1 INFO-VAX	Mon, 03 Jun 2002	Volume 2002 : Issue 305       Contents:* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity?* Re: $WAKE() lost during high AST activity? Re: Adding drives to SAN/HSG80 ANN: SYSLOGCLIENT 1 Re: Assembly generation with OpenVMS's C compiler 1 Re: Assembly generation with OpenVMS's C compiler - Re: has anyone ported this opensource to VMS? - Re: has anyone ported this opensource to VMS? 3 Re: IBM wins supercomputer deal, DELL wants in too.  Install /share question ) Intel to start to advertise even more :-( - Re: Intel to start to advertise even more :-( - Re: Intel to start to advertise even more :-(   Re: OpenVMS FAQ due next week... Re: Shadow sets efficiency  F ----------------------------------------------------------------------  # Date: Sun, 02 Jun 2002 18:31:15 GMT ' From: Jeffrey Chimene <jeff@nospam.net> 3 Subject: Re: $WAKE() lost during high AST activity? * Message-ID: <3CFA636A.BA86310A@nospam.net>   Jason O'Donnell wrote:  e > yehavi@vms.huji.ac.il (Yehavi Bourvine (58-4279)) wrote in message news:<2002Jun2.180239@hujicc>... 
 > > Hello, > > Q > >   I have a program in DEC-C which has the following logic: It has a main loop Q > > which does a few instructions, then sleeps for a while, do a few instructions R > > sleeps and so on. The sleep is done by calling $SETIMR followed by $HIBER (and > > the $SETIMR calls $WAKE).  > > P > >   While the program's main loop is active there is an extensive I/O done viaS > > AST's (terminal I/O). I noticed that during high AST activity the $WAKE is lost 9 > > (i.e. it is called, but the $HIBER does not wake up).  > > A > >   Anyone noticed such behaviour and knows how to overcome it?  > > 9 > >                                     Thanks! __Yehavi:  > @ > This doesn't answer your question, but we do a similar thing a( > different way.  It seems to work fine. > E > We have the timr routine set an event flag and resubmit itself.  So ( > instead of using HIB we use sys$wflor.  e Exactly. I don't see why the $hiber call is even necessary. If all the alarm does is $wake  - why not ] use sleep()? Are there other events that call $wake? Do you need sub-second timer resolution?   d Race conditions like this are easy to build when using AST-driven programs. I assume that "extensiveb I/O done via AST's" does *not* mean you're actually doing I/O in the AST routine. Perhaps you meanc that the AST also issues a $wake as part of the I/O completion logic. In that case, I don't believe a you're not seeing the $wake. My guess is that it's getting buried by all time spent at AST level, b racing with the timer reactivation. Remember, that $wake signalling occurs in User mode, just like your AST routines.  c I'd consider using a work list and queue maniplutlation routines for these kinds of programs (about _ this program I admit to knowing very little). As an alternative, you need a different signaling [ mechanism that won't compete with AST delivery; event flags are such a signaling technique.    -- microsoft free by 2003   ------------------------------   Date: 2 Jun 2002 23:10:34 GMT 7 From: yehavi@vms.huji.ac.il (Yehavi Bourvine (58-4279)) 3 Subject: Re: $WAKE() lost during high AST activity? $ Message-ID: <2002Jun2.231034@hujicc>   Hello,    Forst - thanks for answering me!  N > Exactly. I don't see why the $hiber call is even necessary. If all the alarm > does is $wake  - why notM > use sleep()? Are there other events that call $wake? Do you need sub-second  > timer resolution?   N I tried sleep() but it failed with the same symptoms but much more frequently;D it simply did not return... (I placed a print before and after it).\  M > Race conditions like this are easy to build when using AST-driven programs. G > I assume that "extensive > I/O done via AST's" does *not* mean you're ) > actually doing I/O in the AST routine.    N The AST is fired a s result of $qio() read and then it calls a $qiow() to send" to another channel (a faster one).  7                                       Thanks, __Yehavi:    ------------------------------  % Date: Sun, 02 Jun 2002 15:46:23 -0400 % From: JF Mezei <jfmezei@videotron.ca> 3 Subject: Re: $WAKE() lost during high AST activity? , Message-ID: <3CFA760D.53349C35@videotron.ca>  L It would seem to me that you issue a $SETIMR whose AST gets delivered beforeM the $HIBER executes, so when the AST does a $WAKE, the process isn't sleeping G yet so $WAKE does nothing, and a little later, $HIBER finally executes.   N Another possibility is that somewhere else in your program you (or some systemI routine) may be issuing a $CANTIM which also affects your time requests.  < (note that this is not very likely, but still a possibility)     Two suggestions:  I use a random number to give your $SETIMR request a unique ID. (this ID is J passed as parameter to the AST). This will make it extremely unlikely that1 $CANTIM used elsewhere would cancel your request.   J The other suggestion is to put the $SETIMR inside the AST, right after theH $WAKE. This way, the $WAKE is garanteed to execute every few seconds (or whatever $SETIMR specified).  - You would have code in your mainline such as     wake_flag = 1 ;  status = SYS$HIBER() ; wake_flag = 0 ;   L And in your AST, you would execute $WAKE only if variable wake_flag = 1, but& always issue another $SETIMR request.   N If you issue a $WAKE before a $HIBER, the $HIBER will do nothing if I rememberJ correctly. So the goal of the above wake_flag mechanism is to only issue aJ $WAKE once you are pretty sure than your program has gotten to the $HIBER.   ------------------------------   Date: 2 Jun 2002 12:50:48 -0700 / From: chris@applied-synergy.com (Chris Scheers) 3 Subject: Re: $WAKE() lost during high AST activity? = Message-ID: <754a27c1.0206021150.7a31ec05@posting.google.com>   c yehavi@vms.huji.ac.il (Yehavi Bourvine (58-4279)) wrote in message news:<2002Jun2.180239@hujicc>...  > Hello, > O >   I have a program in DEC-C which has the following logic: It has a main loop O > which does a few instructions, then sleeps for a while, do a few instructions Q > sleeps and so on. The sleep is done by calling $SETIMR followed by $HIBER (and   > the $SETIMR calls $WAKE).  > N >   While the program's main loop is active there is an extensive I/O done viaQ > AST's (terminal I/O). I noticed that during high AST activity the $WAKE is lost 7 > (i.e. it is called, but the $HIBER does not wake up).  > ? >   Anyone noticed such behaviour and knows how to overcome it?  > 7 >                                     Thanks! __Yehavi:     D It is probable that you are getting AST activity between the $SETIMRC and the $HIBER and that the $SETIMR is firing during the AST, i.e.,  before the $HIBER.  D If something in the AST activity consumes the $WAKE, then the $HIBER might not wake up.  E You may be better off using $SCHDWK instead of $SETIMR.  Then, if one 9 of the wakeups misses, the next one will wake the $HIBER.    ------------------------------  % Date: Sun, 02 Jun 2002 16:59:58 -0400 % From: JF Mezei <jfmezei@videotron.ca> 3 Subject: Re: $WAKE() lost during high AST activity? , Message-ID: <3CFA8749.8F8CEA18@videotron.ca>   Chris Scheers wrote:F > If something in the AST activity consumes the $WAKE, then the $HIBER > might not wake up.  # Something I am not sure of anymore.   J I know that if I do a SET PROC/RESUME followed by SET PROC/SUSP, the /SUSP; won't do much because there is already a "RESUME" flag set.   - Does the $HIBER and $WAKE work the same way ?   M If a $WAKE executes before a $HIBER, would the $HIBER simply skip since there  is a wakeup flag already set ?  J I agree with your suggestion of the $SCHDWK which can be set to repeatedly+ issue the $WAKE for you at fixed intervals.    ------------------------------  # Date: Sun, 02 Jun 2002 21:50:32 GMT ' From: Jeffrey Chimene <jeff@nospam.net> 3 Subject: Re: $WAKE() lost during high AST activity? * Message-ID: <3CFA9252.B97D0896@nospam.net>  " "Yehavi Bourvine (58-4279)" wrote:  P > I tried sleep() but it failed with the same symptoms but much more frequently; > it simply did not return...   ( Technically, this is called a "clue" ;=}  K There's no way that sleep() is not returning. The problem lies in your app.  *somewhere*.   >  > O > > Race conditions like this are easy to build when using AST-driven programs. I > > I assume that "extensive > I/O done via AST's" does *not* mean you're * > > actually doing I/O in the AST routine. > P > The AST is fired a s result of $qio() read and then it calls a $qiow() to send$ > to another channel (a faster one).  O There will be those who disagree with this assesment: remove the $qioW from the Q AST routine. The AST should be doing nothing more than indicating to the mainline Q that the $qio has finished. In your implementation, the AST routine cannot return O until the $qioW completes; which design is not what you want in an event-driven Q program. The AST routine is now a *major* bottleneck for the rest of your progam; H the $qio completion ASTs are queuing up waiting for the $qioW to finish.  L Please bear in mind that I only know as much as about this program as you'veO revealed to the list. However, two decades work w/ OpenVMS suggests that I/O in K an AST routine is To Be Avoided At All Costs. At this time. I don't see how M moving the $qioW outside the AST will fix the sleep() issue. However, I would Q *highly recommend* you consider restructuring. As a side-effect you might uncover ! the cause of the sleep() problem.    -- microsoft free by 2003   ------------------------------  % Date: Sun, 02 Jun 2002 19:04:04 -0400 % From: JF Mezei <jfmezei@videotron.ca> 3 Subject: Re: $WAKE() lost during high AST activity? , Message-ID: <3CFAA45A.39B71CB9@videotron.ca>   Jeffrey Chimene wrote:Q > There will be those who disagree with this assesment: remove the $qioW from the S > AST routine. The AST should be doing nothing more than indicating to the mainline  > that the $qio has finished.   @ Note that QIO doesn't actually do the IO, it queues the request.  