This message appeared on the error logs this morning:

 “Backup of MYDB failed. ACO5422E Received the following from the MS SQL server: An exception occurred while executing a Transact-SQL statement or batch. BackupVirtualDeviceSet::SetBufferParms: Request large buffers failure on backup device 'TDPSQL-00002B50-0000'. Operating system error 0x8007000e(error not found). BACKUP DATABASE is terminating abnormally”

 The MemToLeave area had filled up and caused heavy fragmentation. SQL server backup allocates a defined buffer count in a defined size . i.e BUFFERCOUNT and MAXTRANSFERSIZE. Default MAXTRANSFERSIZE is 1M. Default BUFFERCOUNT is defined by SQL server based on database volumes and backup devices. On a 32 bit system – as SQL Server is also allocating VAS for various other functions, such as OLE Automation – it is possible there aren’t x number of contigious 1M chunks of VAS.   (Not to be confused with physical memory).

 As the first allocation fails – the backup retries – with smaller MAXTRANSFERSIZE sizes – until there is no VAS left – which will cause the backup to fail.

 One well documented way to deal with this issue is to increase the MemToLeave area via the –g switch such as –g512 .  Another approach is to decrease the BUFFECRCOUNT to ? and the MAXTRANSFERSIZE to ? – this will have a negative impact on backup performance but needs to be weighed up against failed backups. The BUFFERCOUNT and MAXTRANSFERSIZE options can be defined on the BACKUP command

The BUUFERCOUNT option details the total number of I\O buffers in the backup

The MAXTRANSFERSIZE option details the largest unit of transfer in bytes – for the transfer between the SQL Server and the backup plaftform. Values range from 65536 bytes (64 KB) – 4194304 bytes (4 MB)

 Total buffer space used is : buffercount * maxtransfersize

Ref:Jack Vamvas (http://www.sqlserver-dba.com)