Vol. 5 No. 8
September 11, 1997

1.0     Digital/Oracle Strategic Partnership2.0     Developer 2000, Forms Memory Leak Patch1.0	Digital/Oracle Strategic Partnership	On Mon, 07 Jul 1997 10:08:51 -0400 James Gibbons  wrote:		We often speak of our "alliances" with other technology 		firms:  Microsoft, MCI, etc., but rarely hear it from the 		perspective of our Alliance for Enterprise Computing partners.		Copied here is a letter received from Richard French of 		Oracle Corporation that details, from Oracle's perspective, 		the value they place in Digital's superior platform 		technologies. 		Regards,		James T. Gibbons		Digital Equipment Corporation		gibbonsj@mail.dec.com		(716) 833-0216 - ph		(716) 833-1539 - fax 		From: 	 Richard French		Sent: 	 Friday, June 20, 1997 6:12 PM		Subject: Digital / Oracle Strategic partnership		With reference to our recent conversation, I'd like to assure 	     	you that Digital is clearly a key partner of Oracle.  Our 		Digital relationship is valued not only for its historical 		significance (1979 1st Oracle on PDP-11) but more so for 		its broad support of Oracle's complete product suite.  		Not only are Oracle's industry leading databases, tools 		and application products supported on all Digital platforms 		(x86, VAX and Alpha) but this product family is		further optimized for performance against the OVMS, UNIX, 		and NT operating environments.  This optimization is 		evidenced by our support of 64-bit and TruClusters on 		OVMS, UNIX, and soon NT. Our joint development strategy 		allows both Digital and Oracle to present to our shared 		clients a best-in-class offering tailored to their corporate 		requirements, all without compromise. 		Two other points of evidence of Digital's importance to 		Oracle is Digital's commitment to the Network Computing                 and Digital's commitment to jointly market Oracle's NT 		products to the enterprise. 		With respect to Network Computing, no one at Oracle would 		argue the fact that Digital is the most aligned of our 		"traditional" hardware partners behind our Network Computing 		strategy.  For the last twelve months Digital has demonstrated		its commitment to this new market by promoting the powerful 		StrongARM processor, the NC servers, and committing to deliver		NC services through your highly acclaimed SI organization.  		On June 24th, Oracle is launching Network Computing and		Oracle8.  Oracle has offered Digital the visibility that only 		global partners can expect.  This includes: -- All video                 content for the demos are provided by Digital -- several                 4100s spread around Radio City Music Hall -- Onstage presence                 of Digital's NC and NC servers -- Digital's inclusion in the                 press briefings (only 4 partners invited) --included) -- The                 only web site that showcases Oracle's Network Computing                 Architecture is powered by Digital -- As many NC's as 		Digital can produce will be showcased. 		Everybody at Oracle is impressed by Digital's commitment to 		support our efforts on June 24th.  This type of support is 		guaranteed to further increase Digital's high visibility                 within Oracle. 		Furthermore, Oracle is heavily investing in Digital's NT                 marketing programs.  Oracle has sponsored this year's NT                 symposium which was extremely well received by our joint                 customers.  Going forward, Oracle will be co-sponsoring                 with Digital, a Corporate Leaders Forum around the world.                  At Oracle, we firmly believe that Digital and Oracle's 		shared expertise in key growth segments, including: NT                 technologies such as 64-bit and clusters are unique to                 the Enterprise NT market. 		Finally, from a technology standpoint Digital truly                 stands out among our global partners.  Evidence of                 Oracle's commitment to Digital is the sheer size of our                 Digital Product Division which is the largest porting                 organization at Oracle.  In order to maintain parity                 between our key hardware partners, we created the                 Enterprise Platforms Division comprising the Compaq,		Digital, HP and Sun Product lines.  This has resulted                 in the release of products on these platforms within                 the same week.  Oracle8 beta as well as Oracle8                 production were released on the same day on Digital                 UNIX as other key platforms.  This immediate                 availability comes without any technology compromise.                  