|
From: Aurelienne A. S. J. <aur...@gm...> - 2013-08-23 19:46:53
|
Hello! I'm a post-graduation student and my conclusion work is about Postgres-XC. I've read a lot about the project and I really think this is a powerful tool. Unfortunately, I couldn't achieve the performance I had expected. I've configured a cluster with 4 virtualized servers (all of them on the same hardware using Virtual Box): Server 1: gtm Server 2: 1 gtm_proxy + 1 datanode + 1 coordinator Server 3: 1 gtm_proxy + 1 datanode + 1 coordinator Server 4: 1 gtm_proxy + 1 datanode + 1 coordinator The version of Postgres-XC I'm using is 1.0.3, and the OS in all servers is Debian 7. I used pgbench to compare the number of transactions per second in Postgres-XC and in a standalone PostgreSQL server. Postgres-XC achieved only half TPS compared to PostgreSQL. The test was basically consisted of inserts into distributed tables (I used a custom script in pgbench) simulating many clients simultaneously. I would like to know what configuration would be enough to have a better performance than PostgreSQL. - How many servers would be enough to have a write/read scalability compared to a single PostgreSQL? - Would it be a problem to have all servers on the same hardware? - Should I distribute the workload among all coordinators? I'm basically using one coordinator, once I couldn't find a way to use all of them simultaneously with pgbench. I would be very thankful if you could help me. Regards, Aurelienne Jorge. |
|
From: Mason S. <ma...@st...> - 2013-08-23 22:20:09
|
On Fri, Aug 23, 2013 at 12:46 PM, Aurelienne Aparecida Souza Jorge < aur...@gm...> wrote: > Hello! > > I'm a post-graduation student and my conclusion work is about Postgres-XC. > I've read a lot about the project and I really think this is a powerful > tool. Unfortunately, I couldn't achieve the performance I had expected. > > I've configured a cluster with 4 virtualized servers (all of them on the > same hardware using Virtual Box): > Server 1: gtm > Server 2: 1 gtm_proxy + 1 datanode + 1 coordinator > Server 3: 1 gtm_proxy + 1 datanode + 1 coordinator > Server 4: 1 gtm_proxy + 1 datanode + 1 coordinator > How many cores are on the underlying system? Keep in mind that those resources are going to be shared. Are you CPU bound? Was your base PostgreSQL instance for comparison also running in a VM? In addition, does each server have its own dedicated storage? Or are they all contending for the same physical device underneath? Do you have fsync on (and is it really fsync'ed by the VM)? > > The version of Postgres-XC I'm using is 1.0.3, and the OS in all servers > is Debian 7. > > I used pgbench to compare the number of transactions per second in > Postgres-XC and in a standalone PostgreSQL server. Postgres-XC achieved > only half TPS compared to PostgreSQL. The test was basically consisted of > inserts into distributed tables (I used a custom script in pgbench) simulating > many clients simultaneously. > > I would like to know what configuration would be enough to have a better > performance than PostgreSQL. > > - How many servers would be enough to have a write/read scalability > compared to a single PostgreSQL? > > - Would it be a problem to have all servers on the same hardware? > > - Should I distribute the workload among all coordinators? I'm basically > using one coordinator, once I couldn't find a way to use all of them simultaneously > with pgbench. > > I would be very thankful if you could help me. > > > Regards, > > Aurelienne Jorge. > > > ------------------------------------------------------------------------------ > Introducing Performance Central, a new site from SourceForge and > AppDynamics. Performance Central is your source for news, insights, > analysis and resources for efficient Application Performance Management. > Visit us today! > http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk > _______________________________________________ > Postgres-xc-general mailing list > Pos...@li... > https://lists.sourceforge.net/lists/listinfo/postgres-xc-general > > -- Mason Sharp StormDB - http://www.stormdb.com The Database Cloud Postgres-XC Support and Services |
|
From: Wolfgang K. <fel...@gm...> - 2013-08-25 11:35:01
|
> I'm a post-graduation student Hm. > and my conclusion work is about Postgres-XC. I've read a lot about > the project and I really think this is a powerful tool. > Unfortunately, I couldn't achieve the performance I had expected. > > I've configured a cluster with 4 virtualized servers (all of them on > the same hardware using Virtual Box): You're running several RDBMS server instances on one physical computer and you expect to get more throughput from that than with one single instance running on the same physical computer? Errr, huh? > I used pgbench to compare the number of transactions per second in > Postgres-XC and in a standalone PostgreSQL server. Postgres-XC > achieved only half TPS compared to PostgreSQL. The test was basically > consisted of inserts into distributed tables (I used a custom script > in pgbench) simulating many clients simultaneously. That's what is to be expected. The same hardware minus the additional overhead for virtualisation and replication *has to* give you less throughput than one single RDBMS server instance. And if you emulate the clients by running scripts on the same hardware as the servers... > - How many servers would be enough to have a write/read scalability > compared to a single PostgreSQL? More than one. *Hardware* servers. > - Would it be a problem to have all servers on the same hardware? That's not just "a problem", that's the whole point. If you want more throughput, you need more hardware*. That's what replication is made for in the first place. Sincerely, Wolfgang *Of course you also must optimise your configuration to get the maximum out of the hardware. |