http://www.linuxvirtualserver.org/

“The Linux Virtual Server is a highly scalable and highly available server built on a cluster of real servers, with the load balancer running on the Linux operating system.”

March 31, 2007 nazin

1. Linux Virtual Server(LVS)란?

리눅스 가상 서버란, 한대의 서버로 증가하는 인터넷 사용자를 처리하기가 힘들어 지면서 고가용성 서버를 구축하기 위해 리눅스 머신을 로드 발랜스 하도록 해주는 운영시스템이다.

만약 하나의 노드에서 처리량이 너무 많아서 서비스가 불가능할 경우 간단히 하나의 노드를 병렬 구성으로 추가 함으로써 부하분산을 하도록 하는 것을 말한다. 공개소스로서 이러한 기능을 담당하고 있는 것이다.

2. Linux Virtual Server 스케쥴링

1)라운드-로빈(round-robin)

라운드-로빈 방식은 로드밸런서에 들어오는 요청 패킷들을 차례대로 실제 서버에 할당하는 방식이 다. 이 방식에서 실제 서버의 현재 부하 상황 등은 고려되지 않는다. 단지 차례대로 할당할 뿐이다. 하지만, 이렇게 하더라도 로드밸런싱을 위해 이전에 사용되던 라운드-로빈 DNS 방식에 의해 서버를 할당하는 방식에 비해서는 우수한데, DNS의 경우는 한번 서버가 지정되면 해당 서버에 수많은 요청 패킷이 몰릴 수 있기 때문이다.

2)가중 라운드-로빈(weighted round-robin)

가중 라운드-로빈 방식은 기본적으로 라운드-로빈 방식인데, 각 서버에 서로 다른 가중치를 주어서 할당하는 방식이다. 이 방식을 사용해야 하는 경우는 실제 서버들이 CPU의 수와 성능, 메모리 용량 등 서로 다른 성능을 가지고 있어서, 각 서버를 동등하게 취급할 수 없는 경우이다.

3)최소 연결(least connection)

최소 연결 방식은 실제 서버들 중에서 현재 가장 적은 수의 요청을 처리하고 있는 서버를 선택하여 요청 패킷을 할당하는 방식이다. 이 방식은 실제 서버의 현재 부하 상황을 동적으로 판단하여 요청을 처리하기 때문에, 앞의 두 방식에 비해서 동적으로 우수한 부하 분산 효과를 얻을 수 있다.

4)가중 최소 연결(weighted least connection)

가중 최소 연결 방식은 기본적으로 최소 연결 방식인데, 가중 라운드-로빈 방식과 마찬가지로 각 서버에 서로 다른 가중치를 주어서 할당하는 방식이다.

3. 시스템 구성 및 ipvsadm 설치

1)Load Balancer Server

운영체제 : SULinux1.5

커널 : kernel-2.6.9-42.0.2.ELsmp

IP : 210.224.223.1

가상IP : 210.224.223.10

2)Real Server1(RS1)

운영체제 : Sulinux1.5

커널 : kernel-2.6.9-12.0.2.ELsmp

IP : 210.224.223.11

IP : 192.168.0.11 [NAT 설정시 부여할 사설 ip]

3)Real Server1(RS2)

운영체제 : CentOS4

커널 : kernel-2.6.9-42.0.2.ELsmp

IP : 210.224.223.12

IP : 192.168.0.12 [NAT 설정시 부여할 사설 ip]

4)Load Balancer Server ipvsadm 설치

Load Balancer Server에서 ipvsadm을 다운로드 후 설치를 한다.

커널 2.4.x 이하 버젼에서는 커널 컴파일 및 hidden 패치를 해야 했으나 커널 2.6.x에서는

Virtual Server 기능들이 기본 포함되어 있으므로 아주 간단하다.

shell>wget http://mirror.centos.org/centos/4/extras/i386/RPMS/ipvsadm-1.24-6.i386.rpm

shell>rpm -ivh ipvsadm-1.24-6.i386.rpm

shell>ipvsadm

IP Virtual Server version 1.2.0 (size=4096)

Prot LocalAddress:Port Scheduler Flags

-> RemoteAddress:Port Forward Weight ActiveConn InActConn

설치가 정상적으로 되었다면 위와같이 나타날 것이다.여기서부터 ipvsadm을 이용하여 아래 방식 으로 설정하여 테스트 해보도록 하겠다.