From day one Oracle8 has been engineered to support                 Alpha's 64-bit and TruCluster technology.  Other evidence                 of our immediate support of the Digital platform is                 the rapid delivery of Oracle Applications 10.7. 		I sincerely hope this clarifies Digital's status with                 Oracle and as always, thanks for your continued support.		Sincerely,		Richard French		Enterprise Platforms Division Oracle Corporation	        500 Oracle Parkway, MS 1OP678,		Redwood Shores, CA 94065 	Submitted by Michael Notarius2.0	Developer 2000, Forms Memory Leak Patch	There is a patch for Alpha and VAX to take care of a memory leak 	problem on Developer 2000 1.0, Forms that was brought to my 	attention by Kathy Paranya (THANK YOU KATHY!).  She heard about it at a	conference (see more details in the readme file included with the saves        etc).	Oracle came up with a patch that takes care of some of the problems the        rest of which can be tuned (please see e-mail below from Oracle).	Once you breakout the saveset, you will have access to the readme.txt         files for your particular platform.	It is currently listed in the "THIRD PARTY" category as: 	 	29  ORAMEMLEAK   1071   6/97     Memory Leak Patch	One of the related issues with this patch has to due with tuning memory	management.  Joe Weber has retrieved the following from Digital with         regards to that (there is more related information within the readme         file in the savset).	------------------------ From Digital Support -----------------------                                                                              	[OpenVMS] How To Set MMG_CTLFLAGS To Enable Proactive Memory Management        Any party granted access to the following copyrighted information        (protected under Federal Copyright Laws), pursuant to a duly executed        Digital Service Agreement may, under the terms of such agreement copy        all or selected portions of this information for internal use and        distribution only. No other copying or distribution for any other        purpose is authorized.	Copyright (c) Digital Equipment Corporation 1991, 1995. All rights         reserved.	OP/SYS:     OpenVMS VAX, Versions 5.4-3 and above           								        OpenVMS ALPHA, All Versions        NEW_5.4-3	COMPONENT:  Memory Management        SOURCE:     Digital Customer Support Center	QUESTION:	How is the value for the SYSGEN parameter MMG_CTLFLAGS interpreted?        ANSWER:	The SYSGEN parameter MMG_CTLFLAGS controls the proactive memory	management and memory reclamation mechanisms introduced in         V5.4-3 of OpenVMS VAX.  The value of this parameter is         interpreted as a bit-encoded field which means each binary         bit in the number turns on or off a particular feature.	There are currently only two bits defined for this parameter:	Bit 0, which has a value of 1, enables the "ticker" function.        When set, the ticker becomes active when the size of the free         list drops below the value of 2 x FREEGOAL.  When active, the ticker         looks for processes that are entering a voluntary wait state (LEF, HIB,        or CEF) but have not consumed sufficient CPU time when compared to         their wait time.  These processes are forced to relinquish a portion         of their working set list entries as long as they are not real-time 	processes and do not have any outstanding direct I/O.	Bit 1, which has a value of 2, enables the "troller" function.        When set, the troller becomes active when the size of the free        list drops below the value of FREEGOAL.  When active, the 		troller reclaims memory by outswapping processes that have been        idle for longer than LONGWAIT seconds.  If the free list drops        below FREELIM, processes with more the one half LONGWAIT seconds of 	idle time will become candidates.	Bits 2 through 7 are reserved for future use.  To enable both         functions, set MMG_CTLFLAGS to a value of 3.	In OpenVMS AXP and versions of OpenVMS VAX previous to V6.0,         AUTOGEN will, by default, set this parameter to the value 255.        This value is the equivalent of setting bits 0 through 7 to         one.  So for the current implementation of this parameter, a         value of 255 is equivalent to 3.  In OpenVMS VAX V6.0, the         default value has been set to 3 and AUTOGEN will set the         parameter to its default value.	