Kernel Mode Cache in HTTP 2.0
The HTTP Server version 2.0 API allows applications to cache responses with static content in kernel mode. Increased performance is achieved when requests are served from the kernel cache without transitioning to user mode.
The HTTP Server API applies the appropriate property configurations to all requests served from the kernel cache, including logging responses. Requests that require authentication, however, will not be served from the cache.
The HTTP Server API limits the kernel mode cache to requests that meet the following conditions:
- The request verb is GET and the entire request is received.
- The request must not have an entity body.
- The HTTP protocol is version 1.0 or later.
- The "Translate: f " header is not present.
- Expect headers other than "Expect: 100-Continue" are not present.
- The authorization header is not present.
- The Range and If-Range headers are not present.
In addition to the restrictions on the request, the response must also meet the following conditions:
- Response size is limited to 256 KB, by default. To change the size of the response that is cached, set the UriMaxUriBytes registry value to the required number of bytes.
HKEY_LOCAL_MACHINE System CurrentControlSet Services HTTP Parameters UriMaxUriBytes
- The entire response must be provided in a single call to HttpSendHttpResponse.
- The date header on the response must not be suppressed.
- If the last-modified header is present, the value of the header must have the correct syntax. The time value in this header is used for cache control verification.
- The kernel mode cache has enough space left to store the response.
By default, kernel mode response cache is enabled. If any of the conditions for the request or response listed above are not met, the response will be sent, but it will not be cached. In HTTP Server version 2.0 API, HttpSendHttpResponse includes an optional pCachePolicy parameter to pass the HTTP_CACHE_POLICY structure. Applications use the cache policy structure to configure the cache.