Re: Do I need to be RESIDENT

Would a setting of 1 be the same as -1? TIA, Zev Berezin From: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> To: informix-list@iiug.org Subject: Re: Do I need to be RESIDENT Date sent: Wed, 30 Mar 2005 18:55:38 +0200 Organization: 1&1 Internet AG Send reply to: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> > In this case I would expect a real benefit. The operating system has to > manage every single page of shared memory and has to store and cache > address mappings for every page. Setting "RESIDENT -1" on Solaris means > reducing the number of pages by a factor of 512. > > I also know a case of a comparable system where "onmode -k" and onstat > commands became slow until "RESIDENT -1" was used. This was presumably > caused by the operating system because detaching so many pages of memory > was a major task. > > Michael > > > Colin Dawson wrote: > > Platform is Solaris 9, 16CPU, 32Gb memory > > > > > > > > Regards > > > > Colin > > > > > > > > > > > > > >>From: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> > >>To: Colin Dawson <cjd_1955@hotmail.com> > >>Subject: Re: Do I need to be RESIDENT > >>Date: Wed, 30 Mar 2005 14:39:08 +0200 > >> > >>What platform are you on? > >> > >>Making memory resident that is not paged out anyway may not make sense. But > >>on Solaris the "RESIDENT -1" setting has a second effect: It creates shared > >>memory with a page size of 4 MB instead of 8k pages (using the > >>SHM_SHARE_MMU flag of shmat()). This speeds up physical to virtual page > >>translation in the cpu and has a performance benefit on large systems. > >> > >>Michael > >> > >>Colin Dawson wrote: > >> > >>>We currently have RESIDENT set to -1, if I increase BUFFERS above 2M I get > >>>the message > >>> > >>>07:55:17 Informix Dynamic Server Stopped. > >>>07:56:49 mt_shm_init: can't create resident segment> >>> > >>>When I change RESIDENT to 0, I can increase the buffers to whatever I > >>>want. I know why I'm getting this error, that is not the reason for this > >>>posting. What I'm looking for is an informed opinion of why set RESIDENT > >>>to -1 in the first place, Is there a real benefit or only a percieved one? > >>> > >>> > >>> > >>> > >>> > >>>Regards > >>> > >>>Colin Dawson > >>> > >>> > >>>sending to informix-list > >> > > > > > > sending to informix-list > sending to informix-list

The onconfig value of RESIDENT is the number of segments to make resident, -1 means all, 1 means the first one, etc. (Use onstat -g seg to check. Resident segments are marked with '*' in the class column.) On Solaris the 4MB page size is used for any segment made resident, no matter how many there are. Michael Zev Berezin wrote: > Would a setting of 1 be the same as -1? > > TIA, > Zev Berezin > > From: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> > To: informix-list@iiug.org > Subject: Re: Do I need to be RESIDENT > Date sent: Wed, 30 Mar 2005 18:55:38 +0200 > Organization: 1&1 Internet AG > Send reply to: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> > >>In this case I would expect a real benefit. The operating system has to >>manage every single page of shared memory and has to store and cache >>address mappings for every page. Setting "RESIDENT -1" on Solaris means >>reducing the number of pages by a factor of 512. >> >>I also know a case of a comparable system where "onmode -k" and onstat >>commands became slow until "RESIDENT -1" was used. This was presumably >>caused by the operating system because detaching so many pages of memory >>was a major task. >> >>Michael >> >> >>Colin Dawson wrote: >> >>>Platform is Solaris 9, 16CPU, 32Gb memory >>> >>> >>> >>>Regards >>> >>>Colin >>> >>> >>> >>> >>> >>> >>> >>>>From: Michael Mueller <m.mueller01.delete_me@kay-mueller.de> >>>>To: Colin Dawson <cjd_1955@hotmail.com> >>>>Subject: Re: Do I need to be RESIDENT >>>>Date: Wed, 30 Mar 2005 14:39:08 +0200 >>>> >>>>What platform are you on? >>>> >>>>Making memory resident that is not paged out anyway may not make sense. But >>>>on Solaris the "RESIDENT -1" setting has a second effect: It creates shared >>>>memory with a page size of 4 MB instead of 8k pages (using the >>>>SHM_SHARE_MMU flag of shmat()). This speeds up physical to virtual page >>>>translation in the cpu and has a performance benefit on large systems. >>>> >>>>Michael >>>> >>>>Colin Dawson wrote: >>>> >>>> >>>>>We currently have RESIDENT set to -1, if I increase BUFFERS above 2M I get >>>>>the message >>>>> >>>>>07:55:17 Informix Dynamic Server Stopped. >>>>>07:56:49 mt_shm_init: can't create resident segment>>>>> >>>>>When I change RESIDENT to 0, I can increase the buffers to whatever I >>>>>want. I know why I'm getting this error, that is not the reason for this >>>>>posting. What I'm looking for is an informed opinion of why set RESIDENT >>>>>to -1 in the first place, Is there a real benefit or only a percieved one? >>>>> >>>>> >>>>> >>>>> >>>>> >>>>>Regards >>>>> >>>>>Colin Dawson >>>>> >>>>> >>>>>sending to informix-list >>>> >>> >>>sending to informix-list >> > > > sending to informix-list