Winter is coming !!

일반 텍스트 압축 프로그램

XML은 트리 구조의 데이터를 텍스트로 표시한다. XML 문서를 압축하기 위한 간단한 논리적 접근 방법은 기존의 범용 텍스트 압축 도구를 사용하는 것이다. 지난 수십 년간에 걸쳐 텍스트 데이터를 효율적으로 압축하기 위해 수많은 알고리즘이 고안되었다. 이 그룹에 속하는 압축 프로그램 중 가장 대중적이고 효율적인 대표적 프로그램은 gzip, bzip2 및 PPM 압축 프로그램이다.

gzip 압축 프로그램은 LZ77 알고리즘과 호프만 압축 코딩의 조합을 사용하는 DEFLATE 무손실 데이터 압축 알고리즘을 기반으로 한다. LZ77 알고리즘에서는 데이터 압축을 위해 이미 인코더와 디코더를 모두 통과한 데이터와 일치하는 참조로 데이터 부분을 대체한다. 호프만 압축 코딩에서는 각 기호에 대한 표시를 선택하기 위해 특정 메소드를 사용하며, 이때 공통 소스 기호를 줄이기 위해 더 짧은 비트 문자열을 사용하는 가장 공통적인 문자가 사용된다.

bzip2 압축 프로그램에서는 Burrows-Wheeler 변환을 사용하여 자주 되풀이되는 문자 시퀀스를 같은 문자로 구성된 문자열로 변환한 다음, MTP(move-to-front) 변환을 적용하고, 마지막으로 호프만 압축 코딩을 적용한다. Burrows-Wheeler 변환에서는 원래 문자열에 자주 나오는 하위 문자열이 여러 개 있는 경우 변환된 문자열에 단일 문자가 한 행에 여러 번 반복되는 곳이 여러 군데 있는 식으로, 문자들의 순서가 변경된다. 이 접근 방법은 반복되는 문자들이 이어지는 문자열을 쉽게 압축하는 데 도움이 되므로 압축에 유용하게 사용된다. 실제로, bzip2 압축은 gzip을 이용한 압축보다 높은 압축비로 파일을 압축하지만, 성능은 더 낮다.

PPM은 컨텍스트 모델링 및 예측을 바탕으로 한 적응형 통계 데이터 압축 기술이다. PPM에서는 순서가 고정된 여러 가지 컨텍스트 모델을 함께 조합하여 입력 시퀀스에서 다음 문자를 예측하는 유한 컨텍스트 통계 모델링 기술을 사용한다. 이 모델의 각 컨텍스트에 대한 예측 확률은 빈도 개수로부터 계산된다. 이는 적절히 업데이트됩니다. 실제로 나타나는 기호는 산술 코딩을 사용하여 예측한 분포와 상대적으로 인코드된다. PPM은 간단하고 지금까지 제시한 압축 프로그램 중에서 가장 효율적이지만, 계산에 따른 비용이 가장 많이 들기도 한다.

실제로는, 아카이브 목적으로 또는 데이터 교환 프로세스 중의 네트워크 대역폭 감소를 위해 일반 텍스트 압축 프로그램이 사용된다. 일반적으로, 압축비와 압축/압축 풀기 시간이라는 두 가지 기본 메트릭의 측면에서 이런 압축 프로그램 간에 절충점이 있다. 한편으로는, PPM 압축이 압축비는 가장 높은 반면 gzip은 압축비가 가장 낮다. 다른 한편으로는, gzip이 압축/압축 풀기 시간 면에서 최고의 성능을 발휘하는 반면 PPM의 압축/압축 풀기 시간은 훨씬 오래 걸린다. bzip2는 두 메트릭에 대해 중간 위치를 차지한다. 따라서 사용자는 이 두 가지 메트릭을 기준으로 주로 사용 시나리오의 요구사항에 따라 사용할 압축 프로그램을 선택하게 된다.

쿼리 불가능한(아카이브) XML 압축 프로그램

이 그룹에 속하는 XML 압축 프로그램은 최고의 압축비를 달성하는 것이 주 목적이므로 압축된 형식에 대해 어떤 쿼리도 처리할 수 없다. 이 섹션에서는 다음과 같이 이 그룹의 두 가지 주 클래스의 대표적 압축 프로그램에 대해 설명한다.

스키마에 독립적인 압축 프로그램
스키마에 종속적인 압축 프로그램

스키마에 독립적인 압축 스킴

이 클래스의 압축 스킴에서는 인코딩 및 디코딩 프로세스에 대한 스키마 정보의 가용성이 필수적이지 않다. XMill은 XML 문서의 구조를 데이터에서 분리하고 트리에서 데이터 값의 상대 경로와 데이터 유형을 바탕으로 데이터 값을 같은 유형의 컨테이너로 그룹화한다는 새로운 아이디어를 도입한 XML 인식 압축 프로그램을 최초로 구현한 것이다.

XMill에서는 원본 XML 문서의 구조 및 데이터 값 파트가 모두 수집되고 따로 압축된다. 구조 파트에서는 XML 태그 및 속성이 사전 기반 방식으로 인코드된 후 백엔드 일반 텍스트 압축 스킴으로 인코딩을 전달한다. XMill의 구조적 인코딩 스킴은 각각의 개별 요소와 속성 이름에 정수 코드를 지정하며, 이 코드는 요소 및 속성 이름 사전에 대한 키 역할을 한다. 데이터 파트에서는 데이터 값이 경로와 데이터 유형에 따라 유형이 같고 시맨틱상 관련된 컨테이너로 그룹화된다. 그런 다음 각각의 컨테이너는 이 컨테이너의 데이터 유형에 이상적인 전문 압축 프로그램을 사용하여 따로 압축된다. 이 그룹화 조작을 통해 반복을 현지화하므로 압축비가 향상된다.

최신 버전의 XMill 소스 배포에서는 gzip, bzip2 및 PPM의 세 가지 대체 백엔드 범용 압축 프로그램 중 하나로 압축된 형식의 중간 2진 데이터가 전달될 수 있다. 그림 4는 XML 구문 분석기, 구조 및 데이터 컨테이너, 하나 이상의 압축 스킴, 압축된 XML 파일(압축된 구조와 압축된 데이터가 있음)을 포함한 XMill 압축 프로그램의 일반적인 아키텍처 개요를 나타낸 것이다.

그림 4. XMill 압축 프로그램의 일반적인 아키텍처

그림 5는 XML 파일을 구조 및 데이터 컨테이너로 분할하는 예를 보여준다. 요소 및 속성 테이블은 XML 문서의 구조 컨테이너를 저장한다. 각 고유 경로의 값(요소 또는 속성)은 별도의 테이블(컨테이너)에 저장된다. 따라서 각 컨테이너의 값은 같은 유형이 되어 더욱 효율적으로 압축될 수 있다. 이런 예로는 별개의 요소(customers, customers/customer, customers/customer/firstName, customers/customer/lastName, customers/customer/invoice, customers/customer/invoice/items, customers/customer/invoice/items/item) 및 속성(customers/customer/@id, customers/customer/invoice/@total) 테이블이 포함된다.

