Re: HDR Performance

On Mar 26, 8:25 pm, jpren...@yahoo.com wrote: > On Mar 26, 4:02 am, tc <terance.cam...@gmail.com> wrote: > > > I want to know > > > - How to measure/estimate the Network Bandwidth for a HDR pair > >   over a WAN ? > > If this is what I think you are asking, what you would want to do is > calculate the amount of data you are writing into your logical logs, > which you can roughly do looking at the time it takes you to complete > a logical log from the messages in the online.log and then knowing how > big they are.  From those two things you could get a rough amount of > bytes/sec that would need to be sent from your primary machine to your > secondary machine. > > > - What will happen to the performance of the Primary server if > >   Network slows down ? > > , > If the network restricts the rate at which you can send your data to > the secondary (you generate more log data then can be sent), you would > start to see some bottlenecking on access to the logical logs likely > (more threads waiting to access the logical log and thus slower write > activity on the primary), and then also your checkpoint duration would > also spike up.  As the checkpoint record needs to be sent to the > secondary and ack'd before it can complete on the primary and so there > would be a bit of a backlog of data to be sent before the checkpoint > could be sent.  That backlog time would be included in the checkpoint > duration. > > > - If Primary fails during high trasaction processing will the > > Secondary > >   be in Sync > > Depends on if you use HDR in the sync or async setting.  If you are > using sync, then every logical log buffer flush that occurs on the > primary has to be ack'd by the secondary before the log buffer flush > completes on the primary.  So the most that a sync HDR pair would lose > is whatever was in the current logical log buffer on the primary when > it crashed.  If you are using async mode the logical log buffer > flushes in the primary don't need to be sent and ack'd before > continueing.  In this case there are 12 buffers (of logical log buffer > size) on the primary that are used to store logical log buffer data > that could be full and waiting to be sent, so then that would be the > largest amount of data your secondary server would not receive if your > primary crashed with all 12 of those buffers full. > > Jacques > > > - Are there tools/functions available in Open Admin > >       -  to monitor the trafic between the HDR pair > >       - to see how sync the Secondary is > > > TC Thanks !!! I sent an email to Jacques Roy asking more clarifications.