[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference vaxaxp::vmsnotes

Title:	VAX and Alpha VMS
Notice:	This is a new VMSnotes, please read note 2.1
Moderator:	VAXAXP::BERNARDO

Created:	Wed Jan 22 1997
Last Modified:	Fri Jun 06 1997
Last Successful Update:	Fri Jun 06 1997
Number of topics:	703
Total number of notes:	3722

567.0. "Streamlined vs. Full-checking Performance" by CSC32::D_BROWN (Dave Brown CSC-VSG/INTDRV) Tue May 06 1997 11:59

    
    
    Is there any information available that I can pass on to a customer to
    satisfy his request to find the performance differences between the default
    streamlined spinlock routines and the full-checking spinlock routines?
    While the I&DS manuals have have descriptions of the differences, they
    do not speak to timing differences.
    
    So the question is how much faster are the streamlined reoutines over
    the full-checking routines? Naturally the customer will have to
    understand that his overall system performance differences will depend
    also on his spinlocking activity and perhaps other factors too.
    
    This particular customer is asking about OpenVMS VAX V6.1.
    
    Thank you, 
    
    Dave

T.R	Title	User	Personal Name	Date	Lines
567.1	Look at [SYS]SPINLOCK.MAR variants...	XDELTA::HOFFMAN	Steve, OpenVMS Engineering	`Tue May 06 1997 12:27`	16
	There have been no performance studies in this area that I am aware of, nor would I want to provide a customer any specific performance-related expectations (beyond "full-check will be slower than streamlined" :-), as this area is subject to change without notice. (We do not normally benchmark the OpenVMS debugging support code...) If the customer is interested, then they will want consider performing some performance testing with the local system(s), and determine what the (OpenVMS-version-specific) performance differences are. Given your customer has looked at the IDSM and is familiar with the general differences, the other resource is the source listings CD-ROM, which will show the instruction path differences among the variants, depending on how the [SYS]SPINLOCK.MAR module was compiled.
567.2	difference isn't normally relevant	GIDDAY::GILLINGS	a crucible of informative mistakes	`Tue May 06 1997 20:07`	14
	Dave, Full checking code is a diagnostic tool used to isolate problems. Normally it should be turned off. If you're interested in maximising performance you should certainly turn it off. The difference for a particular workload isn't really that relevant, since it will only be used while debugging. Any differences will also vary significantly depending on workload, so there is very little point in trying to give a "general" quantification. If your customer is really interested, I'd suggest they perform their own measurements. On the other hand, perhaps if we knew why they were asking the question in the first place, someone could provide a useful comment. John Gillings, Sydney CSC
567.3		AUSS::GARSON	DECcharity Program Office	`Tue May 06 1997 23:36`	6
	re .* I can imagine a scenario where the customer reports a problem but Digital can't reproduce it in the "lab" and so we direct the customer to turn on checking. In that case the customer might legitimately ask what performance impact that will have before agreeing.
567.4		UTURBO::utojvdbu1.uto.dec.com::JurVanDerBurg	Change mode to Panic!	`Wed May 07 1997 03:31`	5
	It also depends strongly on the cpu speed and the cpu load. On the more modern and faster processors you will hardly notice any performance degradation under normal circumstances. Jur.
567.5	Nobody noticed here	OPCO::TSG_PLL	Live and Sweaty	`Wed May 07 1997 04:08`	8
	While attempting to find out why so much time was being consumed by MPSYNC the site I am at turned on the non-streamlined SPINLOCK code. No user seemed to even notice. The interrupt stack plus MPSYNC time was around 40% of one CPU on a 4 CPU VAX 7700 with the streamlined code and did not change appreciably with the debug version in use. FWIW Paul