그림 5. XML 파일을 구조 및 데이터 컨테이너로 분할하는 예

그림 6은 XMill 압축 프로그램의 명령행 옵션을 나타낸 것이다.

그림 6. XMill 압축 프로그램의 명령행 옵션

그림 7은 XMill 압축 프로그램을 사용하여 XML 문서(tpc.xml, 크기는 282KB)를 압축한 효과를 보여주는 것으로, 출력된 압축 파일(tpc.xmi, 크기는 41KB) 크기가 원본 XML 파일 크기의 15%에 해당하는 것을 알 수 있다.

그림 7. XMill 압축 프로그램을 사용하여 XML 파일을 압축한 결과물

XMLPPM은 MHM이라는 멀티플렉싱된 계층 구조 PPM 모델을 사용하는 스트리밍 XML 압축 프로그램이다. XMLPPM은 부분 일치 압축 스킴(PPM)에 의해 일반적인 용도로 예측한 결과에 따라 조정된 것으로 간주된다. XMLPPM에서는, 우선 SAX 구문 분석기를 사용하여 XML 파일을 구문 분석하여 SAX 이벤트 스트림을 생성한다. 각각의 이벤트는 ESAX(인코드된 SAX)라는 바이트코드 표시를 사용하여 인코드된다. ESAX 바이트코드는 구문 구조(요소, 문자, 속성 및 기타 기호)를 바탕으로 여러 가지 멀티플렉싱된 PPM 모델 중 하나를 사용하여 인코드된다. XMLPPM 압축 프로그램의 변형으로서 SCMPPM 압축 프로그램이 제안되었다. SCMPPM은 구조 컨텍스트 모델링(SCM)이라는 기술을 PPM 압축 스킴과 결합한 것이다. SCMPPM은 별도의 모델을 사용하여 각 요소 기호 아래에서 텍스트 컨텐츠를 압축하므로 XMLPPM보다 더 큰 PPM 모델 세트를 사용한다.

XMill과 XMLPPM의 압축비 및 압축/압축 풀기 시간은 백엔드 범용 압축 프로그램(gzip, bzip2 또는 PPM)에 매우 종속적이며 관련이 깊다. 따라서 이들도 범용 백엔드 압축 프로그램과 같은 절충 사항이 있다.

스키마에 종속적인 압축 스킴

이 클래스의 압축 프로그램에서는 인코딩 및 디코딩 프로세스 중에 XML 문서 스키마 정보의 가용성이 필수적이다. 예를 들어, XAUST XML 압축 프로그램은 DTC에 있는 각 요소에 대해 하나씩, DTD의 스키마 정보를 결정적 유한 오토마톤(DFA) 세트로 변환한다. 그런 다음, 각각의 전환은 요소별로 레이블이 지정되고 전환과 연관된 조치는 그 전환에 레이블을 지정하는 요소에 대한 DFA의 시뮬레이터에 대한 호출이다. XAUST는 동일 요소에 대한 모든 데이터를 단일 컨테이너로 그룹화하고, 산술 4차 압축 프로그램에 대해 단일 모델을 사용하여 이 컨테이너를 점진적으로 압축한다. XAUST는 DTD 스키마 정보를 사용하여 문서의 구조를 추적하고 예상되는 기호를 정확히 예측할 수 있다. 예상되는 기호가 고유한 경우에는 디코더가 DTD에서 같은 모델을 생성하고 따라서 고유의 예상 기호를 생성할 수 있으므로 기호를 인코드할 필요가 전혀 없다.

RNGzip XML 도구는 주어진 Relax NG 스키마를 준수하는 XML 문서를 압축한다. RNGzip에서는 송신자와 수신자가 정확히 같은 스키마에 대해 미리 합의해야 한다. 이런 점에서, 스키마는 암호화 및 복호화를 위한 공유 키와 같다. RNGzip에서는 Relax NG 스키마 유효성 검증기를 사용하여 지정된 스키마에서 결정적 트리 오토마톤을 빌드한다. 그런 다음, XML 문서가 지정되면 오토마톤에서 XML을 허용하는지 확인한다. 이 오토마톤이 주어지면, 수신자가 아주 약간의 정보를 전송하여 전체 XML 문서를 다시 생성할 수 있다. 오토마톤에 선택 지점이 있는 경우 RNGzip은 선택된 전환만 전송하고, 텍스트 전환이 있는 경우에는 일치하는 텍스트가 전송된다.

이론적으로, 스키마에 종속적인 압축 프로그램은 스키마에 독립적인 압축 프로그램보다 압축비가 약간 더 높을 수 있다. 그러나 XML 문서의 스키마 정보를 항상 사용할 수 있는 것은 아니며 그에 따라 반구조적 데이터를 표시할 수 있는 XML의 유연성이 지닌 이점을 잃게 되므로 이런 압축 프로그램을 선호하지 않거나 흔히 사용하지는 않는다. 이 유형의 압축 프로그램은 미리 정의된 스키마로 XML 문서를 압축하는 데 사용될 때만 효과적일 수 있다.

쿼리 가능한 XML 압축 프로그램

쿼리 가능한 XML 압축 프로그램의 주요 목적은 전체 문서의 압축을 풀지 않고 압축된 형식에 대해 쿼리를 직접 평가할 수 있도록 하는 것이다. 이 그룹의 압축비는 보통 아카이브 XML 압축 프로그램의 압축비보다 낮다. 이런 유형의 압축 프로그램은 모바일 디바이스 및 GPS 시스템과 같이 자원이 한정된 컴퓨팅 디바이스에 호스트되어 있는 수많은 애플리케이션에 매우 중요하다. 이 섹션에서는 다음과 같이 이 그룹의 두 가지 주 클래스의 대표적 압축 프로그램에 대해 설명한다.

준동형 압축 프로그램
비 준동형 압축 프로그램

준동형 쿼리 가능 XML 압축 프로그램

이 클래스의 압축 프로그램은 압축된 형식으로 XML 문서의 원래 구조를 유지하므로, 원본 형식과 같은 방식으로 액세스 및 구문 분석할 수 있다. XGrind는 압축된 XML 문서의 압축을 완전히 풀 필요 없이 쿼리를 지원하는 최초의 XML 인식 압축 스킴이다. XGrind는 구조에서 데이터를 분리하지 않는다. 그래서 XML 문서의 원래 구조가 유지된다.

XGrind 압축 형식의 준동 특성 덕분에 XGrind에서 다음과 같이 다양하고 흥미로운 기능을 사용할 수 있다.