N For "terminal" type of programs that take data from one port and shove it downK the other port, this seems to be a perfectly valid paradigm. When a READ IO I completes, an AST is fired which does a WRITE of the received data on the ) othert port and then queues another READ.   N Yes, there is the issue of what happens when the second channel becomes XOFFEDJ and the QIOW will hang because the buffers are full, thereby stopping yourL program.  But when you think about it, it isn't so bad since when the outputJ channel is full/blocked, you generally also want to stop the input channelM from sending further data until the situation is resolved, and while you wait I for the WRITE to complete, the input typeahead buffer for the channel you L would be reading from will fill up and an XOFF will be sent to it to tell it to stop for a while.  K Now, one should really issue a QIO instead of QIOW for the write operations L with an event flag set to indicate completion, and before you issue the nextL QIO (write) you make sure the event flag from the previous QIO was set. ThisK way, if the previous QIO (write) hasn't completed yet, your program doesn't L freeze and you can take whatever actions you deem necessary to deal with the* case of a output channel which is blocked.  K Note that if the output channel is slower than the input channel, then QIOW I (write) is fine since the operating system will throttle your application J automatically and in doing so, the typeahead buffer for the faster channelL will also auto-throttle the remote data generating system since you won't beQ issuing QIO (read) to empty that typeahead buffer until the writes are completed.     Q > revealed to the list. However, two decades work w/ OpenVMS suggests that I/O in / > an AST routine is To Be Avoided At All Costs.   / What is so wrong with having a QIO in an AST ?    J To me, the esthetics of having a "read loop" that is AST driven (eg: afterM read complete, you issue a new QIO from within the AST which will call itself M when that IO completes) makes for extremely simpler programs, especially when L you have DECNET connections that can handle multiple connections at the sameJ time (server application). The AST uses an argument to determine who it isJ talking to, stores the received data, dispatches the work, and then queues then read on that channel.  I A single routine can then handle multiple clients and that makes for much  simpler coding.    ------------------------------  $ Date: Sun, 2 Jun 2002 19:23:50 -0400  From: John Santos <JOHN@egh.com>3 Subject: Re: $WAKE() lost during high AST activity? 5 Message-ID: <1020602185419.1319B-100000@Ives.egh.com>   # On Sun, 2 Jun 2002, JF Mezei wrote:    > Chris Scheers wrote:H > > If something in the AST activity consumes the $WAKE, then the $HIBER > > might not wake up. > % > Something I am not sure of anymore.  > L > I know that if I do a SET PROC/RESUME followed by SET PROC/SUSP, the /SUSP= > won't do much because there is already a "RESUME" flag set.  > / > Does the $HIBER and $WAKE work the same way ?  > O > If a $WAKE executes before a $HIBER, would the $HIBER simply skip since there   > is a wakeup flag already set ?  @ If you $HIBER after a $WAKE, the $HIBER will return immediately.F So if your AST's are causing the timer to expire and execute the $WAKE@ before you actually hibernate, you will see your program wake upI instantly.  I don't think this is the symptom you (OP) were seeing, IIRC. @ You were instead seeing the process hibernate and never wake up.F (Have you verified this with $SHOW SYSTEM or $SHOW PROCESS showing the1 process in HIB, rather than LEF or RW??? states?)   L > I agree with your suggestion of the $SCHDWK which can be set to repeatedly- > issue the $WAKE for you at fixed intervals.   A While this should work (cause an eventual wakeup even if you miss B one), it might be masking a more fundamental problem.  Although ifB you understand why wakeups are being missed, maybe you will decide that this is fine.  D Are you sure nothing is calling $CANWAK or $CANTIM behind your back?A Maybe some language RTL is doing this (I don't know what the C or  BASIC sleep() routines do.)   D Are you sure you aren't running out of TQELM quota, and causing your $SETIMR to fail?  F Also, I strongly agree with Jeffrey Chimene's other followup regardingB eliminating $qioW from your AST routines.  If you do use $qio (not> $qioW), make sure all your buffers, IOSBs, etc. are statically> allocated, or at least persistent across the life of the $qio.   --   John Santos  Evans Griffiths & Hart, Inc. 781-861-0670 ext 539   ------------------------------  # Date: Mon, 03 Jun 2002 04:13:12 GMT * From: "Bill Todd" <billtodd@metrocast.net>3 Subject: Re: $WAKE() lost during high AST activity? B Message-ID: <s1CK8.101135$Kp.10369701@bin7.nnrp.aus1.giganews.com>  2 "JF Mezei" <jfmezei@videotron.ca> wrote in message& news:3CFAA45A.39B71CB9@videotron.ca...   ...   0 > What is so wrong with having a QIO in an AST ?  & Nothing.  A QIOW, on the other hand...   - bill   ------------------------------  # Date: Mon, 03 Jun 2002 03:16:07 GMT  From: danco@pebble.org3 Subject: Re: $WAKE() lost during high AST activity? - Message-ID: <slrnaflol2.i80.danco@pebble.org>   H In article <1020602185419.1319B-100000@Ives.egh.com>, John Santos wrote:  F > Are you sure you aren't running out of TQELM quota, and causing your > $SETIMR to fail?  D Right on.  Be certain that you're checking all your return statuses.H The default TQELM quota is very low, something on the order of 16 or so.   - Dan    ------------------------------  $ Date: Sun, 2 Jun 2002 17:46:45 -0400  From: John Santos <JOHN@egh.com>' Subject: Re: Adding drives to SAN/HSG80 5 Message-ID: <1020602173711.1319B-100000@Ives.egh.com>     On 1 Jun 2002, John Koska wrote:  ^ > John Santos <JOHN@egh.com> wrote in message news:<1020530183040.359A-100000@Ives.egh.com>...* > > On Thu, 30 May 2002, Rick Dyson wrote: > .  > .C > .tF > > I had the same question about adding disks to an HSZ80 a few weeks- > > ago, and someone pointed me a document ath > > g > > http://wwss1pro.compaq.com/support/reference_library/viewdocument.asp?source=OD011115_CW01.xml&dt=3K > >  > > (That probably wrapped.) > .W > .  > . B > > P.S.  Thanks again, whoever the helpful person was who pointed > > me at this.t > E > I think that was me from what checking I did in the archives.  Your H > thanks is welcome.  comp.os.vms has almost always been a helpful place> > for my questions, and I like to return something when I can. >  > John Koska > Matthew Bender & Co., Inc.  8 Yes, John, it was you.  Please accept my belated thanks.  I (I use Pine on VMS for newsgroup postings, and one of it's rough edges isoI you can't save a partially composed message, check your mail, and go backtI in to complete the message.  Actually, this works fine if you are writingnE a mail message, but if you try this in a news post, it forgets it's a-F news post and instead tries to mail it to the newsgroup.  Ugh.  So theH polite thing would have been to go back into my saved messages, find outD who deserved the credit, and thank him (i.e. you) properly, but thisH was too much of a PITA :-)  Someday, when I have nothing to do for a fewI weeks (fat chance), maybe I'll fix this and some other bugs.  But I think 7 the latest VMS version of Pine is years out of date...)    --   John Santosi Evans Griffiths & Hart, Inc. 781-861-0670 ext 539   ------------------------------  # Date: Mon, 03 Jun 2002 02:12:15 GMT ( From: Mark Hemker <hemker@insightbb.com> Subject: ANN: SYSLOGCLIENT8 Message-ID: <i1klfucs5sqeja75o2qd9ecqs34trdtv48@4ax.com>  ; I finally got around to putting up a website and putting myS7 SYSLOGCLIENT on it.  If anyone is interested, it is at:tC http://home.insightbb.com/~hemker/vms.html.  This is a small client B that I wrote since I couldn't find one anywhere else.  There is anC included text file that has a much more detailed description of thes@ program and how to use it.  If you have question, suggestions or. comments, please feel free to send them to me.   Mark Hemker  hemker@insightbb.com   ------------------------------  # Date: Sun, 02 Jun 2002 21:54:20 GMTw- From: "John E. Malmberg" <wb8tyw@qsl.network>e: Subject: Re: Assembly generation with OpenVMS's C compiler* Message-ID: <3CFA9143.7080104@qsl.network>   Steffen Grnneberg wrote:o7 > How would I make the c compiler output assembly code?i  - For the Compaq or DEC C compilers on OpenVMS:r   /LIST/MACHINE_CODE  A > And, how would I compile the generated code afterwords with cc?   H The assembly language output is not directly suitable for recompilation.  C In particular, you will probably need to fix up some of the global '@ symbol definitions, psects, and the start address of the module.  F Also the declaration of storage for string constants were not present  the last time I did this.u  I But with a small amount of editing, you can use the MACRO-32 compiler to iD compile on ALPHA the machine code of from a Compaq C listing on VAX.     -Johno wb8tyw@qsl.network Personal Opinion Only    ------------------------------  # Date: Sun, 02 Jun 2002 22:01:45 GMTeB From: Steffen =?ISO-8859-15?Q?Gr=F8nneberg?= <steffen@illumina.no>: Subject: Re: Assembly generation with OpenVMS's C compiler7 Message-ID: <dBwK8.8497$_15.253534@news4.ulv.nextra.no>.  4 Interesting.. Thanx :-) Ill hack around with it.. :)   John E. Malmberg wrote:i   > Steffen Grnneberg wrote:08 >> How would I make the c compiler output assembly code? > / > For the Compaq or DEC C compilers on OpenVMS:n >  > /LIST/MACHINE_CODE > B >> And, how would I compile the generated code afterwords with cc? > J > The assembly language output is not directly suitable for recompilation. > D > In particular, you will probably need to fix up some of the globalB > symbol definitions, psects, and the start address of the module. > G > Also the declaration of storage for string constants were not presents > the last time I did this.n > J > But with a small amount of editing, you can use the MACRO-32 compiler toF > compile on ALPHA the machine code of from a Compaq C listing on VAX. >  >  > -Johno > wb8tyw@qsl.network > Personal Opinion Only    ------------------------------  $ Date: Mon, 3 Jun 2002 00:16:05 -0400( From: "Mike Foley" <mike-spam@yelof.com>6 Subject: Re: has anyone ported this opensource to VMS?