible for the response packet is sent back to the client. And tunnel mode is different, direct routing mode (DR mode) and back-end server requires the dispatcher must be in the same LAN, VIP address needs to be shared among all the servers and back-end scheduler, because in the end the real server to the client We need to set the source when the responding packet IP for the VIP address, destination IP client IP, so that clients can access the VIP address scheduler, in response to the source address also remains the VIP address (VIP on the real server), the client is imperceptible backend server exists. Since many computers are provided with the same a VIP address, so the requirements of the scheduler in direct routing mode, the VIP address is outside visible, the client needs to request packet to the scheduler host, VIP addresses of all of the real server must be configured on Non-ARP network devices, that is, the network device does not broadcast outside its own MAC and the corresponding IP address, VIP real server is not visible to the outside world, but the real server has accepted the target address VIP network requests, and in response to the data packet the source address as the VIP address. After the selected scheduler real server, without modifying the data packets, data frames will change the MAC address MAC address of the real server according to the selected algorithm, the data frame to the switch by the real server. Throughout the process, VIP real server does not need to be visible to the outside world.
Three, LVS load balancing scheduling algorithm
According to previous reports, we learned about the LVS of three operating modes, but regardless of the actual environment in which mode is used, strategy and scheduling algorithm scheduling algorithm is the core technology of LVS, LVS main achievement in the kernel a bit ten kinds of scheduling algorithms.
1. round robin scheduling
Round robin scheduling (Round Robin referred to as 'RR') algorithm is the manner in cycles of the request to schedule a different server, the algorithm is the biggest feature is simple. Polling algorithm assumes that all the capabilities of the server processes the request are the same, the scheduler will average all requests assigned to each real server.
2. Weighted round robin scheduling
Weighted round robin (Weight Round Robin referred to as 'WRR') is mainly an optimization algorithm and supplement polling algorithm, the LVS will consider the performance of each server, and to add a weight to each server, to the server if the A value 1, value 2 to the server B, then the scheduler schedules the request to the server B, server a would be twice. The higher the weight of the server, the more processing the request.
3. The minimum connection scheduling
The minimum connection scheduling (Least Connections referred to as 'LC') algorithm is to assign a new connection request to the server with the smallest number of currently connected. The minimum connection scheduling is a dynamic scheduling algorithm that estimates the number of the server by connecting to the server is currently active. Scheduler needs to record the number of connections to each server has been established, when a request is dispatched to a server, which is 1 plus the number of connections; when the connection is lost or a timeout, which is 1 minus the number of connections.
(Real server cluster systems have similar performance, with a minimum connection scheduling algorithm can better balance the load.)
4. weighted least connections scheduling
Weighted least connections (Weight Least Connections referred to as 'WLC') connected to the scheduling algorithm is the smallest superset corresponding weight represents the processing performance of each server. The default value of 1 to the server, the system administrator can dynamically set the weight of the server. Weighted least-connection scheduling in scheduling a new connection so that the server has been established as a value proportional to the number of connections and their weights. Queries scheduler may automatically load the real server, and dynamically adjusts the weight.
The local connection based minimum
Based on local minimum connection scheduling (Locality-Based Least Connections referred to as 'LBLC') algorithm for load balancing scheduling request packet destination IP address, the key for Cache cluster system, because in Cache Clusters customers request packet destination IP address is changed. It is assumed that any backend server can handle a request for any, the algorithm is designed in a substantially balanced load the server will request the same destination IP address are dispatched to the same server, each server to improve locality and Cache hit rate, so as to enhance the processing capability of the whole cluster system. LBLC scheduling algorithm first find the IP address of the target server according to the most recently used destination IP address request, if the server is available and is not overloaded, send a request to the server; if the server does not exist, or if the server has a server is overloaded and workload in half, using the principle of "least connections' to select an available server, sends a request to the server.
6. band replication connection based minimum Locality
With a copy of the Locality-Based Least-Connection (Locality-Based Least Connections with Replication referred to as 'LBLCR') algorithm is load balancing target IP address, the key for Cache cluster system, it LBLC algorithm is that it is different to maintaining the mapping from a target IP address to a group of servers, and LBLC algorithm maintains a mapping from a target IP address to a server. Press 'minimum connection' principle selected from eleven servers in the server group, if the server is not overloaded, send a request to the server; if the server is overloaded, press 'minimum connection' a principle selected from the entire server cluster the server is then added to the server group, sends a request to the server. Meanwhile, when the server group for some time not been modified, the busiest server is removed from the server group, in order to reduce the degree of replication.
7. destination address hashing scheduling
Scheduling target address hash (Destination Hashing abbreviation 'DH') algorithm according to the request destination IP address as the hash key (Hash Key) to find the corresponding server from the list of hash static allocation, if the server is available and and not overloaded, the request is sent to the server, otherwise empty.
8. The source address hashing scheduling U
Scheduling the source address hash (Source Hashing abbreviation 'SH') algorithm according to the source IP address of the request, as the hash key (Hash Key) to find the corresponding server from the list of hash static allocation, if the server is available and and not overloaded, the request is sent to the server, otherwise empty. Hash function with the hash of the destination address it uses the same scheduling algorithm, which is substantially similar to the target address hash algorithm process scheduling algorithm.
9. shortest desired delay
The shortest expected delay scheduling (Shortest Expected Delay referred to as 'SED') WLC algorithm Algorithm. To give an example, the right to ABC three servers weight 1,2,3, respectively. So if you use WLC algorithm, then a new request comes in it may give any one of ABC. We will perform a computation algorithm after using SED
A: (1 + 1) / 1 = 2 B: (1 + 2) / 2 = 3/2 C: (1 + 3) / 3 = 4/3 puts the request to the server which yields a minimum calculation result.
10. The queue scheduling Minimum
Minimum queue scheduling (Never Queue referred to as 'NQ') algorithm, without queue. If the number of connections is equal to 0 realserver directly assigned past, no operation is performed SED.