압축된 XML 문서는 태그와 해당 인코딩으로 대체된 요소/속성 값을 가진 원본 XML 문서로 볼 수 있다. 그러므로 XGrind를 확장된 SAX 구문 분석기로 간주할 수 있다.
XML 인덱싱 기술은 일반 XML 문서에서 빌드할 수 있는 것과 유사한 방법으로 압축된 문서에서 빌드할 수 있다. XGrind에서는 요소 및 속성 이름이 사진 기반 인코딩 스킴을 사용하여 인코드되고 문자 데이터는 반 적응형 호프만 압축 코딩을 사용하여 압축된다. XGrind의 쿼리 프로세서는 압축된 값에 대한 정확한 일치 및 접두부 일치 쿼리와 압축이 풀린 값에 대한 부분 일치 및 범위 쿼리만 처리할 수 있다. 그러나 XGrind에서는 압축된 도메인에서의 비 동등 선택과 같은 여러 가지 조작을 지원하지 않는다. 따라서 XGrind는 결합, 집계, 중첩된 쿼리 또는 생성 조작을 수행할 수 없다.

XPress는 특성의 조합을 사용하여 XML 데이터를 효율적으로 압축하고 검색하는 또 다른 준동형 쿼리 가능 XML 압축 프로그램이다. XML 문서의 레이블과 경로를 인코드하기 위해, XPress는 서로 다른 간격으로 역 산술 인코딩 방법을 사용한다. 간격 사이에 제약 관계를 사용하여 압축된 XML 데이터에 대해 경로 표현식이 평가된다. XPress의 압축 스킴은 입력 파일의 임시 스캔을 사용하여 통계 정보를 수집하는 반 적응형 스킴이고, 데이터에 대한 인코딩 규칙은 데이터의 위치에 상관없다. 또한, 이 스킴에서는 자동 유추 유형 정보를 바탕으로 데이터 값에 알맞은 인코더를 사용한다.

비 준동형 쿼리 가능 XML 압축 프로그램

이 클래스의 압축 프로그램은 XML 문서의 인코딩 프로세스 중에 데이터 파트에서 구조 파트를 분리한다. 따라서 준동형 클래스와는 달리, 압축된 형식의 구조는 원본 XML 문서의 구조와 다르고 압축 풀기 프로세스 중 다른 방식으로 구문 분석되어야 한다. 하지만, 더 높은 압축비를 달성할 수 있다. 예를 들어, XSeq는 문법 기반 쿼리 가능 XML 압축 스킴으로서, 유명한 문법 기반 텍스트 문자열 압축 알고리즘인 Sequitur를 적절히 조정한 것으로 간주된다.

XSeq에서는 입력 XML 파일의 토큰이 컨테이너 세트로 분리되고 그 각각은 Sequitur를 사용하여 압축된다. Sequitur 압축 알고리즘은 주어진 문자열 입력에 대해 컨텍스트가 없는 문법을 형성하는 선형 시간 온라인 알고리즘이다. XSeq는 정의된 컨텍스트가 없는 문법을 사용하여 부적절한 압축 데이터의 순차적 스캔을 피하고 주어진 쿼리에 의해 일치되는 데이터 값만 처리한다. 그 밖에도, XSeq는 컨텍스트가 없는 문법을 통해 전체 또는 부분적으로 압축을 풀지 않고 압축된 파일에 대해 직접 쿼리를 처리할 수 있다. 다른 컨테이너에 저장된 데이터 값을 연결하고 쿼리 평가 시간을 단축하기 위해, XSeq는 압축된 파일 내에 저장되고 메모리에 로드된 후 규칙 컨텐츠를 처리하는 인덱스 세트를 사용한다. 예를 들어, XSeq는 압축을 풀지 않고 컨테이너에 있는 각각의 데이터 값을 빠르게 찾을 수 있지만 헤더 인덱스에는 파일에 있는 각 컨테이너의 입구에 대한 포인터 목록이 있는 구조적 인덱스를 사용한다.

TREECHOP XML 압축 프로그램은 XML 문서의 SAX 기반 구문 분석을 수행하여 압축 프로세스를 시작하고 구문 분석된 토큰은 첫 번째 레벨의 깊이에 있는 압축 스트림에 작성되는 또 다른 쿼리 가능 XML 압축 프로그램이다. 각 노드에 대한 코드 워드에는 상위의 코드 워드가 접두부로 표시되고, XML 문서 트리에 있는 두 개의 노드가 같은 경로를 가진 경우 같은 코드 워드를 공유한다. 각각의 CDATA 섹션, 주석, 처리 명령어 및 리프가 아닌 노드에 2진 코드 워드가 지정된다. 이 코드 워드는 트리 노드의 경로를 바탕으로 고유하게 지정된다. 트리 노드 인코딩이 첫 번째 레벨의 깊이에서 압축 스트림에 작성되기 때문에, 압축 풀기 프로그램이 적응형 인코딩 정보를 점진적으로 사용하여 원본 XML 문서를 다시 생성할 수 있다. TREECHOP에서는 압축 스트림을 통해 단일 스캔을 사용하여 정확한 일치 및 범위 쿼리를 수행할 수 있다.

XQuec 시스템에서는 XML 문서 내부의 분리 구조 및 컨텐츠를 바탕으로 압축된 XML 문서에 대해 단편화 및 스토리지 모델을 사용한다. 그 밖에도, XQuec 시스템은 같은 그룹에 속한 컨테이너가 쿼리 조건부에도 함께 나타나도록 컨테이너를 그룹화하는 방법의 올바른 선택에 의존한다. 압축된 도메인 내에서 조건부의 평가를 수행하기 위해, 이 시스템에서는 조건부에서 관련된 컨테이너가 같은 그룹에 속하고 압축된 도메인에서 그 조건부를 지원하는 알고리즘으로 압축되도록 한다. 조건부에 대한 정보는 사용 가능한 쿼리 워크로드를 사용하여 유추된다. XQueC에서는 쿼리 워크로드 정보를 이용해 소스 모델에 따라 컨테이너를 세트로 파티션하고 가장 적합한 압축 알고리즘을 각 세트에 적절히 지정한다. XQueC는 XML 쿼리를 평가하기 위한 대수도 고안했다. 일반 연산자와 압축 인식 연산자를 자유롭게 혼합할 수 있는 비용 기반 최적화 프로그램에서 이 대수를 이용한다.

결론

개요를 설명하는 본 기사에서는 최첨단 XML 압축 기술에 대해 살펴보았다. 이 도메인에서 XMill에 의한 첫 번째 구현에서 XML 압축 메커니즘의 주요 혁신 기술이 제시되었다. 여기서는 데이터 파트에서 XML 문서의 구조 파트를 분리한 다음 따로 압축할 수 있는 같은 유형의 컨테이너로 관련 데이터 항목을 그룹화하는 개념을 도입했다. 중복 데이터를 더 쉽게 발견할 수 있기 때문에, 이런 분리를 통해 범용 압축 프로그램 또는 기타 압축 메커니즘으로 유형이 같은 이런 컨테이너를 압축하는 추가 단계를 개선할 수 있다.

