ignite-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sergi Vladykin <sergi.vlady...@gmail.com>
Subject Re: Local Node Query Optimization
Date Mon, 17 Aug 2015 19:10:32 GMT
Hi,

First of all I would not expect that general purpose solution (Apache
Ignite) can be fairly compared in performance to any highly optimized
custom implementation (yours with cache friendly containers). Just because
Ignite is a distributed highly parallel system which supports much more use
cases than you implemented.

There are multiple ways we can try to solve your problem:

1. Use custom Indexing implementation with your optimized low latency C++
implementation inside. See
https://ignite.incubator.apache.org/releases/1.3.0/javadoc/org/apache/ignite/spi/indexing/IndexingSpi.html

2. I would suggest to try H2 Geospatial extension which can be used in
Ignite instead of z-order curve and check if it will work better for you.
See
https://github.com/apache/incubator-ignite/blob/f9fe99963fd3defab639e9700dffc965cdd39bb1/modules/geospatial/src/test/java/org/apache/ignite/internal/processors/query/h2/GridH2IndexingGeoSelfTest.java

In addition to Val's comments I also would not recommend to run local
queries with partitioned cache because it is by design not a fault tolerant
solution.

Sergi



2015-08-17 21:13 GMT+03:00 vkulichenko <valentin.kulichenko@gmail.com>:

> Hi Johnny,
>
> I'm not sure why it's so much slower than with some C++ implementation
> (what
> did you use, BTW?). Most likely there are ways to improve local execution
> and I think Sergi Vladykin should have some comments on this, as he is an
> expert in this part of Ignite.
>
> But anyway, if performance scales, why do you want to use local queries?
> Your web app can start an embedded client node which will execute queries
> on
> a remote cluster which can have as many nodes as you need. You can even
> have
> all the nodes on one nodes if it's dictated by your environment.
> This approach is much more flexible, because you can add nodes at any
> moment
> (for example, your data set size increases and performance gets worse). In
> case of local query you will need to switch to a bigger physical box to
> store the data and you can do nothing with performance.
>
> -Val
>
>
>
> --
> View this message in context:
> http://apache-ignite-users.70518.x6.nabble.com/Local-Node-Query-Optimization-tp996p1000.html
> Sent from the Apache Ignite Users mailing list archive at Nabble.com.
>

Mime
View raw message