• Direct Routing

• NAT 방식

• IP Tunnling

A.Direct Routing 방식

실제 서버와 부하분산 서버에서 가상 IP 주소를 부여하여 서로 공유한다. 부하분산 서버에도

마찬가지로 가상 IP 주소를 설정한 인터페이스가 있어야하며 이 인터페이스를 이용, 요청

패킷을 받아들이고 선택한 서버에 직접 라우팅할 수 있게 된다.

모든 실제 서버는 가상 IP주소로 설정한 non-arp alias 인터페이스가 있거나 가상 IP 주소로

향하는 패킷을 지역 소켓으로 재지향한다. 그래서 실제 서버 에서 패킷을 지역적으로 처리할

수 있는 구조를 가지게 된다.

Direct Routing 방식의 구조도는 아래와 같다.

A-1.Load Balancer Server 설정법

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경후

shell>sysctl -p

커널파라미터 적용

shell>ifconfig eth0:1 210.224.223.10 up

가상아이피 등록

shell>ipvsadm -A -t 210.224.223.10:80 -s rr

라운드-로빈방식으로 설정

shell>ipvsadm -a -t 210.224.223.10:80 -r 210.224.223.11:80 -g

RS1 서버로 다이렉트 라우팅 설정

shell>ipvsadm -a -t 210.224.223.10:80 -r 210.224.223.12:80 -g

RS2 서버로 다이렉트 라우팅 설정

A-2.Real Server(RS1) 설정법

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경하여야 하며

#for ipvs

net.ipv4.conf.lo.arp_ignore = 1

net.ipv4.conf.lo.arp_announce = 2

net.ipv4.conf.all.arp_ignore = 1

net.ipv4.conf.all.arp_announce = 2

#end of ipvs

#for ipvs ~ #end of ipvs 부분을 추가한후

shell>sysctl -p

커널파라미터 적용

shell>ifconfig lo:0 210.224.223.10 up

가상아이피 등록

A-3.Real Server(RS2) 설정법

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경하여야 하며

#for ipvs

net.ipv4.conf.lo.arp_ignore = 1

net.ipv4.conf.lo.arp_announce = 2

net.ipv4.conf.all.arp_ignore = 1

net.ipv4.conf.all.arp_announce = 2

#end of ipvs

#for ipvs ~ #end of ipvs 부분을 추가한후

shell>sysctl -p

커널파라미터 적용

shell>ifconfig lo:0 210.224.223.10 up

가상아이피 등록

클라이언트에서 가상아이피 210.224.223.10으로 접속하면 우리는 느끼지 못하지만

rr(라운드-로빈)방식을 이용해서 RV1,RV2 서버에 접속하게 된다.

RV1 서버에 장애가 있다면 RV2 서버로 접속하게 된다.

B.NAT 방식

NAT는 특정한 IP 주소를 한 그룹에서 다른 그룹으로 매핑하는 기능이다.

클라이언트가 가상 아이피로 접근시 사설 아이피로 구성된 부하분산 서버 RS1,RS2로 패킷이 향 하게 된다. 사설 아이피로 구성된 부하분산 서버는 패킷의 목적지와 포트 번호를 검사한다.