REFERENCES:	"VMS Version 5.4-3 Release Notes", Sept 1991, (AA-PHUFA-TE),         Page 5-1 "OpenVMS System Management Utilities Reference Manual:        M-Z", May 1993,  (AA-PV5QA-TK), Page C-20	[OpenVMS] Details On How Proactive Memory Reclamation Works        Any party granted access to the following copyrighted information         (protected under Federal Copyright Laws), pursuant to a duly executed         Digital Service Agreement may, under the terms of such agreement copy         all or selected portions of this information for internal use and         distribution only. No other copying or distribution for any other         purpose is authorized. 	Copyright (c) Digital Equipment Corporation 1992, 1995.  All rights         reserved.	PRODUCT:    OpenVMS VAX Versions 5.4-3 and above                    OpenVMS AXP All Versions                    NEW_5.4-3	            COMPONENT:  Memory Management	            SOURCE:     Digital Customer Support Center		    SUBJECT:	This article provides a detailed discussion of the proactive         memory reclamation mechanisms first introduced in OpenVMS VAX         V5.4-3. 	DESCRIPTION:	Prior to V5.4-3, VMS only began to reclaim memory if there was        a critical shortage.  If the size of the free page list got         below the SYSGEN parameter FREELIM, the SWAPPER would begin         first level trimming (i.e. trimming processes back from         WSEXTENT to WSQUOTA).  If this did not regain enough memory,         then it would go on to second level trimming (back to         SWPOUTPGCNT) and eventually begin swapping processes.  	Since processes that were swapped now had working set sizes         that were very small (SWPOUTPGCNT defaults to 288 pages), once        they were brought back into the system they would have to page        fault enough to justify growing again to their original size.         Since this was done when the system was already very short on         memory, performance quickly deteriorated and rarely recovered.	With V5.4-3, the above algorithm is still true.  If we get         down to FREELIM, the SWAPPER will still do first and second         level trimming and then its normal swapping.  However, a 	new proactive memory reclamation was introduced that activates        sooner with the intention of avoiding very low free memory         conditions.	There are two mechanisms to proactive memory reclamation.  One        is referred to as the "ticker".  The ticker becomes active         when the free list is lower than two times FREEGOAL and trims         processes that are mostly dormant but awaken periodically.	The other mechanism is referred to as the "troller".  It         becomes active when the free list drops below FREEGOAL and         reclaims memory by outswapping dormant processes.	TICKER OPERATION:	The TICKER routine is enabled by setting bit 0 of the SYSGEN         parameter MMG_CTLFLAGS.  It is flagged as "active" when the         system sees that the free list is less than FREEGOAL*2.        When any process is placed into HIB, LEF or CEF the system         will check to see if the TICKER is flagged as active.  If the         TICKER is flagged as active, and the process is a candidate         for TICKER trimming, the TICKER will be called.  A TICKER         candidate is a process that has accumulated at least 30 seconds        of idle time, is not a real-time process, and has no outstanding         direct I/O.	The TICKER routine itself then checks to see if the process         has used less than 1% of it's time in the CPU (vs. waiting).          If so, 25% of the pages are removed from the process's 			working set.	TROLLER OPERATION:	The TROLLER is enabled by setting bit 1 of the SYSGEN parameter         MMG_CTLFLAGS.  When the system detects that it has fewer than         FREEGOAL pages in the free list, the SWAPPER will be activated to         execute the TROLLER routine and begin outswapping process that have        been waiting for longer than LONGWAIT seconds in HIB or LEF state.        The SWAPPER will first remove pages from the process' working set         (down to SWPOUTPGCNT) without reducing the size of the actual         working set list.  This allows the process to quickly pagefault         needed pages back into its working set when it is swapped back into        the system.        --------------- End Note from Digital Support ----------------------	Please let me know if you have any questions / concerns.	Submitted by Mike Notarius                           Editor:  Barbara Boquard                                    Boquarba@itec.suny.edu                 ITEC's General Office     716/878-4832                 Director's Office         716/878-4832                 Operations                716/878-5122                 After hours               716/878-6642                 FAX                       716/878-4235                 web page                  http://www.itec.suny.eduITEC HOME PAGE