Microsoft Technology Licensing, LLC

Venkataraman Ramanathan

Yun Lin

Nk Srinivas

David Matthew Kruse

Abolade Gbadegesin

Vadim Eydelman

Shishir P Pardikar

Ramanathan Ganapathy

Melur K Raghuraman

A method of purging cached open files from a client-side delayed close list on a remote file system. The remote file system stores a unique file identifier for each cached open file which is independent of the cached file's namespace. In response to a delete or rename command of a remote target file, the remote file system obtains a target file identifier from the target file's server. The remote file system then compares the target file identifier with each of the cached file's unique file identifier. If a cached file's unique file identifier matches the target file identifier, the cached file is purged from the delayed close list.

Selective file purging for delete or rename

An opportunistic directory cache and method of usage reduce the number of calls between a client and server for file information, reducing the wastage of network resources and server resources. The expirable directory cache can be refreshed any time an appropriate directory enumeration call is made to the server, and is capable of being partially rather than completely invalidated in response to an indication that the file information of a file listed in the cache has or will change. If the affected file is identifiable, the entry in the cache for that file is invalidated, i.e. it is no longer useful for serving file information. If the affected file is not identifiable, then a first character projection of the file name is invalidated in the cache. In this manner, the directory cache is maintained with minimum resources and is persisted to the greatest extent possible, increasing its likelihood of usefulness.

Opportunistic directory cache

A system and method are presented that enhance the performance of single and multiple-processor servers by taking advantage of the half-duplex nature of many HTTP requests. Upon receipt of an HTTP Get resource request, a single send and disconnect IRP is generated. The semantics of the send and disconnect IRP completion recognizes the half-duplex nature of the transaction by indicating a graceful close of the TCP/IP connection. This send and disconnect IRP is also completed without a queuing stage (queuelessly) on the processor that serviced the client request. To further enhance performance, the server FIN message to the client is included in the last data frame transmitted to the server. The invention also contemplates transmission of a single HTTP get and disconnect request by a client to allow acknowledgementless disconnection of the TCP/IP connection once a requested resource has been sent.

System and method of enhancing web server throughput in single and multiple processor systems

A system and method to improve server throughput is presented. This system and method stores only selected information from the server's TCP/IP Transmission Control Block (TCB) in the time-wait state. This allows for the release of a portion of the non-paged pool (NPP) memory that was used by the TCB to previously service the now closing connection. This newly released NPP memory may then be used by the server to service additional TCP/IP connections, thus improving the throughput of the server. This small time-wait TCB (TWTCB) stores only an amount of information necessary to uniquely identify the connection to which it relates. Information that was stored to allow servicing of an active connection, but that is not needed to merely identify the connection, is not kept in the TWTCB. By maintaining information that allows for a unique identification of a closing connection, late routed packets for that connection are prevented from establishing a new connection with the server. By minimizing the amount of information to accomplish this unique identification, additional connections to new client may be established without reaching a NPP memory limit.

System and method of enhancing server throughput by minimizing timed-wait TCP control block (TWTCB) size

A method of tracing data on a network that detects receive and send events in a protocol stack is provided. The method provides key information for capacity planners for correctly charging network traffic to the individual services and applications. The TCP/IP stack is instrumented at the transport layer, so that Input/Output Request packets (IRP) representing sends and receives can be detected as they pass through the stack. When such packets are detected an appropriate send or receive event is recorded in a trace log.

Growing community of inventors

Redmond, WA, United States of America

Venkataraman Ramanathan