/ Message-ID: <uflrc8namrus24@corp.supernews.com>D       Sue,  @     Wouldn't help that much. I don't think there's alot of solid;     OpenGL support on an Alpha VMS workstation. :) Besides,h:     VMS already has a flight simulator on the Freeware CD.<     It's called FLIGHT. Written by Tom Dahl and John Buehler     of DECwrite fame.l  ?     I have the email John Buehler, if you ever need it. We keept@     in touch. I helped them with hours of beta testing. Probably;     our most fun night was using my VAX 9000-440 in the VMSl<     lab and a couple of high-end VAXstations in my office to     dogfight for many hours.  >     Oh, and also the late nights preparing for DECworld '88 in8     Cannes. I got my 5th missile lock and became the ace@     for the night. Much cursing was heard all over the floor and     in the pubs later. :)e     mikeE                                                              formerlyv STAR::MFOLEY    = "Sue Skonetski" <susan.skonetski@compaq.com> wrote in messagef, news:SbPJ8.30$_R4.388138@news.cpqcorp.net... > http://www.flightgear.orgu >u1 > I am thinking this would be pretty cool on VMS.t >l > sues >a >h >$   ------------------------------  % Date: Mon, 03 Jun 2002 00:59:29 -0400o- From: JF Mezei <jfmezei.spamnot@videotron.ca>o6 Subject: Re: has anyone ported this opensource to VMS?, Message-ID: <3CFAF791.3A17B1E7@videotron.ca>   Mike Foley wrote:EA >     I have the email John Buehler, if you ever need it. We keep0 >     in touch.   H I had setup a world for Montreal, and got into a problem of having "trueL north" coordinates for the geography, but needing the runways and compass to be magnetic north.  M For instance, runway 6R needs to be at 41 tru north, and 60 magnetic north..K If I put the runway at 60, then the planes will hit the mountain :-(. If I:M put it at 41, then the plane's compass is out of wack because you would want 1 to have a heading of 60 on the approach to land.e  M Is there any resource that really explains this dilemma ? I checked the world)L reference, but the explanations do not deal with the difference between real north and magnetic north.    ------------------------------   Date: 2 Jun 2002 23:23:57 -0600i+ From: young_r@encompasserve.org (Rob Young)e< Subject: Re: IBM wins supercomputer deal, DELL wants in too.3 Message-ID: <ROVZcYoEG$C8@eisner.encompasserve.org>   T In article <3CF9B87A.5B6593D7@videotron.ca>, JF Mezei <jfmezei@videotron.ca> writes:N > With the death of Alpha, it makes some interesting developments. Dell on the6 > 8086, IBM on the Power, and IA64 nowhere to be seen. >    	IA64 nowhere to be seen?      	JF in typical fashion.a  5 http://news.com.com/2100-1001-271382.html?legacy=cneti  +  NSF awards $53 million supercomputing bid l     By Stephen Shankland - Staff Writer, CNET News.coma August 9, 2001, 1:50 PM PT    G The National Science Foundation has awarded contracts worth $53 millionuN to build a grid that connects supercomputer clusters across the country into aK single large computing resource called the Distributed Terascale Facility. h  G The supercomputers will be made from Intel's "McKinley" CPU, the second O generation model of the Itanium line, the National Science Foundation said in aoO statement. In addition, Qwest will link the computers with a high-speed network 2 that can transfer data at 40 gigabits per second.     	 =========t  3 http://news.com.com/2100-1001-884297.html?tag=cd_mhr  !  HP to build Linux supercomputer s     By David Becker  Staff Writer, CNET News.com  April 16, 2002, 4:45 PM PT    D Hewlett-Packard announced a high-profile contract Tuesday to build a? Linux-powered supercomputer for the U.S. Department of Energy. cJ The $24.5 million supercomputer will be capable of processing 8.3 trillionN calculations per second (8.3 teraflops), according to HP. That would place theL system among the current top 10 of the world's fastest supercomputers, if it were operating today.   J HP expects to have the supercomputer running at the Department of Energy'sN Pacific Northwest National Laboratory in Richland, Wash., early next year. TheO supercomputer will be used for biological and environmental research, includingtN genomics work, at the laboratory's Environmental Molecular Sciences Laboratory (EMSL).   n [snip]  I The HP supercomputer will replace an outdated IBM system installed at thee laboratory in 1997.n   =====t  8 	You can bet IBM got beat on the RFP for the new system.  B 	Here's an official prognostication.  In 2 to 3 years, Itanium andD 	follow-ons will win the bulk of next generation supercomputer bids.0 	Price and performance wins the day most always.   				