I think some info that I will be posting below would be relevant to this topic.
[1] How large can and should the lock table be configured?
As of Release 4.0 the default size for the lock table is 4 MB. For medium-sized systems this value is absolutely sufficient. As of 4.6, it appears that a size of 10-20 MB is required for some background jobs, and a size of 32-200 MB may be required for large systems, although this is the exception.
As a lock table that is too small causes transaction terminations, but resources for the lock table are relatively low, a size of 20 MB should be specified initially. no further changes are generally required to the layout of the shared memories with this size (except for AIX 32 bit). You can enter this setting using the profile parameter:
enque/table_size = 20000
You can monitor the lock table in transaction SM12 via the menu option "Extras -> Statistics".
The lock table is a shared memory, not a database table.
[2] What does the expired queue time in the syslog mean for locks?
Locks can usually only be set in the R/3 system if they
are available. If an object is already locked, requests to lock it are
refused and the error message " ... locked by user ..." is issued. This
prevents the dead lock situations familiar from databases. However, a job
should not terminate in background processing if a lock happens to be unavailable.
In such cases, the locks are requested with the addition "and wait". The
queue time incurred is then logged in the syslog. The maximum queue time
is set using the profile parameter enque/delay_max.
[3] How does communication take place in lock management?
This question is especially important for troubleshooting since it indicates the characters at which errors can occur. In the Central Instance, all work
processes can access the lock table directly. Therefore, the ENQ work process is not needed, and so no communication occurs.
An application server sets and deletes locks via the Message
Server. There is therefore communication from work process-> dispatcher->server
-> dispatcher -> ENQ work process. The lock table is read by RFC, that
is, work process-> dialog process -> gateway-> gateway-> dialog process
-> dialog process. Here again, the ENQ process is not required, although
two dialog processes are needed for RFC communication. Central instances
with few dialog processes can be overloaded quickly in this way. The same
also applies to pure background or update server processing. The number
of dialog work processes must therefore be at least as high as the number
of remaining processes.
[4] What are "black" and "blue" locks?
Black locks are normal transaction locks. Blue locks are inherited by the update system and deleted with the corresponding update request. Blue locks are also saved in the file system and restored when the system is restarted.