나머지 XML 압축 프로그램은 대부분 이 아이디어를 다른 식으로 시뮬레이션했다. 압축 시간 및 압축 풀기 시간 메트릭은 여러 가지 XML 압축 기술을 구분하는 데 결정적인 역할을 한다. 원칙적으로, XML 문서의 스키마 정보를 항상 사용할 수 있는 것은 아니며 필요한 형식(DTD, XML 스키마, RElaxNG)으로 되어 있는 것도 아니기 때문에 실제로는 스키마에 종속적인 XML 압축 프로그램을 선호하지 않거나 흔히 사용하지는 않는다. 쿼리 가능 XML 압축 프로그램은 수많은 애플리케이션에 매우 중요하지만, 문법 기반 XML 압축 기술과 쿼리 가능 XML 압축 프로그램에 대한 확실한 구현은 공개적으로 제공되지 않는다. 이 두 가지 영역은 추가적인 연구와 개발을 위한 다양하고 흥미로운 접근 수단을 제공한다.

자료출처 : IBM(원문)

참고자료 : http://www.http-compression.com/

Best Practices for Speeding Up Your Web Site

2011-11-09T13:57:00.003+09:00

Yahoo 사이트에 흥미있는 글이 올라 있네요.

Minimize HTTP Requests

tag: content

80% of the end-user response time is spent on the front-end. Most of this time is tied up in downloading all the components in the page: images, stylesheets, scripts, Flash, etc. Reducing the number of components in turn reduces the number of HTTP requests required to render the page. This is the key to faster pages.

One way to reduce the number of components in the page is to simplify the page's design. But is there a way to build pages with richer content while also achieving fast response times? Here are some techniques for reducing the number of HTTP requests, while still supporting rich page designs.

Combined files are a way to reduce the number of HTTP requests by combining all scripts into a single script, and similarly combining all CSS into a single stylesheet. Combining files is more challenging when the scripts and stylesheets vary from page to page, but making this part of your release process improves response times.

CSS Sprites are the preferred method for reducing the number of image requests. Combine your background images into a single image and use the CSS background-image and background-positionproperties to display the desired image segment.

Image maps combine multiple images into a single image. The overall size is about the same, but reducing the number of HTTP requests speeds up the page. Image maps only work if the images are contiguous in the page, such as a navigation bar. Defining the coordinates of image maps can be tedious and error prone. Using image maps for navigation is not accessible too, so it's not recommended.

Inline images use the data: URL scheme to embed the image data in the actual page. This can increase the size of your HTML document. Combining inline images into your (cached) stylesheets is a way to reduce HTTP requests and avoid increasing the size of your pages. Inline images are not yet supported across all major browsers.

Reducing the number of HTTP requests in your page is the place to start. This is the most important guideline for improving performance for first time visitors. As described in Tenni Theurer's blog postBrowser Cache Usage - Exposed!, 40-60% of daily visitors to your site come in with an empty cache. Making your page fast for these first time visitors is key to a better user experience.

Use a Content Delivery Network

tag: server

The user's proximity to your web server has an impact on response times. Deploying your content across multiple, geographically dispersed servers will make your pages load faster from the user's perspective. But where should you start?

As a first step to implementing geographically dispersed content, don't attempt to redesign your web application to work in a distributed architecture. Depending on the application, changing the architecture could include daunting tasks such as synchronizing session state and replicating database transactions across server locations. Attempts to reduce the distance between users and your content could be delayed by, or never pass, this application architecture step.

Remember that 80-90% of the end-user response time is spent downloading all the components in the page: images, stylesheets, scripts, Flash, etc. This is the Performance Golden Rule. Rather than starting with the difficult task of redesigning your application architecture, it's better to first disperse your static content. This not only achieves a bigger reduction in response times, but it's easier thanks to content delivery networks.

A content delivery network (CDN) is a collection of web servers distributed across multiple locations to deliver content more efficiently to users. The server selected for delivering content to a specific user is typically based on a measure of network proximity. For example, the server with the fewest network hops or the server with the quickest response time is chosen.

Some large Internet companies own their own CDN, but it's cost-effective to use a CDN service provider, such as Akamai Technologies, EdgeCast, or level3. For start-up companies and private web sites, the cost of a CDN service can be prohibitive, but as your target audience grows larger and becomes more global, a CDN is necessary to achieve fast response times. At Yahoo!, properties that moved static content off their application web servers to a CDN (both 3rd party as mentioned above as well as Yahoo’s own CDN) improved end-user response times by 20% or more. Switching to a CDN is a relatively easy code change that will dramatically improve the speed of your web site.

Add an Expires or a Cache-Control Header

tag: server

There are two aspects to this rule:

For static components: implement "Never expire" policy by setting far future Expires header
For dynamic components: use an appropriate Cache-Control header to help the browser with conditional requests

Web page designs are getting richer and richer, which means more scripts, stylesheets, images, and Flash in the page. A first-time visitor to your page may have to make several HTTP requests, but by using the Expires header you make those components cacheable. This avoids unnecessary HTTP requests on subsequent page views. Expires headers are most often used with images, but they should be used onall components including scripts, stylesheets, and Flash components.

Browsers (and proxies) use a cache to reduce the number and size of HTTP requests, making web pages load faster. A web server uses the Expires header in the HTTP response to tell the client how long a component can be cached. This is a far future Expires header, telling the browser that this response won't be stale until April 15, 2010.

      Expires: Thu, 15 Apr 2010 20:00:00 GMT

If your server is Apache, use the ExpiresDefault directive to set an expiration date relative to the current date. This example of the ExpiresDefault directive sets the Expires date 10 years out from the time of the request.

      ExpiresDefault "access plus 10 years"

Keep in mind, if you use a far future Expires header you have to change the component's filename whenever the component changes. At Yahoo! we often make this step part of the build process: a version number is embedded in the component's filename, for example, yahoo_2.0.6.js.

Using a far future Expires header affects page views only after a user has already visited your site. It has no effect on the number of HTTP requests when a user visits your site for the first time and the browser's cache is empty. Therefore the impact of this performance improvement depends on how often users hit your pages with a primed cache. (A "primed cache" already contains all of the components in the page.) We measured this at Yahoo! and found the number of page views with a primed cache is 75-85%. By using a far future Expires header, you increase the number of components that are cached by the browser and re-used on subsequent page views without sending a single byte over the user's Internet connection.

Gzip Components

tag: server

The time it takes to transfer an HTTP request and response across the network can be significantly reduced by decisions made by front-end engineers. It's true that the end-user's bandwidth speed, Internet service provider, proximity to peering exchange points, etc. are beyond the control of the development team. But there are other variables that affect response times. Compression reduces response times by reducing the size of the HTTP response.

Starting with HTTP/1.1, web clients indicate support for compression with the Accept-Encoding header in the HTTP request.

      Accept-Encoding: gzip, deflate

If the web server sees this header in the request, it may compress the response using one of the methods listed by the client. The web server notifies the web client of this via the Content-Encoding header in the response.

      Content-Encoding: gzip

Gzip is the most popular and effective compression method at this time. It was developed by the GNU project and standardized by RFC 1952. The only other compression format you're likely to see is deflate, but it's less effective and less popular.

Gzipping generally reduces the response size by about 70%. Approximately 90% of today's Internet traffic travels through browsers that claim to support gzip. If you use Apache, the module configuring gzip depends on your version: Apache 1.3 uses mod_gzip while Apache 2.x uses mod_deflate.