Rob    ------------------------------  % Date: Sun, 02 Jun 2002 19:13:44 -0400i% From: JF Mezei <jfmezei@videotron.ca>   Subject: Install /share question, Message-ID: <3CFAA69D.A12FB53C@videotron.ca>  N Trying to ensure that TPU starts up faster, I checked which files are accessed0 (both with set watch as well as show dev/files).  & I started 4 TPU (decwindows) sessions.  M SHOW DEV/FILES reveals that SYS$MESSAGE:DECW$TERMINALMSG.EXE is opened by allt 4 processes.   BUT:  < INSTALL LIST/FULL SYS$MESSAGE:DECW$TERMINALMSG.EXE  reveals: 	Open Hdr Shar    Lnkblg 	Entry access count		= 0! 	Current /Maximum Shared 	= 1 / 0o 	Global section count 		= 2f  J There is also DECW$TRANSPORTMSG.EXE and XLIBMSG.EXE which exhibit the sameI behavour, and the other installed images will show current access to be 4@% (with maximum shared also 4 for now).s  J If the file is opened, how come INSTALL doesn't have an accurate "Current" count ?h   (This is VAX VMS 7.2)l   ------------------------------  % Date: Sun, 02 Jun 2002 17:39:15 -0400S% From: JF Mezei <jfmezei@videotron.ca>i2 Subject: Intel to start to advertise even more :-(, Message-ID: <3CFA907C.2AE7ADEA@videotron.ca>  N In light of discussions on why HP should mention VMS in mainstream advertsing,' I found the following most interesting:n  N The URL was too long on so I am pasting the article whole here. (Yahoo: if youN want extra eyeball's for your advertising, make oit easy for folks to refer to your pages).    ) Intel Kicks Off Biggest-Ever Ad Campaign u  Sun Jun 2, 3:32 PM ET e  G SANTA CLARA, Calif. (Reuters) - Seeking to bolster its brand with largeaL corporations and the  people who run them, Intel Corp. , the world's largestK chipmaker that is best known for its  Pentium line of microprocessors, willa: this week kick off its largest-ever advertising campaign.   A   The Santa Clara, California-based chipmaker, which invented themN microprocessor -- most widely  known as the brains of personal computers -- isN seeking to build on its brand awareness with  consumers and extend it to large businesses.   N  Starting on Tuesday, ads will appear in print and online in such publicationsM as BusinessWeek,  CIO Magazine, Fortune and The Wall Street Journal, an IntelsN spokesman said. The campaign is  slated to last as long as three years, a goodF amount of time for an advertising push, and will run into "the tens of millions of dollars."   G  The ads will pose questions such as, "'Can a company that doesn't makeeJ computers change the course of computing history?' and 'Can a  microscopicK piece of technology solve enterprise-sized problems? The answer is 'Yes."' t  M  Intel's brand is already among the most recognized in the world, at No. 6 in K 2001 with a brand value of $34.7 billion, according to New York consultancy N Interbrand Corp. The most valuable brand last year was Coca-Cola , followed byI Microsoft , International Business Machines ,  General Electric and Nokiat (news - web sites) . T  H  Intel already has a strong presence in the server-computer market, and,M according to researcher IDC, servers using Intel processors  accounted for 89i* percent of all servers shipped last year.   L  Servers based on Intel chips are typically low- to middle-end, ranging fromF approximately $2,000 up to approximately $15,000. But Intel,  with itsJ high-end Itanium 2 chip set to ship to computer makers soon, is seeking toH move higher up the server food chain, where profit  margins are fatter.   I  Also due out this year from Intel are advanced versions of its Pentium 4 I processor for desktop and mobile PCs, as well as new Intel Xeon chips fors	 servers.    N  The company's "Intel Inside" campaign, which debuted in 1991, has been widelyK credited with building Intel's brand to where it is today. Intel subsidizeshM the cost of advertising spots bought by PC makers, who agree to let the Inteln? Pentium logo appear in their television, print and online ads.    H  Ads in recent years for Intel's Pentium chips have included performanceK ensemble The Blue Man Group, as well as dancers dressed in brightly-coloredaL "bunny suits," the protective suits worn by workers in chip factories' clean rooms.   ------------------------------  # Date: Mon, 03 Jun 2002 00:31:04 GMTK1 From: "David J. Dachtera" <djesys.nospam@fsi.net>e6 Subject: Re: Intel to start to advertise even more :-(' Message-ID: <3CFABC53.E0886B40@fsi.net>y   JF Mezei wrote:W > P > In light of discussions on why HP should mention VMS in mainstream advertsing,) > I found the following most interesting:a > P > The URL was too long on so I am pasting the article whole here. (Yahoo: if youP > want extra eyeball's for your advertising, make oit easy for folks to refer to > your pages). > * > Intel Kicks Off Biggest-Ever Ad Campaign >  Sun Jun 2, 3:32 PM ET > I > SANTA CLARA, Calif. (Reuters) - Seeking to bolster its brand with large N > corporations and the  people who run them, Intel Corp. , the world's largestM > chipmaker that is best known for its  Pentium line of microprocessors, wille; > this week kick off its largest-ever advertising campaign.p  ' Hhmmm... Is