Insert performance

[yann Blazart <ya...@gmail.com>]

Hello,

I'm doing tests to get the best performance whily using Ignite.

I've to insert a lot of data, then doing some requests.

First, in my tests, I'm using objects with @SqlFields annotations,
cache.putAll is faster than IgniteDataStream, is it normal ? I'm using
cache.putAll to send 5000 objects at the same time.

Second, when I'm doing my insertion test, 30 millions of objects, using two
servers node and one client node on same machine, I raise 500 insert per
milliseconds, wich is excellent !

But if I use two differents machines, threw my network, performances go
down to 200 inserts per milliseconds. But my network is far away to be busy
(only 80Mb/s used on 200Mb/s available on simple test).

Is there any reason ?

Thanks in advance,

Regards.

Re: Insert performance

[Ilya Kasnacheev <il...@gmail.com>]

Hello!

This might be the point of IgniteDataStreamer (bigger latency, bigger
throughput).
However, you should be feeding data in parallel to IgniteDataStreamer, from
multiple threads, for optimal performance.

There are a lot of tuning considerations, such as thread pools, etc.

Regards,
-- 
Ilya Kasnacheev


пн, 14 янв. 2019 г. в 23:34, yann Blazart <ya...@gmail.com>:

> Hello,
>
> I'm doing tests to get the best performance whily using Ignite.
>
> I've to insert a lot of data, then doing some requests.
>
> First, in my tests, I'm using objects with @SqlFields annotations,
> cache.putAll is faster than IgniteDataStream, is it normal ? I'm using
> cache.putAll to send 5000 objects at the same time.
>
> Second, when I'm doing my insertion test, 30 millions of objects, using
> two servers node and one client node on same machine, I raise 500 insert
> per milliseconds, wich is excellent !
>
> But if I use two differents machines, threw my network, performances go
> down to 200 inserts per milliseconds. But my network is far away to be busy
> (only 80Mb/s used on 200Mb/s available on simple test).
>
> Is there any reason ?
>
> Thanks in advance,
>
> Regards.
>