NAT 방식의 구조도는 아래와 같다

B-1.Balancer Server 설정법

네트웍 디바이스 eth0,eth1 정보

DEVICE=eth0

onBOOT=yes

BOOTPROTO=static

IPADDR=210.224.223.11

NETMASK=255.255.255.0

GATEWAY=210.223.223.1

DEVICE=eth1

onBOOT=yes

BOOTPROTO=static

IPADDR=192.168.0.1

NETMASK=255.255.255.0

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경후

shell>sysctl -p

커널파라미터 적용

shell>ifconfig eth0:1 210.224.223.10 up

가상아이피 등록

shell>ipvsadm -A -t 210.224.223.10:80 -s lc

최소연결 방식으로 설정

shell>ipvsadm -a -t 210.224.223.10:80 -r 192.168.0.11:80 -m

RS1 서버로 NAT 설정

shell>ipvsadm -a -t 210.224.223.10:80 -r 192.168.0.12:80 -m

RS2 서버로 NAT 설정

B-2.Real Server(RS1,RS2) 설정법

RS1

DEVICE=eth0

onBOOT=yes

BOOTPROTO=static

IPADDR=192.168.0.11

NETMASK=255.255.255.0

GATEWAY=192.168.0.1

RS2

DEVICE=eth0

onBOOT=yes

BOOTPROTO=static

IPADDR=192.168.0.12

NETMASK=255.255.255.0

GATEWAY=192.168.0.1

클라이언트에서 가상아이피 210.224.223.10으로 접속하면 우리는 느끼지 못하지만

최소연결 방식을 이용해서 RV1,RV2 서버 중에 접속자가 작은 서버로 된다.

RV1 서버에 장애가 있다면 RV2 서버로 접속하게 된다.

NAT 방식의 경우 패킷 Load Balancer Server를 반드시 거쳐가야 하므로 20대 이상 구성해야

할때는 문제가 발생할 수 있는 단점이 있다.

C.IP Tunneling 방식

IP 터널링 (IP encapsulation)은 IP 데이터그램안에 IP 데이터그램을 넣는 기술로서, 어떤 IP 주 소를 향하는 데이터그램을 감싸 다른 IP 주소로 재지향할 수 있다. IP encapsulation은 현재 엑 스트라넷, 모빌-IP, IP-멀티캐스트, tunnled 호스트나 네트웍 등에 일반적으로 사용되고 있다.

IP Tunneling 방식의 구조도는 아래와 같다

C-1.Balancer Server 설정법

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경후

shell>sysctl -p

커널파라미터 적용

shell>ifconfig eth0:1 210.224.223.10 up

가상아이피 등록

shell>ipvsadm -A -t 210.224.223.10:80 -s wlc

가중치 최소연결 방식으로 설정

shell>ipvsadm -a -t 210.224.223.10:80 -r 210.224.223.11:80 -i

RS1 서버로 Tunneling

shell>ipvsadm -a -t 210.224.223.10:80 -r 210.224.223.12:80 -i

RS2 서버로 Tunneling

C-2.Real Server(RS1,RS2) 설정법

/etc/sysctl.conf에 net.ipv4.ip_forward = 0 부분을 1로 변경하여야 하며

#for ipvs

net.ipv4.conf.lo.arp_ignore = 1

net.ipv4.conf.lo.arp_announce = 2

net.ipv4.conf.all.arp_ignore = 1

net.ipv4.conf.all.arp_announce = 2

#end of ipvs

#for ipvs ~ #end of ipvs 부분을 추가한후

shell>sysctl -p

커널파라미터 적용

RS1

DEVICE=tunl0

onBOOT=yes

BOOTPROTO=static

IPADDR=210.224.223.10

NETMASK=255.255.255.0

RS2

DEVICE=tunl0

onBOOT=yes

BOOTPROTO=static

IPADDR=210.224.223.10

NETMASK=255.255.255.0

또는 각 리얼서버에서

RS1

shell>ifconfig tunl0 210.224.223.10 up

가상아이피를 IPIP Tunnel에 등록한다.

RS2

shell>ifconfig tunl0 210.224.223.10 up

가상아이피를 IPIP Tunnel에 등록한다.

클라이언트에서 가상아이피 210.224.223.10으로 접속하면 우리는 느끼지 못하지만

가중치 최소연결 방식을 이용해서 RV1,RV2 서버 중에 접속자가 작은 서버로 된다.

RV1 서버에 장애가 있다면 RV2 서버로 접속하게 된다.