AMD nipping at their heels?k  ' Perhaps fallout from Hammer Vs. Itanic?e   -- t David J. Dachtera  dba DJE Systemsi http://www.djesys.com/  ( Unofficial Affordable OpenVMS Home Page: http://www.djesys.com/vms/soho/h   ------------------------------  # Date: Mon, 03 Jun 2002 01:42:57 GMTn1 From: "Terry C. Shannon" <terryshannon@attbi.com>o6 Subject: Re: Intel to start to advertise even more :-(/ Message-ID: <BQzK8.153935$L76.239059@rwcrnsc53>y  < "David J. Dachtera" <djesys.nospam@fsi.net> wrote in message! news:3CFABC53.E0886B40@fsi.net...t > JF Mezei wrote:o > >lF > > In light of discussions on why HP should mention VMS in mainstream advertsing, + > > I found the following most interesting:- > >-K > > The URL was too long on so I am pasting the article whole here. (Yahoo:l if youI > > want extra eyeball's for your advertising, make oit easy for folks to- refer to > > your pages). > >7, > > Intel Kicks Off Biggest-Ever Ad Campaign > >  Sun Jun 2, 3:32 PM ET > > K > > SANTA CLARA, Calif. (Reuters) - Seeking to bolster its brand with largeiH > > corporations and the  people who run them, Intel Corp. , the world's largest-J > > chipmaker that is best known for its  Pentium line of microprocessors, will= > > this week kick off its largest-ever advertising campaign.s >-) > Hhmmm... Is AMD nipping at their heels?n >t) > Perhaps fallout from Hammer Vs. Itanic?o  H My guess is that it's a preparatory attack designed to promote McKinley.   ------------------------------  # Date: Sun, 02 Jun 2002 21:17:30 GMTu3 From: Jeff Coffield <Jeffrey@DigitalSynergyInc.com>Q) Subject: Re: OpenVMS FAQ due next week...e5 Message-ID: <3CFA8B76.BCACFDD9@DigitalSynergyInc.com>-   Hoff Hoffman wrote:2  G >   I'm putting the finishing work on a massive overhaul to the OpenVMSrE >   FAQ, and as a result of this work can now directly generate text,m! >   Postscript, and HTML formats.t > F >   The downside of all this work is that the new FAQ is substantiallyG >   larger than previous editions, requiring 1415 blocks versus 887 for.H >   one of the most recent older versions.  The format has also changed. >nG >   One content question: any last-minute updates for the next edition?aB >   (I expect to be passing the new FAQ files along to the OpenVMS >   webmaster early next week) >dF >   One distribution question: do y'all want to see the eight articlesF >   that it will take to post the new FAQ -- this due to a two-hundredG >   block limit in the local news server -- or will a notification heremD >   of its availability, and where you can access the new FAQ and/orG >   download it suffice?  (Or are there other distribution alternativesr9 >   that might be prefered?  Ok, that's two questions...)h > P >  ---------------------------- #include <rtfaq.h> -----------------------------L >       For additional, please see the OpenVMS FAQ -- www.openvms.compaq.comP >  --------------------------- pure personal opinion ---------------------------N >    Hoff (Stephen) Hoffman   OpenVMS Engineering   hoffman#xdelta.zko.dec.com  Q If you wold like to add the answer to "Where can I get FastCGI software for VMS", 	 it can beI# found at www.DigitalSynergyInc.com.3   ------------------------------  $ Date: Sun, 2 Jun 2002 17:29:44 -0400  From: John Santos <JOHN@egh.com># Subject: Re: Shadow sets efficiency 5 Message-ID: <1020602171644.1319A-100000@Ives.egh.com>n  , On 1 Jun 2002 mckinneyj@cpva.saic.com wrote:  1 > In article <3CF79958.8040406@xs4all.nospam.nl>, . >  Bart Zorn <B.Zorn@xs4all.nospam.nl> writes: > > Alan E. Feldman wrote:p > >> "Marc Van Dyck" <marc.vandyck@skynet.be> wrote in message news:<3cf5df5e$0$6963$ba620e4c@news.skynet.be>... > >> s > >>>Hello,A > >>>SN > >>>    We are replacing the old storageworks disks of one of our clusters by > >>>new universalJ > >>>disks. This means we'll have to copy about 1 TB net of data. The data > >>>storage is entirelyQ > >>>on shadow sets. What is the most efficient way (elapsed time optimisation) :- > >>>-I > >>>1) initialize the new shadow sets and copy the data directly to themf > >>>i > >>>ora > >>>jN > >>>2) copy the data on normal disks and re-mount them as shadow sets (with a > >>>shadow copy) after ?  > >>>iN > >>>Also, when initializing a new shadow set, is there a way to do it without( > >>>having a shadow copy taking place ? >  > No.a >  > >>>h > >>>Thanks in advance,h > >>>y > >>>Marc Van Dyck.l > >> y > >> eK > >> In my previous post I said BACKUP/PHYSICAL does the same thing as doesfG > >> a shadow copy operation. I stand corrected. It would be faster, inlG > >> general. However, I still haven't tried it to see if the shadowingdJ > >> software would consider them both to be current members. I'll have to  > >> try it when I get a chance. > >> i > >> Disclaimer: JMHOh > >> Alan E. Feldman' > >> afeldman atski gfigroup dotski comh > > M > > I can't imagine that the shadowing software would accept the result of a tM > > BACKUP/PHYSICAL as a valid shadowset member just because it has the same t > M > It won't... the generation