Block size (BLKSIZE) specifies the maximum length, in bytes, of a physical block of storage in MVS. If BLKSIZE(0) is specified, the system will determine the optimal block size based on the maximum record length (LRECL) and the physical characteristics of the disk, or approximately half of a physical track. Half track blocking is optimal in that it reduces the amount of wasted storage on the disk. If BLKSIZE(0) is specified for Direct Access (DA) data sets, the system does not determine the block size. z/OS NFS Server uses a formula to calculate the block size; see blksize(0 | quan) in Data set creation attributes syntax.

The BLKSIZE is also important to the performance of the z/OS NFS, which performs physical I/O on a block basis. While half track blocking is generally optimal, for z/OS NFS server you should be aware of the relationship between the block size, the file record length, the network packet size and the NFS client mount write and read buffer sizes, wsize and rsize.

A network packet size will be negotiated between the NFS client and the NFS server when the NFS mount is processed. In general, the NFS server will determine the packet size for write operations, and the NFS client will determine the packet size for read operations. For z/OS NFS, this packet size is 8192 bytes, when processing with Sun NFS V2 protocol. For several NFS client implementations the default packet size is 8192. However, there are some implementations with a 4096 byte maximum packet size. The maximum packet size is 65536 bytes when processing with Sun NFS V3 protocol, and larger with the V4 protocol. However, z/OS NFS supports a maximum packet size of 32768 bytes for both the NFS V3 and V4 protocols.

The packet size can also be automatically reset to a smaller value when a constraint is detected. Such a reduction in packet size causes an increase in the number of NFS requests for the server to process the same amount of data. This increases the load on the server and may or may not have been the initial constraint causing the packet size reduction. This can have an adverse effect on z/OS NFS server performance, particularly when processing write requests.

Regardless of the packet size, the z/OS NFS server must reassemble the packets, possibly arriving out of sequence, into records. The z/OS NFS server buffers (see Ordering out-of-sequence data) and orders these packets until a block of data can be written to disk. Write requests, especially those for large files, are processed more efficiently when the likelihood of packets spanning records and blocks is reduced.

When the z/OS NFS server is processing read requests, the physical I/O is again on a block basis. Consequently, read throughput is improved by increasing the block size up to half track blocking.

If files are allocated with a default block size (BLKSIZE=0) on RAMAC 3 or Enterprise Storage Server (ESS™) 2105 DASD with 3390-3 format, the actual block sizes will probably vary between 24 and 27 KB, depending on the specified record length and record format. Table 1 shows the actual allocation block sizes for physical sequential data sets allocated on a RAMAC 3 or ESS 2105 DASD with a variety of record lengths and with both fixed and variable length records.