Conference makrel::msg_solutions

684.0. "MCM log file too fragmented !" by BACHUS::COLLART (Li p'ti fouineu - Dtn 856-8796) Thu Feb 20 1997 09:00

			Hello,



	How can we control what is logged in MCM$TRACE.LOG file for
	MCM 1.0-001 ?

	We are trying also to push customer to upgrade to 1.1 but he
	doesn't want to go through this quite complicated upgrade for
	them.

	In the manual of MCM 1.1, I found a description how to define
	which level of message is logged by MCM:

		define a global symbol mcm$cvt_trace:==n in LOGIN.COM

	Does it work for v1.0 ? Does it exist other usefull symbols ?

	The problem behind this question is the MCM disk fragmentation
	caused by a big MCM$TRACE.LOG file that reach more than 600 extents
	dayly (for +/- 3500 blocks size). This log file badly impact the
	system performance.

	I must say that the disk used is a RZ26 (1 GB disk) defined with a
	cluster size of 3. The SYSGEN parameter RMS_EXTEND_SIZE is 0.
	The disk is dayly defragmented (with all applications stopped).

	Is there a way to pre-allocate space for MCM$TRACE.LOG to avoid so
	many extents creation (using a FDL file for instance) ?

	The final goal is to find a compromise between what needs to be logged
	and acceptable performance (the MCM disk is also used by Message Router
	and MailWorks server). The average ammount of dayly messages is about
	20.000 (inbound + outbound) per cluster node (2 nodes cluster)
	including about 10.000 going through Message Router (and MCM).

					Thanks for advise

					Eric Collart
					MCS Brussels

    Note 317 has info on the various un-documented V1.1 features. MCM V1.0 
    is too far back for me to remember the particulars as to what was or
    wasn't available (most of the symbols WEREN'T in V1.0). With V1.1 you 
    can also do stuff like define logicals in the login.com to redirect 
    various bits around (for example to distribute I/O across spindles). 
    Running MCM on the same disk as MR is a killer, since fetching/posting 
    messages are in effect file copies (ie lots of seeks if MR$NBS and 
    MCM$MSG areas are on same spindle).
    
    The V1.1 IVP procedure uses LOT's of 'tricks' (eg defining a logical 
    to redirect MCM$TRACE.LOG to the terminal, using search lists, etc).
    Simply examining MCM$FE.COM/MCM$BE.COM/MCM$CVT.COM/MCM$START.COM will 
    reveal a lot about how things work and the hooks that exist. 
    
    Make sure when they upgrade that they are using the V1.1-002 kit 
    (July 95) as it includes a number of important bug-fixes.
    
    Dave

			Hello Dave,



	Thanks for the info, I will check note 317 (I missed it)

					Eric

    BTW, MCM$START.COM is where MCM$TRACE.LOG creation/roll-over occurs.
    
    Dave