RE: Checkpoint Duration - HDR - Sync/Async

Thanks Alexey, i really appreciate your help I think that the dss query was taking disk resources, but regardless of cpu or disk resources, I think that when a ckeckpoint occurs, Informix stops running any query (even already started querys) and doesn't execute new ones, and then execute the checkpoint, am i right? Could you explain me a little more about what you mention about rolling forward logical log records received from the primary ? I am sure that i am not the first person who implements HDR for high availability purposes, and for executing DSS querys on the secondary server, so i think that i am missing something. Thanks Regards -----Mensaje original----- De: Alexey Sonkin [mailto:alexeis@grandvirtual.com] Enviado el: Lunes, 06 de Octubre de 2003 02:47 p.m. Para: 'Francisco Roldan'; informix-list@iiug.org Asunto: RE: Checkpoint Duration - HDR - Sync/Async Hi, Francisco, The problem is that DSS query on secondary might take a lot of CPU/disk resources that a requred to roll forward logical log records received from the primary. The first thing to do is to define whether it's CPU or disk bottneck on the secondary ------------------------------------------ Alexey Sonkin Senior Database Administrator > -----Original Message----- > From: Francisco Roldan [mailto:froldan@5b.com.gt] > Sent: Friday, October 03, 2003 11:48 AM > To: informix-list@iiug.org > Subject: RV: Checkpoint Duration - HDR - Sync/Async > > Hi everyboy, > > I really would appreciate your comments about this subject. > > Thanks in advance > > -----Mensaje original----- > De: Francisco Roldan [mailto:froldan@5b.com.gt] > Enviado el: Mi'rcoles, 01 de Octubre de 2003 12:30 p.m. > Para: informix-list@iiug.org > Asunto: Checkpoint Duration - HDR - Sync/Async > > Hi everybody, > my environment : > > Two identical servers running AIX 4.3.3 - IDS 7.3.1 UD1 running HDR > between both database servers > DRINTERVAL : 360 segs (=> ASYNCHRONOUS)> I have only one production database on Buffered Logging Mode. > > I implemented HDR in order to protect the data changed after the last > backup, and to be able to execute reports/DSS-querys on the secondary > server (in that order). > > I read that the checkpoint between primary and secondary servers > running HDR are synchronous, regardless of DRINTERVAL parameter is > -1 (=> SYNCHRONOUS) or not. > > When i execute a very heavy dss query on my secondary server, it > affects the performance on this server and i have noticed that it > affects the duration of the checkpoint on both servers (even getting 6 > times the normal duration). > It is logic if i consider that the checkpoints between primary and > secondary > servers are synchronous, am i right ? or am i missing something else that > could be affecting the checkpoints duration on both servers? > > Is it possible to minimize the impact on the primary server's > checkpoint duration ? For examble by tunning some parameters like > DRINTERVAL, CKPTINTVL(240 segs. on my servers), or the difference > between these values (I am just guessing). Or the Logging mode on my > only production database. > > Thanks in advance > > Regards > > > sending to informix-list > > sending to informix-list sending to informix-list

On Mon, 06 Oct 2003 18:12:36 -0400, Francisco Roldan wrote: Hi. > Thanks Alexey, > i really appreciate your help > > I think that the dss query was taking > disk resources, but regardless of > cpu or disk resources, I think that > when a ckeckpoint occurs, Informix stops running any query (even already > started querys) and doesn't execute new ones, and then execute the checkpoint, > am i right? <SNIP> Checkpoints officially only block threads that require a critical section, ie those updating pages. Unofficially there is/was a design flaw at least through 7.30 wherein any non-critical threads running in CPU VP #1 were suspended while that VP managed the disk flushes, roughly for the first half of the checkpoint duration. This was supposed to be addressed in 7.31 and 9.21 but I do not know if the problem was ever solved. Non-critical threads running on other CPU VPs are not affected. Art S. Kagel