In fact this test shows 3 problems:
1. Bad scalabity even with embedded server:  H2 can't use more then 10% of 
PU due to locking 
2. Slow TCP exchange even with localhost
3. Degradation of TCP server performance, probably due to a leak

I would very appreciate your input for each of the problems

My final goal - 2000 tps with TCP server. Each transaction includes approx. 
10 inserts, 10 updates by primary key, 50 short selects. Database size 
>1TB. CPU cores are not limited, we can have 30-100 cores
Do you think it is achivable?  Do you have such experience?
  

On Sunday, December 11, 2016 at 7:41:34 PM UTC, Noel Grandin wrote:
>
> thanks.
>
> looks like we are bottlenecking on the locking on the undoLog object in 
> TransactionStore, particularly in the commit() method.
>
> I see a comment there
>    // TODO could synchronize on blocks (100 at a time or so)
> which means perhaps Thomas had some ideas how this could be improved.
>
> Bit late here now, will thank about this more tomorrow.
>
> Still don't understand why the TCP server is so much slower than direct, 
> but perhaps it's a combined latenct thingof the locking and the socket 
> connection limiting the throughput?
>
> ​
>

-- 
You received this message because you are subscribed to the Google Groups "H2 
Database" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/h2-database.
For more options, visit https://groups.google.com/d/optout.

Reply via email to