There are known issues with browsers and proxies that may cause a mismatch in what the browser expects and what it receives with regard to compressed content. Fortunately, these edge cases are dwindling as the use of older browsers drops off. The Apache modules help out by adding appropriate Vary response headers automatically.

Servers choose what to gzip based on file type, but are typically too limited in what they decide to compress. Most web sites gzip their HTML documents. It's also worthwhile to gzip your scripts and stylesheets, but many web sites miss this opportunity. In fact, it's worthwhile to compress any text response including XML and JSON. Image and PDF files should not be gzipped because they are already compressed. Trying to gzip them not only wastes CPU but can potentially increase file sizes.

Gzipping as many file types as possible is an easy way to reduce page weight and accelerate the user experience.

Put Stylesheets at the Top

tag: css

While researching performance at Yahoo!, we discovered that moving stylesheets to the document HEAD makes pages appear to be loading faster. This is because putting stylesheets in the HEAD allows the page to render progressively.

Front-end engineers that care about performance want a page to load progressively; that is, we want the browser to display whatever content it has as soon as possible. This is especially important for pages with a lot of content and for users on slower Internet connections. The importance of giving users visual feedback, such as progress indicators, has been well researched and documented. In our case the HTML page is the progress indicator! When the browser loads the page progressively the header, the navigation bar, the logo at the top, etc. all serve as visual feedback for the user who is waiting for the page. This improves the overall user experience.

The problem with putting stylesheets near the bottom of the document is that it prohibits progressive rendering in many browsers, including Internet Explorer. These browsers block rendering to avoid having to redraw elements of the page if their styles change. The user is stuck viewing a blank white page.

The HTML specification clearly states that stylesheets are to be included in the HEAD of the page: "Unlike A, [LINK] may only appear in the HEAD section of a document, although it may appear any number of times." Neither of the alternatives, the blank white screen or flash of unstyled content, are worth the risk. The optimal solution is to follow the HTML specification and load your stylesheets in the document HEAD.

Put Scripts at the Bottom

tag: javascript

The problem caused by scripts is that they block parallel downloads. The HTTP/1.1 specificationsuggests that browsers download no more than two components in parallel per hostname. If you serve your images from multiple hostnames, you can get more than two downloads to occur in parallel. While a script is downloading, however, the browser won't start any other downloads, even on different hostnames.

In some situations it's not easy to move scripts to the bottom. If, for example, the script usesdocument.write to insert part of the page's content, it can't be moved lower in the page. There might also be scoping issues. In many cases, there are ways to workaround these situations.

An alternative suggestion that often comes up is to use deferred scripts. The DEFER attribute indicates that the script does not contain document.write, and is a clue to browsers that they can continue rendering. Unfortunately, Firefox doesn't support the DEFER attribute. In Internet Explorer, the script may be deferred, but not as much as desired. If a script can be deferred, it can also be moved to the bottom of the page. That will make your web pages load faster.

Avoid CSS Expressions

tag: css

CSS expressions are a powerful (and dangerous) way to set CSS properties dynamically. They were supported in Internet Explorer starting with version 5, but were deprecated starting with IE8. As an example, the background color could be set to alternate every hour using CSS expressions:

      background-color: expression( (new Date()).getHours()%2 ? "#B8D4FF" : "#F08A00" );

As shown here, the expression method accepts a JavaScript expression. The CSS property is set to the result of evaluating the JavaScript expression. The expression method is ignored by other browsers, so it is useful for setting properties in Internet Explorer needed to create a consistent experience across browsers.

The problem with expressions is that they are evaluated more frequently than most people expect. Not only are they evaluated when the page is rendered and resized, but also when the page is scrolled and even when the user moves the mouse over the page. Adding a counter to the CSS expression allows us to keep track of when and how often a CSS expression is evaluated. Moving the mouse around the page can easily generate more than 10,000 evaluations.

One way to reduce the number of times your CSS expression is evaluated is to use one-time expressions, where the first time the expression is evaluated it sets the style property to an explicit value, which replaces the CSS expression. If the style property must be set dynamically throughout the life of the page, using event handlers instead of CSS expressions is an alternative approach. If you must use CSS expressions, remember that they may be evaluated thousands of times and could affect the performance of your page.

Make JavaScript and CSS External

tag: javascript, css

Many of these performance rules deal with how external components are managed. However, before these considerations arise you should ask a more basic question: Should JavaScript and CSS be contained in external files, or inlined in the page itself?

Using external files in the real world generally produces faster pages because the JavaScript and CSS files are cached by the browser. JavaScript and CSS that are inlined in HTML documents get downloaded every time the HTML document is requested. This reduces the number of HTTP requests that are needed, but increases the size of the HTML document. On the other hand, if the JavaScript and CSS are in external files cached by the browser, the size of the HTML document is reduced without increasing the number of HTTP requests.

The key factor, then, is the frequency with which external JavaScript and CSS components are cached relative to the number of HTML documents requested. This factor, although difficult to quantify, can be gauged using various metrics. If users on your site have multiple page views per session and many of your pages re-use the same scripts and stylesheets, there is a greater potential benefit from cached external files.

Many web sites fall in the middle of these metrics. For these sites, the best solution generally is to deploy the JavaScript and CSS as external files. The only exception where inlining is preferable is with home pages, such as Yahoo!'s front page and My Yahoo!. Home pages that have few (perhaps only one) page view per session may find that inlining JavaScript and CSS results in faster end-user response times.

For front pages that are typically the first of many page views, there are techniques that leverage the reduction of HTTP requests that inlining provides, as well as the caching benefits achieved through using external files. One such technique is to inline JavaScript and CSS in the front page, but dynamically download the external files after the page has finished loading. Subsequent pages would reference the external files that should already be in the browser's cache.

Reduce DNS Lookups

tag: content

The Domain Name System (DNS) maps hostnames to IP addresses, just as phonebooks map people's names to their phone numbers. When you type www.yahoo.com into your browser, a DNS resolver contacted by the browser returns that server's IP address. DNS has a cost. It typically takes 20-120 milliseconds for DNS to lookup the IP address for a given hostname. The browser can't download anything from this hostname until the DNS lookup is completed.

DNS lookups are cached for better performance. This caching can occur on a special caching server, maintained by the user's ISP or local area network, but there is also caching that occurs on the individual user's computer. The DNS information remains in the operating system's DNS cache (the "DNS Client service" on Microsoft Windows). Most browsers have their own caches, separate from the operating system's cache. As long as the browser keeps a DNS record in its own cache, it doesn't bother the operating system with a request for the record.

Internet Explorer caches DNS lookups for 30 minutes by default, as specified by the DnsCacheTimeoutregistry setting. Firefox caches DNS lookups for 1 minute, controlled by thenetwork.dnsCacheExpiration configuration setting. (Fasterfox changes this to 1 hour.)

When the client's DNS cache is empty (for both the browser and the operating system), the number of DNS lookups is equal to the number of unique hostnames in the web page. This includes the hostnames used in the page's URL, images, script files, stylesheets, Flash objects, etc. Reducing the number of unique hostnames reduces the number of DNS lookups.

