00705: UNIX Data Server speed is really slow with reads/writes or large records, but very fast when dealing with smaller records

Title:

UNIX Data Server speed is really slow with reads/writes or large records, but very fast when dealing with smaller records

Description:

1) It could be a problem with the operating system’s TCP/IP stack. We’ve had this problem before with SCO Unix and Linux. One of the older SCO versions had a problem where the transmission time would be really slow if the packet was over 1024 bytes. Because of this, whenever you read smaller records from the DS, it would go fast. When the records were larger, causing the total packet size with all of the TCP/IP and framing info to exceed 1024 bytes, the transmission would be really slow. 

2) It could be that the TCP/IP configuration for the UNIX box has the nagle algorithm enabled by default. If copying large files from the OS goes quickly, it could be that the SMB protocol (or whatever is used when copying the file) disables it. The nagle algorithm is used to reduce the 
number of small packets sent by a host (i.e., by buffering unacknowledged send data until a full-size packet can be sent). This means that it could be buffering the packets, and the usual wait time is in the neighborhood of 50ms. This can add substantially to the transmission time.

Last Modified: 11/17/1998 Product: PRO/5 Data Server Operating System: UNIX

BASIS structures five components of their technology into the BBx Generations.