4. 참고 문헌 & 사이트 정보

http://www.linuxvirtualserver.org

http://linux-ha.org/

http://eelco.maljaars.net/ultramonkey-rhel4.html

Inheritance is a powerful way to achieve code reuse, but it is not always the best tool for the job. Used inappropriately, it leads to fragile software. It is safe to use inheritance within a package, where the subclass and the superclass implementation are under the control of the same programmers. It is also safe to use inheritance when extending classes specifically designed and documented for extension (Item 15). Inheriting from ordinary concrete classes across package boundaries, however, is dangerous. As a reminder, this book uses the word
“inheritance” to mean implementation inheritance (when one class extends another).

The problems discussed in this item do not apply to interface inheritance (when a class implements an interface or where one interface extends another).

Powerful:

Powerful and simple:

The ideal Windows Vista®-based application is both powerful and simple. Of course you want your application to be powerful and of course you want it to be simple, but can you achieve both? There is a natural tension between these goals, but that tension is far from irreconcilable. You can achieve both power and simplicity through carefully balanced feature selection and presentation.

Powerful

What does "power" really mean in terms of software? An application might be considered powerful if it is jam-packed full of features, having a tremendous breadth of functionality in an attempt to be all things to all users. Such a design is not likely to be successful because an untargeted feature set is unlikely to satisfy the needs of anyone. This is not the type of power that we are after.

An application is powerful when it has the right combination of these characteristics:

• Enabling. The application satisfies the needs of its target users, enabling them to perform tasks that they couldn't otherwise do and achieve their goals effectively.
• Efficient. The application enables users to perform tasks with a level of productivity and scale that wasn't possible before.
• Versatile. The application enables users to perform a wide range of tasks effectively in a variety of circumstances.
• Direct. The application feels like it is directly helping users achieve their goals, instead of getting in the way or requiring unnecessary steps. Features like shortcuts, keyboard access, and macros improve the sense of directness.
• Flexible. The application allows users complete, fine-grained control over their work.
• Integrated. The application is well integrated with Microsoft® Windows®, allowing it to share data with other applications.
• Advanced. The application has extraordinary, innovative, start-of-the-art features that are not found in competing solutions.

Some of these characteristics depend upon the perception of the user and are relative to users' current capabilities. What is considered powerful may change over time, so today's advanced search feature might be commonplace tomorrow.

All these characteristics can be combined into our definition of power:

An application is powerful when it enables its target users to realize their full potential efficiently.

Thus, the ultimate measure of power is productivity, not the number of features.

Different users need help in achieving their full potential in different ways. What is enabling to some users might harm versatility, directness, and control for others. Well-designed software must balance these characteristics appropriately. For example, a desktop publishing system designed for nonprofessionals might use wizards to walk users through complex tasks. Such wizards enable the target users to perform tasks that they otherwise wouldn't be able to perform. By contrast, a desktop publishing system for professionals might focus on directness, efficiency, and complete control. For users of such an application, wizards may be limiting and frustrating.

If you do only one thing...
Understand your target users' goals and craft a feature set that enables them to achieve those goals productively.

Simple

We define simplicity as follows:

Simplicity is the reduction or elimination of an attribute of a design that target users are aware of and consider unessential.

In practice, simplicity is achieved by selecting the right feature set and presenting the features in the right way. This reduces the unessential, both real and perceived.

Simplicity is dependent upon the perception of users. Consider how the effect of an automatic transmission depends on a user's perspective:

• For the typical driver (the target user), an automatic transmission eliminates the need for a manual gear shift and clutch, making a car much easier to drive. A manual gear shift and clutch are not essential to the task of driving, so they are removed to achieve simplicity.
• For a professional race car driver, having direct control over the transmission is essential to being competitive. An automatic transmission negatively affects the car's performance, so it is not regarded as resulting in simplicity.
• For a mechanic, an automatic transmission is a more complex mechanism, and therefore isn't easier to repair or maintain than a manual transmission. Unlike the mechanic, the target user is blissfully unaware of this internal complexity.