Reducing the number of unique hostnames has the potential to reduce the amount of parallel downloading that takes place in the page. Avoiding DNS lookups cuts response times, but reducing parallel downloads may increase response times. My guideline is to split these components across at least two but no more than four hostnames. This results in a good compromise between reducing DNS lookups and allowing a high degree of parallel downloads.

Minify JavaScript and CSS

tag: javascript, css

Minification is the practice of removing unnecessary characters from code to reduce its size thereby improving load times. When code is minified all comments are removed, as well as unneeded white space characters (space, newline, and tab). In the case of JavaScript, this improves response time performance because the size of the downloaded file is reduced. Two popular tools for minifying JavaScript code areJSMin and YUI Compressor. The YUI compressor can also minify CSS.

Obfuscation is an alternative optimization that can be applied to source code. It's more complex than minification and thus more likely to generate bugs as a result of the obfuscation step itself. In a survey of ten top U.S. web sites, minification achieved a 21% size reduction versus 25% for obfuscation. Although obfuscation has a higher size reduction, minifying JavaScript is less risky.

In addition to minifying external scripts and styles, inlined <script> and <style> blocks can and should also be minified. Even if you gzip your scripts and styles, minifying them will still reduce the size by 5% or more. As the use and size of JavaScript and CSS increases, so will the savings gained by minifying your code.

Avoid Redirects

tag: content

Redirects are accomplished using the 301 and 302 status codes. Here's an example of the HTTP headers in a 301 response:

      HTTP/1.1 301 Moved Permanently
      Location: http://example.com/newuri
      Content-Type: text/html

The browser automatically takes the user to the URL specified in the Location field. All the information necessary for a redirect is in the headers. The body of the response is typically empty. Despite their names, neither a 301 nor a 302 response is cached in practice unless additional headers, such asExpires or Cache-Control, indicate it should be. The meta refresh tag and JavaScript are other ways to direct users to a different URL, but if you must do a redirect, the preferred technique is to use the standard 3xx HTTP status codes, primarily to ensure the back button works correctly.

The main thing to remember is that redirects slow down the user experience. Inserting a redirect between the user and the HTML document delays everything in the page since nothing in the page can be rendered and no components can start being downloaded until the HTML document has arrived.

One of the most wasteful redirects happens frequently and web developers are generally not aware of it. It occurs when a trailing slash (/) is missing from a URL that should otherwise have one. For example, going to http://astrology.yahoo.com/astrology results in a 301 response containing a redirect tohttp://astrology.yahoo.com/astrology/ (notice the added trailing slash). This is fixed in Apache by usingAlias or mod_rewrite, or the DirectorySlash directive if you're using Apache handlers.

Connecting an old web site to a new one is another common use for redirects. Others include connecting different parts of a website and directing the user based on certain conditions (type of browser, type of user account, etc.). Using a redirect to connect two web sites is simple and requires little additional coding. Although using redirects in these situations reduces the complexity for developers, it degrades the user experience. Alternatives for this use of redirects include using Alias and mod_rewrite if the two code paths are hosted on the same server. If a domain name change is the cause of using redirects, an alternative is to create a CNAME (a DNS record that creates an alias pointing from one domain name to another) in combination with Alias or mod_rewrite.

Remove Duplicate Scripts

tag: javascript

It hurts performance to include the same JavaScript file twice in one page. This isn't as unusual as you might think. A review of the ten top U.S. web sites shows that two of them contain a duplicated script. Two main factors increase the odds of a script being duplicated in a single web page: team size and number of scripts. When it does happen, duplicate scripts hurt performance by creating unnecessary HTTP requests and wasted JavaScript execution.

Unnecessary HTTP requests happen in Internet Explorer, but not in Firefox. In Internet Explorer, if an external script is included twice and is not cacheable, it generates two HTTP requests during page loading. Even if the script is cacheable, extra HTTP requests occur when the user reloads the page.

In addition to generating wasteful HTTP requests, time is wasted evaluating the script multiple times. This redundant JavaScript execution happens in both Firefox and Internet Explorer, regardless of whether the script is cacheable.

One way to avoid accidentally including the same script twice is to implement a script management module in your templating system. The typical way to include a script is to use the SCRIPT tag in your HTML page.

      <script type="text/javascript" src="menu_1.0.17.js"></script>

An alternative in PHP would be to create a function called insertScript.

      <?php insertScript("menu.js") ?>

In addition to preventing the same script from being inserted multiple times, this function could handle other issues with scripts, such as dependency checking and adding version numbers to script filenames to support far future Expires headers.

Configure ETags

tag: server

Entity tags (ETags) are a mechanism that web servers and browsers use to determine whether the component in the browser's cache matches the one on the origin server. (An "entity" is another word a "component": images, scripts, stylesheets, etc.) ETags were added to provide a mechanism for validating entities that is more flexible than the last-modified date. An ETag is a string that uniquely identifies a specific version of a component. The only format constraints are that the string be quoted. The origin server specifies the component's ETag using the ETag response header.

      HTTP/1.1 200 OK
      Last-Modified: Tue, 12 Dec 2006 03:03:59 GMT
      ETag: "10c24bc-4ab-457e1c1f"
      Content-Length: 12195

Later, if the browser has to validate a component, it uses the If-None-Match header to pass the ETag back to the origin server. If the ETags match, a 304 status code is returned reducing the response by 12195 bytes for this example.

      GET /i/yahoo.gif HTTP/1.1
      Host: us.yimg.com
      If-Modified-Since: Tue, 12 Dec 2006 03:03:59 GMT
      If-None-Match: "10c24bc-4ab-457e1c1f"
      HTTP/1.1 304 Not Modified

The problem with ETags is that they typically are constructed using attributes that make them unique to a specific server hosting a site. ETags won't match when a browser gets the original component from one server and later tries to validate that component on a different server, a situation that is all too common on Web sites that use a cluster of servers to handle requests. By default, both Apache and IIS embed data in the ETag that dramatically reduces the odds of the validity test succeeding on web sites with multiple servers.

The ETag format for Apache 1.3 and 2.x is inode-size-timestamp. Although a given file may reside in the same directory across multiple servers, and have the same file size, permissions, timestamp, etc., its inode is different from one server to the next.

IIS 5.0 and 6.0 have a similar issue with ETags. The format for ETags on IIS isFiletimestamp:ChangeNumber. A ChangeNumber is a counter used to track configuration changes to IIS. It's unlikely that the ChangeNumber is the same across all IIS servers behind a web site.

The end result is ETags generated by Apache and IIS for the exact same component won't match from one server to another. If the ETags don't match, the user doesn't receive the small, fast 304 response that ETags were designed for; instead, they'll get a normal 200 response along with all the data for the component. If you host your web site on just one server, this isn't a problem. But if you have multiple servers hosting your web site, and you're using Apache or IIS with the default ETag configuration, your users are getting slower pages, your servers have a higher load, you're consuming greater bandwidth, and proxies aren't caching your content efficiently. Even if your components have a far future Expires header, a conditional GET request is still made whenever the user hits Reload or Refresh.

