Re: Anybody seen a very fast wild cat out there?

> > Also how can does it determine how much activity can be performed > > in the time period? That must depend on how busy other processes are > > on the host > > It's not so much the amount of activity, but the amount which we need to > recover. If we sense that you want to recover the server in 5 min in > all cases, and that means that we need to issue a checkpoint after 10 > min., then we do so. Again, since we now have non-blocking checkpoints, > then the cost of a checkpoint is not a big expense. > How do you calculate the time taken to recover for a given amount of stuff to recover? What stuff does the formulae behind this use? Different machines will recover the same amount of stuff in different times! > > > > and (for me) how busy the SAN is from activity from other hosts > > (yes 2 hosts can share a disk array!). > > > PS SHMVIRTALLOCSEG having an alarm feature is cool! > > > Also ontape to directories (finally!) is cool but the ability to list > > multiple (colon seperated, like DBSPACETEMP) directories and have > > individual volumes to always go to the directory that is on the > > filesystem with the most free space would be nice. I have a HUGE > > script to > > do this with ontape at present and it takes a LOT of testing! (See I > > should have been in the closed beta!). > > > Finally the ability to limit the amount of memory used by an > > individual session would be nice, again with an alarm). If I have a > > server with 64Gb of RAM > > and a sessions goes rogue I don't want it allocating >60GB of RAM!- Hide quoted text - > > - Show quoted text -

david@smooth1.co.uk wrote: >>> Also how can does it determine how much activity can be performed >>> in the time period? That must depend on how busy other processes are >>> on the host >> It's not so much the amount of activity, but the amount which we need to >> recover. If we sense that you want to recover the server in 5 min in >> all cases, and that means that we need to issue a checkpoint after 10 >> min., then we do so. Again, since we now have non-blocking checkpoints, >> then the cost of a checkpoint is not a big expense. >> > > How do you calculate the time taken to recover for a given amount > of stuff to recover? > > What stuff does the formulae behind this use? > > Different machines will recover the same amount of stuff in > different times! AFAIK, the calculation is based on statistics collected during engine activity, like buffer cleans per disk, (K)AIO rate etc. It's not a static value. It will be based on measures taken on the actual system and Informix instance. Regards. -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently...

david@smooth1.co.uk wrote: >>> Also how can does it determine how much activity can be performed >>> in the time period? That must depend on how busy other processes are >>> on the host >> It's not so much the amount of activity, but the amount which we need to >> recover. If we sense that you want to recover the server in 5 min in >> all cases, and that means that we need to issue a checkpoint after 10 >> min., then we do so. Again, since we now have non-blocking checkpoints, >> then the cost of a checkpoint is not a big expense. >> > > How do you calculate the time taken to recover for a given amount > of stuff to recover? > > What stuff does the formulae behind this use? > > Different machines will recover the same amount of stuff in > different times! But generally the recovery time will be in proportion to the amount of time that they did their original work. > > > >> >>> and (for me) how busy the SAN is from activity from other hosts >>> (yes 2 hosts can share a disk array!). >>> PS SHMVIRTALLOCSEG having an alarm feature is cool! >>> Also ontape to directories (finally!) is cool but the ability to list >>> multiple (colon seperated, like DBSPACETEMP) directories and have >>> individual volumes to always go to the directory that is on the >>> filesystem with the most free space would be nice. I have a HUGE >>> script to >>> do this with ontape at present and it takes a LOT of testing! (See I >>> should have been in the closed beta!). >>> Finally the ability to limit the amount of memory used by an >>> individual session would be nice, again with an alarm). If I have a >>> server with 64Gb of RAM >>> and a sessions goes rogue I don't want it allocating >60GB of RAM!- Hide quoted text - >> - Show quoted text - > >

On 17 Feb, 02:28, Fernando Nunes <s...@domus.online.pt> wrote: > d...@smooth1.co.uk wrote: > >>> Also how can does it determine how much activity can be performed > >>> in the time period? That must depend on how busy other processes are > >>> on the host > >> It's not so much the amount of activity, but the amount which we need to > >> recover. If we sense that you want to recover the server in 5 min in > >> all cases, and that means that we need to issue a checkpoint after 10 > >> min., then we do so. Again, since we now have non-blocking checkpoints, > >> then the cost of a checkpoint is not a big expense. > > > How do you calculate the time taken to recover for a given amount > > of stuff to recover? > > > What stuff does the formulae behind this use? > > > Different machines will recover the same amount of stuff in > > different times! > > AFAIK, the calculation is based on statistics collected during engine activity, like buffer cleans per disk, (K)AIO rate etc. > It's not a static value. It will be based on measures taken on the actual system and Informix instance. > Yes, can we get these in the documentation? And a way to view all this information via onstat/SMI? > Regards. > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently...- Hide quoted text - > > - Show quoted text -

david@smooth1.co.uk wrote: > > Yes, can we get these in the documentation? And a way to view all > this information via onstat/SMI? Check onstat -g iof and onstat -g ckp (new option). I believe (may be wrong), that they contain much of this information. Not sure if it's in the format you'd like... Regards. -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently...

On 18 Feb, 18:41, Fernando Nunes <s...@domus.online.pt> wrote: > d...@smooth1.co.uk wrote: > > > Yes, can we get these in the documentation? And a way to view all > > this information via onstat/SMI? > > Check onstat -g iof and onstat -g ckp (new option). I believe (may be wrong), that they contain much of this information. > Not sure if it's in the format you'd like... > What this the formulae used on these figures to calculate when a checkpoint needs to be done? > Regards. > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently...

david@smooth1.co.uk wrote: > On 18 Feb, 18:41, Fernando Nunes <s...@domus.online.pt> wrote: >> d...@smooth1.co.uk wrote: >> >>> Yes, can we get these in the documentation? And a way to view all >>> this information via onstat/SMI? >> Check onstat -g iof and onstat -g ckp (new option). I believe (may be wrong), that they contain much of this information. >> Not sure if it's in the format you'd like... >> > > What this the formulae used on these figures to calculate when a > checkpoint needs to be done? I don't have that level of detail. You can try to ask at the Cheetah Forum at: http://www.ibm.com/developerworks/forums/dw_forum.jsp?forum=1071&cat=19 ... or wait that someone answers here. I don't know if this information will be provided, but the forum is constantly monitored by support staff. They can get you an official answer I hope. Regards. -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently...