Anxious to read the [technology to support the server/infrastructure] in 5 minutes

Recently it has been increasingly able to see the system configuration diagram in the business.

Because it is often do not know may be seen, I will try to learn something,

read the [technology to support the server/infrastructure].

Because much trouble, tried together to the extent that the bearded hair in the table of contents.

[Chapter 1] server/infrastructure construction Introduction

  • The system must be designed so that failure is good even when going on
  • For this purpose, router, load balancer, server, needs to be redundant all
  • The important thing when you have redundancy,

    • Detection of failure (health check)
    • Smooth switching (takeover of the IP address)
  • Scheduling algorithm, VRRP, and various mechanisms are used

Construction of [Chapter 2] up a notch server/infrastructure

  • Using cache server as a reverse proxy can load distribution to the AP server

    • The content of the request, it is possible to switch the server (L7 switch role)
    • Return HTML, CSS, JS, static content such as images from the cache server
    • [HTTP protocol level: Squid], [Web application data: memcached]
  • MySQL also redundant (replication), and effective use of slave

    • Request only by referring to data may be reference to the slave
    • Of course, keep to be in order in the slave also the internal load balancer (Linux)
  • By using HTTP as the storage protocol, to avoid the risk of site is totally stopped

    • Necessary redundancy to prevent single point of failure
    • Failure-prone and leave synchronize files on multiple servers
    • These two problems in Chapter 3

[Chapter 3] further devised that aims to infrastructure that does not stop

  • Although failure of the DNS server is not easily occur, is wide much more difficult impact of range Cause identified when that happened

    • To have a VIP to the load balancer, redundant in the Active/Active configuration. (IPVS, keepalive)
  • Storage server synchronization is difficult -> solved using DRBD

    • DRBD: rather than a file-by-file basis, to replicate in real-time block device unit
    • To control the keepalivd in daemontools
    • Weakness that someone will be reflected in the backup as soon as you delete a file by mistake
  • With Bonding driver, L1,/L2 also redundant (links, switches)

    • If redundant too loop she can (broadcast storm)
    • Resolved RSTP (blocking the automatic redundant connection)
  • Flexibility to the network using a VLAN

    • Port VLAN: assign a VLAN identifier for each port on the switch
    • Tag VLAN: to insert the VLAN identification information in the Ethernet frame. Tend to be complicated logical configuration.
    • Use of a server farm, the use of the tag VLAN, Nakuseru the physical waste
    • The port VLAN as a basic, to perform the set of tag VLAN where needed ideal
    • Also VLAN configuration becomes complicated, physical configuration determine the simplicity

[Chapter 4] performance improvement, tuning

  • In accordance with the redundancy of the past, there is a sense of first load distribution by drawing performance of a single host

    • Load is not to guess, to measure: see the load average
    • There are two types of CPU load and I/O load, find out which one is the bottleneck
    • ps, sar, vmstat, with the top ,,, various commands to make sure
    • By looking at the kernel code, deepen the understanding of the load calculation method
    • Tuning and discovers a bottle neck, only to remove it
  • But can also be tuning in Web server (Apache), most of the cause exists in a different location

    • If by force increase, what to choose to MPM at the time of the parallel processing is important
    • Determining the MaxClients by the average memory consumption of physical memory and a process server installed
    • perfork: multi-process. Everyday use if good here.
    • worker: multi-threaded. It is a fast, lightweight, complex settings.
  • Effective MySQL tuning table design and optimization of SQL is

    • It can be on the server side, the parameter tuning of memory relationship
    • Buffer: innodb_buffer_pool_size Toka Toka

[Chapter 5] labor-saving operation

  • Used a variety of tools for labor-saving operation
  • Nagios: service of the operation monitoring

    • Life-and-death states
    • Load state
    • Measurement of capacity utilization
  • Ganglis: Monitoring of server resources

    • CPU usage, memory usage
    • Load balancer
    • Ne network traffic
  • Puppet: efficiency of server management

    • Introduction of new server
    • Setting change of an existing server
    • Where the number is large, manual priority is a place setting leakage is likely to occur is
  • deamontools: operation management of the daemon

    • Me re-start automatically when the process fell
    • Make the easy daemon
  • PXE, initramfs: take advantage of the network boot

    • To read from a server on a network, there is no need to secondary storage device
    • Load balancer and DB, is often utilized in the file server
  • Remote maintenance

    • Although it easier and remote login with SSH, can not be used with the OS is not running
    • To deal with a serial console, IPMI, etc.
  • Handling of Web server log

    • To make it easier to handle the logs, aggregated in one place, there is a need to collect
    • It is used in the summary and analysis of the problems of access