While different users regard the automatic transmission differently, it's successful because it eliminates the need for unessential knowledge, skill, and effort from its target users. For the typical driver, automatic transmission is a great feature because it just works.

Simplicity vs. ease of use

Simplicity, when correctly applied, results in ease of use. But simplicity and ease of use are not the same concepts. Ease of use is achieved when users can perform a task successfully on their own without difficulty or confusion within a suitable amount of time. There are many ways to achieve ease of use, and simplicity—the reduction of the unessential—is just one of them.

All users, no matter how sophisticated, want to get their work done with a minimum amount of unnecessary effort. All users—even advanced users—are primarily motivated to get their work done, not to learn about computers or your application.

Simplicity is the most effective way to achieve ease of use, and ease of use equals use. Complex, hard-to-use features just don't get used. By contrast, simple, elegant designs that perform their function well are a joy to use. They invoke a positive, emotional response.

For example, consider the wireless networking support in Microsoft Windows XP. Microsoft could have added a wizard to walk users through the configuration process. This approach would have resulted in ease of use but not simplicity, because an unessential feature (the wizard) would have been added. Instead, Microsoft designed wireless networking to configure itself automatically. Users ultimately don't care about the configuration details, so long as it "just works" reliably and securely. This combination of power and simplicity in wireless networking technology has led to its popularity and rapid adoption.

If you do only one thing...
Start your design process with the simplest designs that do the job well.

If you're not satisfied with your current design, start by stripping away all the unessential elements. You will find that what remains is usually quite good.

Obtaining simplicity while maintaining power

Design principles

To obtain simplicity, always design for the probable, not the possible.

The possible

Design decisions based on what's possible lead to complex user interfaces like the Registry Editor, where all actions are equally possible and as a result require equal effort. Since anything can happen, user goals aren't considered in design decisions.

The probable

Design decisions based on the probable lead to simplified, goal- and task-based solutions, where the likely scenarios receive focus and require minimal effort to perform.

The simplicity design principle

To obtain simplicity, focus on what is likely; reduce, hide, or remove what is unlikely; and eliminate what is impossible.

What users will do is far more relevant to design than what they might do.

Design techniques

To obtain simplicity while maintaining power, choose the right set of features, locate the features in the right places, and reduce the effort to use them. This section gives some common techniques to achieve these goals.

Choosing the right feature set

"Perfection is achieved, not when there is nothing more to add,
but when there is nothing left to take away." —Antoine de Saint-Exupery

The following design techniques give your users the features they need while achieving simplicity through actual reduction or removal:

• Determine the features your users need. Understand your users' needs through goal, scenario, and task analysis. Determine a set of features that realizes these objectives.
• Remove unnecessary elements. Remove elements that aren't likely to be used or have preferable alternatives.
• Remove unnecessary redundancy. There might be several effective ways to perform a task. To achieve simplicity, make the hard decision and choose the best one for your target users instead of providing all of them and making the choice an option.
• Make it "just work" automatically. The element is necessary, but any user interaction to get it to work is not because there is an acceptable default behavior or configuration. To achieve simplicity, make it work automatically and either hide it from the user completely or reduce its exposure significantly.

Streamlining the presentation

"The ability to simplify means to eliminate the unnecessary
so that the necessary may speak." —Hans Hofmann

Use the following design techniques to preserve power, while achieving simplicity through the perception of reduction or removal:

