sequoiadb-discuss

Re: [Sequoiadb-discuss] [Sequoia] Sequioa 2.10.10 extremly slow...

[Emmanuel C. <ma...@fr...>]

Hi Robert,

Note that the mailing list has moved to 
seq...@li...
To better understand what the benchmark is doing, could you tell us more 
about the transactions that are being executed? How many reads/writes 
per transaction? How many threads executing transactions in parallel?
About your setup, what network connectivity do you have between the 
machines? What group communication configuration do you use between your 
controllers?
When you perform your reference run on a single Postgres, does Postgres 
run in a VM as well? Does JDBCBench run in a remote VM?

I have never used VirtualBox so I am not aware of any performance 
problem that could come from it, but running a group communication 
benchmark on this setup seems like a logical thing to do to make sure 
that this is not the root cause of the latency overhead on every single 
query.

Keep us posted with your findings
Emmanuel

> I'm currently evaluating Sequoia as a PostgreSQL-Cluster Solution, but
> something in my configuration files seems to be wrong. JDBCBench says:
>
> 	Time to execute 100 transactions: 57.821 seconds.
> 	Max/Min memory usage: 2640936 / 489680 kb
> 	0 / 100 failed to complete.
> 	Transaction rate: 1.729475450095986 txn/sec.
>
> And i already had to set the "number of accounts in 1 tps db" to 1000
> it would have taken hours to init the accounts table with 100.000.
> I'm currently using 2 Controllers and 2 Backends running in VirtualBox
> Virtual Machines with 1GB of Ram on some quadcore servers. If i'm
> running JDBCBench on a single node postgresql database the result would
> be:
>
> 	Time to execute 1000 transactions: 9.679 seconds.
> 	Max/Min memory usage: 2890896 / 638312 kb
> 	0 / 1000 failed to complete.
> 	Transaction rate: 103.31645831180907 txn/sec.
>
> When i'm doing the JDBCBench-Test on the cluster, the cpu load is mostly
> at 0.8 to 1, with about 30% idle and 20% user and 50% sys. Does someone
> has a clue what i've done wrong?
>
>
>
> Sincerly yours,
>
> Robert
>
>   
> ------------------------------------------------------------------------
>
> _______________________________________________
> Sequoia mailing list
> Se...@li...
> http://forge.continuent.org/mailman/listinfo/sequoia


-- 
Emmanuel Cecchet
FTO @ Frog Thinker 
Open Source Development & Consulting
--
Web: http://www.frogthinker.org
email: ma...@fr...
Skype: emmanuel_cecchet

Re: [Sequoiadb-discuss] [Sequoia] Sequoia 2.10.10 extremely slow...

[Robert D. <rob...@bp...>]

> Hi Robert,

Hi,



> Note that the mailing list has moved to 
> seq...@li...

Ok, so i kept the whole Email.



> To better understand what the benchmark is doing, could you tell us more 
> about the transactions that are being executed? How many reads/writes 
> per transaction? How many threads executing transactions in parallel?

It's a TPC-B Benchmark and was later modified by Guillaume Smet
(sequoia-list) for having a read-only benchmark. There are three
UPDATEs, one INSERT and one SELECT per transaction and i tried several
settings of 10 or more clients doing one transaction at the same time.



> About your setup, what network connectivity do you have between the 
> machines?

100Mbit.



> What group communication configuration do you use between your 
> controllers?

I'm using the hedera jgroup communication like in the distributed demo.



> When you perform your reference run on a single Postgres, does Postgres 
> run in a VM as well? Does JDBCBench run in a remote VM?

All the PostgreSQL-servers are running on virtualmachines on different
servers and JDBCBench is running on my box. 



Sincerly yours,

Robert


---
Original Message:

> > I'm currently evaluating Sequoia as a PostgreSQL-Cluster Solution, but
> > something in my configuration files seems to be wrong. JDBCBench says:
> >
> > 	Time to execute 100 transactions: 57.821 seconds.
> > 	Max/Min memory usage: 2640936 / 489680 kb
> > 	0 / 100 failed to complete.
> > 	Transaction rate: 1.729475450095986 txn/sec.
> >
> > And i already had to set the "number of accounts in 1 tps db" to 1000
> > it would have taken hours to init the accounts table with 100.000.
> > I'm currently using 2 Controllers and 2 Backends running in VirtualBox
> > Virtual Machines with 1GB of Ram on some quadcore servers. If i'm
> > running JDBCBench on a single node postgresql database the result would
> > be:
> >
> > 	Time to execute 1000 transactions: 9.679 seconds.
> > 	Max/Min memory usage: 2890896 / 638312 kb
> > 	0 / 1000 failed to complete.
> > 	Transaction rate: 103.31645831180907 txn/sec.
> >
> > When i'm doing the JDBCBench-Test on the cluster, the cpu load is mostly
> > at 0.8 to 1, with about 30% idle and 20% user and 50% sys. Does someone
> > has a clue what i've done wrong?
> >
> >
> >
> > Sincerly yours,
> >
> > Robert