If you're not taking advantage of the flexible validation model that ETags provide, it's better to just remove the ETag altogether. The Last-Modified header validates based on the component's timestamp. And removing the ETag reduces the size of the HTTP headers in both the response and subsequent requests. This Microsoft Support article describes how to remove ETags. In Apache, this is done by simply adding the following line to your Apache configuration file:

      FileETag none

Make Ajax Cacheable

tag: content

One of the cited benefits of Ajax is that it provides instantaneous feedback to the user because it requests information asynchronously from the backend web server. However, using Ajax is no guarantee that the user won't be twiddling his thumbs waiting for those asynchronous JavaScript and XML responses to return. In many applications, whether or not the user is kept waiting depends on how Ajax is used. For example, in a web-based email client the user will be kept waiting for the results of an Ajax request to find all the email messages that match their search criteria. It's important to remember that "asynchronous" does not imply "instantaneous".

To improve performance, it's important to optimize these Ajax responses. The most important way to improve the performance of Ajax is to make the responses cacheable, as discussed in Add an Expires or a Cache-Control Header. Some of the other rules also apply to Ajax:

Let's look at an example. A Web 2.0 email client might use Ajax to download the user's address book for autocompletion. If the user hasn't modified her address book since the last time she used the email web app, the previous address book response could be read from cache if that Ajax response was made cacheable with a future Expires or Cache-Control header. The browser must be informed when to use a previously cached address book response versus requesting a new one. This could be done by adding a timestamp to the address book Ajax URL indicating the last time the user modified her address book, for example, &t=1190241612. If the address book hasn't been modified since the last download, the timestamp will be the same and the address book will be read from the browser's cache eliminating an extra HTTP roundtrip. If the user has modified her address book, the timestamp ensures the new URL doesn't match the cached response, and the browser will request the updated address book entries.

Even though your Ajax responses are created dynamically, and might only be applicable to a single user, they can still be cached. Doing so will make your Web 2.0 apps faster.

Flush the Buffer Early

tag: server

When users request a page, it can take anywhere from 200 to 500ms for the backend server to stitch together the HTML page. During this time, the browser is idle as it waits for the data to arrive. In PHP you have the function flush(). It allows you to send your partially ready HTML response to the browser so that the browser can start fetching components while your backend is busy with the rest of the HTML page. The benefit is mainly seen on busy backends or light frontends.

A good place to consider flushing is right after the HEAD because the HTML for the head is usually easier to produce and it allows you to include any CSS and JavaScript files for the browser to start fetching in parallel while the backend is still processing.

Example:

      ... <!-- css, js -->
    </head>
    <?php flush(); ?>
    <body>
      ... <!-- content -->

Yahoo! search pioneered research and real user testing to prove the benefits of using this technique.

Use GET for AJAX Requests

tag: server

The Yahoo! Mail team found that when using XMLHttpRequest, POST is implemented in the browsers as a two-step process: sending the headers first, then sending data. So it's best to use GET, which only takes one TCP packet to send (unless you have a lot of cookies). The maximum URL length in IE is 2K, so if you send more than 2K data you might not be able to use GET.

An interesting side affect is that POST without actually posting any data behaves like GET. Based on theHTTP specs, GET is meant for retrieving information, so it makes sense (semantically) to use GET when you're only requesting data, as opposed to sending data to be stored server-side.

Post-load Components

tag: content

You can take a closer look at your page and ask yourself: "What's absolutely required in order to render the page initially?". The rest of the content and components can wait.

JavaScript is an ideal candidate for splitting before and after the onload event. For example if you have JavaScript code and libraries that do drag and drop and animations, those can wait, because dragging elements on the page comes after the initial rendering. Other places to look for candidates for post-loading include hidden content (content that appears after a user action) and images below the fold.

Tools to help you out in your effort: YUI Image Loader allows you to delay images below the fold and theYUI Get utility is an easy way to include JS and CSS on the fly. For an example in the wild take a look atYahoo! Home Page with Firebug's Net Panel turned on.

It's good when the performance goals are inline with other web development best practices. In this case, the idea of progressive enhancement tells us that JavaScript, when supported, can improve the user experience but you have to make sure the page works even without JavaScript. So after you've made sure the page works fine, you can enhance it with some post-loaded scripts that give you more bells and whistles such as drag and drop and animations.

Preload Components

tag: content

Preload may look like the opposite of post-load, but it actually has a different goal. By preloading components you can take advantage of the time the browser is idle and request components (like images, styles and scripts) you'll need in the future. This way when the user visits the next page, you could have most of the components already in the cache and your page will load much faster for the user.

There are actually several types of preloading:

Unconditional preload - as soon as onload fires, you go ahead and fetch some extra components. Check google.com for an example of how a sprite image is requested onload. This sprite image is not needed on the google.com homepage, but it is needed on the consecutive search result page.
Conditional preload - based on a user action you make an educated guess where the user is headed next and preload accordingly. On search.yahoo.com you can see how some extra components are requested after you start typing in the input box.
Anticipated preload - preload in advance before launching a redesign. It often happens after a redesign that you hear: "The new site is cool, but it's slower than before". Part of the problem could be that the users were visiting your old site with a full cache, but the new one is always an empty cache experience. You can mitigate this side effect by preloading some components before you even launched the redesign. Your old site can use the time the browser is idle and request images and scripts that will be used by the new site

Reduce the Number of DOM Elements

tag: content

A complex page means more bytes to download and it also means slower DOM access in JavaScript. It makes a difference if you loop through 500 or 5000 DOM elements on the page when you want to add an event handler for example.

A high number of DOM elements can be a symptom that there's something that should be improved with the markup of the page without necessarily removing content. Are you using nested tables for layout purposes? Are you throwing in more <div>s only to fix layout issues? Maybe there's a better and more semantically correct way to do your markup.

A great help with layouts are the YUI CSS utilities: grids.css can help you with the overall layout, fonts.css and reset.css can help you strip away the browser's defaults formatting. This is a chance to start fresh and think about your markup, for example use <div>s only when it makes sense semantically, and not because it renders a new line.

The number of DOM elements is easy to test, just type in Firebug's console:
document.getElementsByTagName('*').length

And how many DOM elements are too many? Check other similar pages that have good markup. For example the Yahoo! Home Page is a pretty busy page and still under 700 elements (HTML tags).

Split Components Across Domains

tag: content

Splitting components allows you to maximize parallel downloads. Make sure you're using not more than 2-4 domains because of the DNS lookup penalty. For example, you can host your HTML and dynamic content on www.example.org and split static components between static1.example.org andstatic2.example.org

For more information check "Maximizing Parallel Downloads in the Carpool Lane" by Tenni Theurer and Patty Chi.

Minimize the Number of iframes

tag: content

Iframes allow an HTML document to be inserted in the parent document. It's important to understand how iframes work so they can be used effectively.

