HP Brocade 8/12c Brocade Fabric OS Documentation Updates - Supporting Fabric O - Page 63

Documentation updates for Fabric OS v6.4.0 and v6.4.1 8 1. The initiator sends a write command. 2. The PT responds with a Transfer Ready, enabling the host to send more data. 3. The target device processes the write command, and sends a Transfer Ready to solicit more data. The PI intercepts the Transfer Ready, and begins sending data received from the host. 4. The PI continues to stage data received from the initiator, respond locally to a Transfer Ready, and send the data to the target device until the target device sends an FCP_RSP. FIGURE 2 How FCFW works FCFW can improve Write performance. Read performance is unaffected. The gains seen from enabling FCFW depend on several factors, including the following: • The size of I/O vs. Transfer Ready. In general, the more times a target device sends a Transfer Ready, the greater the performance gain. • The number of outstanding I/Os (both Write and Read), link speed, and link congestion. FCFW may not result in significant improvement if these factors suggest that the write data is delayed because it is sharing bandwidth. • Target response latency - If the target is slow in responding to the write command, the data must be held by the remote switch. FCFW flow configuration requirements FCFW is enabled on both initiator and target ports. There is a possibility where either initiator or target ports may have flows with other devices that do not support FCFW. An FCFW-specific zone configuration is used to filter FCFW flows. Hardware considerations for FCFW FCFW is implemented in a hardware configuration consisting of two Brocade 7500 switches or two Brocade 48000 or DCX directors with FR4-18i blades connected by Fibre Channel ISLs. Consider the following hardware characteristics and requirements when planning to implement FCFW: Fabric OS Documentation Updates 51 53-1002063-09