We have been doing some significant testing of the TDP for SAP in a Solaris 8 environment. Backup performance with just 2 tape drives and multiplexing of 3 has provided backup speeds of 200GB per hour (55MB/sec). Very successful.
The rub comes during the restore. What we have been able to determine is that the tape drives run so fast on the restore they apparently overrun the TCP/IP buffers used to transfer data from the tapes to the multiple threads necessary to support the restore. The restore fails every time unless the number of drives and sessions are set to 1 during the restore. The configuration looks like this: 2 3590-E1A Fibre Channel Drives (SAN Connected) 2 HBAs in the Solaris E4500 (6 CPUs, 2 Pathways to the ESS Disk Subsystem (Shark), SAN Connected SAN Client 4.2.1.9 (Solaris) TSM Server 4.2.1.9 (AIX) In a mainframe world we have such things as VPACING and numbers of buffers. Does UNIX have anything like this for TCP/IP to prevent the system from blowing out memory? Has anyone else seen this kind of problem? Unfortunately, shared memory is not supported on the Solaris. Tivoli Support acknowledges there is an issue with multiplexing when doing restores. We are working with them, but it would be nice if someone can give some experiences and how they resolved this issue. Paul D. Seay, Jr. Technical Specialist Naptheon, INC 757-688-8180