<iframe> pros:

Helps with slow third-party content like badges and ads
Security sandbox
Download scripts in parallel

<iframe> cons:

Costly even if blank
Blocks page onload
Non-semantic

No 404s

tag: content

HTTP requests are expensive so making an HTTP request and getting a useless response (i.e. 404 Not Found) is totally unnecessary and will slow down the user experience without any benefit.

Some sites have helpful 404s "Did you mean X?", which is great for the user experience but also wastes server resources (like database, etc). Particularly bad is when the link to an external JavaScript is wrong and the result is a 404. First, this download will block parallel downloads. Next the browser may try to parse the 404 response body as if it were JavaScript code, trying to find something usable in it.

Reduce Cookie Size

tag: cookie

HTTP cookies are used for a variety of reasons such as authentication and personalization. Information about cookies is exchanged in the HTTP headers between web servers and browsers. It's important to keep the size of cookies as low as possible to minimize the impact on the user's response time.

For more information check "When the Cookie Crumbles" by Tenni Theurer and Patty Chi. The take-home of this research:

Eliminate unnecessary cookies
Keep cookie sizes as low as possible to minimize the impact on the user response time
Be mindful of setting cookies at the appropriate domain level so other sub-domains are not affected
Set an Expires date appropriately. An earlier Expires date or none removes the cookie sooner, improving the user response time

Use Cookie-free Domains for Components

tag: cookie

When the browser makes a request for a static image and sends cookies together with the request, the server doesn't have any use for those cookies. So they only create network traffic for no good reason. You should make sure static components are requested with cookie-free requests. Create a subdomain and host all your static components there.

If your domain is www.example.org, you can host your static components on static.example.org. However, if you've already set cookies on the top-level domain example.org as opposed towww.example.org, then all the requests to static.example.org will include those cookies. In this case, you can buy a whole new domain, host your static components there, and keep this domain cookie-free. Yahoo! uses yimg.com, YouTube uses ytimg.com, Amazon uses images-amazon.com and so on.

Another benefit of hosting static components on a cookie-free domain is that some proxies might refuse to cache the components that are requested with cookies. On a related note, if you wonder if you should use example.org or www.example.org for your home page, consider the cookie impact. Omitting www leaves you no choice but to write cookies to *.example.org, so for performance reasons it's best to use the www subdomain and write the cookies to that subdomain.

Minimize DOM Access

tag: javascript

Accessing DOM elements with JavaScript is slow so in order to have a more responsive page, you should:

Cache references to accessed elements
Update nodes "offline" and then add them to the tree
Avoid fixing layout with JavaScript

For more information check the YUI theatre's "High Performance Ajax Applications" by Julien Lecomte.

Develop Smart Event Handlers

tag: javascript

Sometimes pages feel less responsive because of too many event handlers attached to different elements of the DOM tree which are then executed too often. That's why using event delegation is a good approach. If you have 10 buttons inside a div, attach only one event handler to the div wrapper, instead of one handler for each button. Events bubble up so you'll be able to catch the event and figure out which button it originated from.

You also don't need to wait for the onload event in order to start doing something with the DOM tree. Often all you need is the element you want to access to be available in the tree. You don't have to wait for all images to be downloaded. DOMContentLoaded is the event you might consider using instead of onload, but until it's available in all browsers, you can use the YUI Event utility, which has an onAvailablemethod.

For more information check the YUI theatre's "High Performance Ajax Applications" by Julien Lecomte.

Choose <link> over @import

tag: css

One of the previous best practices states that CSS should be at the top in order to allow for progressive rendering.

In IE @import behaves the same as using <link> at the bottom of the page, so it's best not to use it.

Avoid Filters

tag: css

The IE-proprietary AlphaImageLoader filter aims to fix a problem with semi-transparent true color PNGs in IE versions < 7. The problem with this filter is that it blocks rendering and freezes the browser while the image is being downloaded. It also increases memory consumption and is applied per element, not per image, so the problem is multiplied.

The best approach is to avoid AlphaImageLoader completely and use gracefully degrading PNG8 instead, which are fine in IE. If you absolutely need AlphaImageLoader, use the underscore hack_filter as to not penalize your IE7+ users.

Optimize Images

tag: images

After a designer is done with creating the images for your web page, there are still some things you can try before you FTP those images to your web server.

You can check the GIFs and see if they are using a palette size corresponding to the number of colors in the image. Using imagemagick it's easy to check using
identify -verbose image.gif
When you see an image useing 4 colors and a 256 color "slots" in the palette, there is room for improvement.
Try converting GIFs to PNGs and see if there is a saving. More often than not, there is. Developers often hesitate to use PNGs due to the limited support in browsers, but this is now a thing of the past. The only real problem is alpha-transparency in true color PNGs, but then again, GIFs are not true color and don't support variable transparency either. So anything a GIF can do, a palette PNG (PNG8) can do too (except for animations). This simple imagemagick command results in totally safe-to-use PNGs:
convert image.gif image.png
"All we are saying is: Give PiNG a Chance!"
Run pngcrush (or any other PNG optimizer tool) on all your PNGs. Example:
pngcrush image.png -rem alla -reduce -brute result.png
Run jpegtran on all your JPEGs. This tool does lossless JPEG operations such as rotation and can also be used to optimize and remove comments and other useless information (such as EXIF information) from your images.
jpegtran -copy none -optimize -perfect src.jpg dest.jpg

Optimize CSS Sprites

tag: images

Arranging the images in the sprite horizontally as opposed to vertically usually results in a smaller file size.
Combining similar colors in a sprite helps you keep the color count low, ideally under 256 colors so to fit in a PNG8.
"Be mobile-friendly" and don't leave big gaps between the images in a sprite. This doesn't affect the file size as much but requires less memory for the user agent to decompress the image into a pixel map. 100x100 image is 10 thousand pixels, where 1000x1000 is 1 million pixels

Don't Scale Images in HTML

tag: images

Don't use a bigger image than you need just because you can set the width and height in HTML. If you need
<img width="100" height="100" src="mycat.jpg" alt="My Cat" />
then your image (mycat.jpg) should be 100x100px rather than a scaled down 500x500px image.

Make favicon.ico Small and Cacheable

tag: images

The favicon.ico is an image that stays in the root of your server. It's a necessary evil because even if you don't care about it the browser will still request it, so it's better not to respond with a 404 Not Found. Also since it's on the same server, cookies are sent every time it's requested. This image also interferes with the download sequence, for example in IE when you request extra components in the onload, the favicon will be downloaded before these extra components.

So to mitigate the drawbacks of having a favicon.ico make sure:

It's small, preferably under 1K.
Set Expires header with what you feel comfortable (since you cannot rename it if you decide to change it). You can probably safely set the Expires header a few months in the future. You can check the last modified date of your current favicon.ico to make an informed decision.

Imagemagick can help you create small favicons

Keep Components under 25K

tag: mobile

This restriction is related to the fact that iPhone won't cache components bigger than 25K. Note that this is the uncompressed size. This is where minification is important because gzip alone may not be sufficient.

For more information check "Performance Research, Part 5: iPhone Cacheability - Making it Stick" by Wayne Shea and Tenni Theurer.