Write preference pthread_rwlock_wrlock

Db2 pureScale provides extreme availability, extreme scalability, and workload consolidations for mission critical OLTP applications that demand 7x24x365 availability. Today, Db2 pureScale is supported on AIX and Linux on POWER and Intel Linux. A typical Db2 pureScale cluster consists of multiple processing nodes known as members to handle and execute application requests, a primary cluster caching facility (CF) responsible for providing data coherency and consistency across all members, and a secondary CF serving as a standby to the primary CF providing transient failover capability. Each CF server is a multi-threaded process with each thread referred to as CF worker running on a dedicated logical processor handling requests from multiple members in a pureScale deployed with RDMA network. Within each CF server, the global buffer pool (GBP) is the component responsible to keep page caching consistent. It holds data that has been fetched from disks, allowing member to avoid unnecessary I/O. The global lock manager (GLM) is another component in the CF server that provides overall consistency locking protocol, granting or denying individual request for a page in the GBP from each member. In order to provide optimal throughput, the size of the GBP and GLM must be tuned automatically so that memory is allocated or reallocated to area where the CF is in need most. The procedure of resizing of GBP memory is referred to as GBP expansion or shrinkage. These operations require serialization in such a way that allows the resize procedure to complete in time while not blocking other requests which may impact throughput. This serialization, also referred to as GBP resize locking protocol, is at the center of this feature request.