number in the shadow control block matches, but, M > the mount flag is also set. Since the SCB's mount flag is set the shadowingsN > software will decide that a merge is required. I don't know if the shadowingI > software will make the correct decision as to which device contains thetM > authoritative data. This would be a problem if the desired source has been,oI > or, is active other than the backup and merge operations. Any potentialnJ > problem here could be avoided by simply initializing the volume that wasO > created using a physical backup. An unqualified INITIALIZE will just recreate.N > the volume label and reserved files. The remainder of the disk is unchanged.J > If this is done then a shadow copy will be performed rather than a mergeN > since the SCB is now absent. The copy will require very few write operationsN > since, as others have pointed out, both the copy and merge functions performI > a compare prior to any write and only write when a difference is found.   D I don't understand the benefit of this.  Why is it faster to read A,H compare with B (with the write A to B if different rarely if ever done)  than it is to read A, write B?  E In other words don't two reads from different disks take just as long0* as a read from one and a write to another?  F (You could use two different methods to compare the disk blocks.  ReadI both blocks into separate buffers and use the CPU to compare the buffers, K or read one block into a buffer and then use a read/compare (or readcheck?)eE function on the second drive to let the disk hardware do the compare.oD Clearly the second method should be faster since the compare is doneJ during the second transfer instead of after it, but maybe all drives don't support this function.)l  E It should take just as long (at the disk head) to read a block (or toeH do a read/compare) as it does to write a block.  The head is moving pastC the surface at the same speed, so the bit rate should be identical.l  I If this has something to do with rotational latency and missed transfers,bD wouldn't using multiple large buffers and async I/O (as BACKUP does)" mask the effect almost completely?  K > Whenever the composition of a shadow set changes the generation number ofDJ > the current members' SCB is incremented. So if a volume is dismounted orF > removed due to a crash it's generation number would be less than theC > remaining shadow members (whose generation numbers have just beenV? > incremented). When it re-joins the shadow set it will requiresG > a copy. While each shadow member is mounted into it's shadow set it'seO > SCB's mount flag is set. If it is dismounted the flag is cleared. If a systemrL > that is serving a disk crashes then the mount flag remains set. So too, ifH > there are files open on this volume during system shutdown so that theE > disk can not be properly dismounted. In either event, when the diskoH > re-joins the shadow set a merge is required. Simple right? But what ifE > the mount flag is set and the generation number differs? Well, the fI > differing generation numbers take precedence. A copy would be performedrD > in all cases where the generation number is different (or absent). > G > I suggest using the strategy from the following example if you decide E > to use BACKUP/PHYSICAL (unless you know that the shadow set will be-/ > static during the entire reformation process.  >  >  > $ SHOW DEVICE DSA2 > R > Device                  Device           Error    Volume         Free  Trans MntR >  Name                   Status           Count     Label        Blocks Count CntR > DSA2:                   Mounted              0  USER01          823815     1   1C > $1$DIA2:      (R5IMFG)  ShadowSetMember      0  (member of DSA2:)t > ) > $ MOUNT/FOREIGN/OVERIDE=SHADOW $1$DIA3:o" > $ BACKUP/PHYSICAL DSA2: $1$DIA3: > $ DISMOUNT/NOUNLOAD $1$DIA3: > $ INITIALIZE $1$DIA3: SHADOW- > $ MOUNT/SYSTEM DSA2:/SHADOW=$1$DIA3: USER02e >  > $ SHOW DEVICE DSA2 > R > Device                  Device           Error    Volume         Free  Trans MntR >  Name                   Status           Count     Label        Blocks Count CntR > DSA2:                   Mounted              0  USER01          823815     1   1C > $1$DIA2:      (R5IMFG)  ShadowSetMember      0  (member of DSA2:)iO > $1$DIA3:      (R5QLSA)  ShadowCopying        0  (copy trgt DSA2:   0% copied)e > I > > generation number. Equal generation numbers are only valid when more  L > > than one member gets mounted into an empty (or not yet existing) shadow L > > set. In the case that a volume gets added to a shadow set, the software M > > will invalidate the new member and subject it to a copy operation. Being IJ > > the result of said BACKUP/PHYSICAL it will be an efficient copy since A > > only a relatively small number of blocks needs to be written.  > > 
 > > Bart Zorn  > >  > >  > >  > -- l > - Jimp >  >    -- , John Santosu Evans Griffiths & Hart, Inc. 781-861-0670 ext 539   ------------------------------   End of INFO-VAX 2002.305 ************************