• Combine what should be combined. Put the essential features that support a task together so that a task can be performed in one place. The task's steps should have a unified, streamlined flow. Break down complex tasks into a set of easy, clear steps, so that "one" place might consist of several UI surfaces, such as a wizard.
• Separate what should be separated. Not everything can be presented in one place, so always have clear, well-chosen boundaries. Make features that support core scenarios central and obvious, and hide optional functionality or make it peripheral. Separate individual tasks and provide links to related tasks. For example, tasks related to manipulating photos should be clearly separated from tasks related to managing collections of photos, but they should be readily accessible from each other.
• Eliminate what can be eliminated. Take a printout of your design and highlight the elements used to perform the most important tasks. Even highlight the individual words in the UI text that communicate useful information. Now review what isn't highlighted and consider removing it from the design. If you remove the item, would anything bad happen? If not, remove it!
  Consistency, configurability, and generalization are often desirable qualities, but they can lead to unnecessary complexity. Review your design for misguided efforts in consistency (such as having redundant text), generalization (such as having any number of time zones when two is sufficient), and configurability (such as options that users aren't likely to change), and eliminate what can be eliminated.
• Put the elements in the right place. Within a window, an element's location should follow its utility. Essential controls, instructions, and explanations should all be in context in logical order. If more options are needed, expose them in context by clicking a chevron or similar mechanism; if more information is needed, display an infotip on mouse hover. Place less important tasks, options, and Help information outside the main flow in a separate window or page. The technique of displaying additional detail as needed is called progressive disclosure.
• Use meaningful high-level combinations. It is often simpler and more scalable to select and manipulate groups of related elements than individual elements. Examples of high-level combinations include folders, themes, styles, and user groups. Such combinations often map to a user goal or intention that isn't apparent from the individual elements. For example, the intention behind the High Contrast Black color scheme is far more apparent than that of a black window background.
• Select the right controls. Design elements are embodied by the controls you use to represent them, so selecting the right control is crucial to efficient presentation. For example, the font selection box used by Microsoft Word shows both a preview of the font as well as the most recently used fonts. Similarly, the way Word shows potential spelling and grammar errors in place is much simpler than the dialog box alternative, as shown in the beginning of this article.

Reducing effort

"Simple things should be simple.
Complex things should be possible."—Alan Kay

The following design techniques result in reduced effort for users:

• Make tasks discoverable and visible. All tasks, but especially frequent tasks, should be readily discoverable within the user interface. The steps required to perform tasks should be visible and should not rely on memorization.
• Present tasks in the user's domain. Complex software requires users to map their problems to the technology. Simple software does that mapping for them by presenting what is natural. For example, a red-eye reduction feature maps directly to the problem space and doesn't require users to think in terms of details like hues and gradients.
• Put domain knowledge into the program. Users shouldn't be required to access external information to use your application successfully. Domain knowledge can range from complex data and algorithms to simply making it clear what type of input is valid.
• Use text that users understand. Well-crafted text is crucial to effective communication with users. Use concepts and terms familiar to your users. Fully explain what is being asked in plain language so that users can make intelligent, informed decisions.
• Use safe, secure, probable defaults. If a setting has a value that applies to most users in most circumstances, and that setting is both safe and secure, use it as the default value. Make users specify values only when necessary.
• Use constraints. If there are many ways to perform a task, but only some are correct, constrain the task to those correct ways. Users should not be allowed to make readily preventable mistakes.

Simplicity does not mean simplistic

"Everything should be made as simple as possible,
but not simpler."—Albert Einstein

We believe that simplicity is crucial to an effective, desirable user experience—but it is always possible to take a good thing too far. The essence of simplicity is the reduction or elimination of the unessential. Removal of the essential just produces a poor design. If your "simplification" results in users becoming frustrated, confused, unconfident, or unable to complete tasks successfully, you have removed too much.

Simplicity does mean more effort for you

"I have only made this letter longer because I have
not the time to make it shorter."—Blaise Pascal

Obtaining simplicity while preserving power often requires significant internal complexity. It is usually easier to design software that exposes all the technology plumbing than to design one that hides it—the latter requires an excellent understanding of your target users and their goals. Removing a feature requires discipline, as does deciding against adding that cool feature that really isn't practical. Simplicity requires making hard design choices instead of making everything configurable. Complex software often results from a misconception about users: that they value unused features or overly complex features they can't understand.

Powerful and simple

Power is all about enabling your users and making them productive. Simplicity is all about removing the unessential and presenting features the right way. By understanding your target users and achieving the right balance of features and presentation, you can design Windows Vista-based applications that do both.