Re: Understanding Write/Read IOPS during Ignite data load through Streamer and JDBC SQL

[krkumar24061...@gmail.com]

Hi Ilya - We do not have this message in the logs that means we have to
increase the resources I guess. thanx for the help in this regard.

Thanx and Regards,.
KR Kumar



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Understanding Write/Read IOPS during Ignite data load through Streamer and JDBC SQL

[Ilya Kasnacheev]

Hello!

Do you have any "Indexed columns of a row cannot be fully inlined" messages
in your logs?

Otherwise, I'm afraid answer is no, mostly. Either increase off-heap size
or live with reduced performance.

Regards,
-- 
Ilya Kasnacheev


ср, 11 дек. 2019 г. в 06:57, krkumar24061...@gmail.com <
krkumar24061...@gmail.com>:

> HI ilya - Thanx for the reply. What you said is exactly what's happening I
> guess. We do see this warning that pages will be rotated from disk and the
> write process will slow down.
>
> Here is the message that we see in the logs:
>
> WARN ][data-streamer-stripe-22-#59][PageMemoryImpl] Page replacements
> started, pages will be rotated with disk, this will affect storage
> performance (consider increasing DataRegionConfiguration#setMaxSize).
>
> We do have inline indexes on the SQL tables, that's also probably  causing
> some of this I guess.
>
> Is there any solution other than increasing the off heap size ??
>
> Thanx and Regards,
> KR Kumar
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>

Re: Understanding Write/Read IOPS during Ignite data load through Streamer and JDBC SQL

[krkumar24061...@gmail.com]

HI ilya - Thanx for the reply. What you said is exactly what's happening I
guess. We do see this warning that pages will be rotated from disk and the
write process will slow down. 

Here is the message that we see in the logs:

WARN ][data-streamer-stripe-22-#59][PageMemoryImpl] Page replacements
started, pages will be rotated with disk, this will affect storage
performance (consider increasing DataRegionConfiguration#setMaxSize).

We do have inline indexes on the SQL tables, that's also probably  causing
some of this I guess.

Is there any solution other than increasing the off heap size ??

Thanx and Regards,
KR Kumar



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Understanding Write/Read IOPS during Ignite data load through Streamer and JDBC SQL

[Ilya Kasnacheev]

Hello!

When you are writing only, this means you can write to memory now and
persist it to disk later. This is fast.

When you are reading, this means that page replacement has started, you do
not have all data available in RAM and you have to read pages from disk
just to update them and put back.

This may be further compound by under-inlined SQL indexes, so please check
that you don't have any. We have a warning about it in Ignite 2.7.

Regards,
-- 
Ilya Kasnacheev


чт, 5 дек. 2019 г. в 05:06, krkumar24061...@gmail.com <
krkumar24061...@gmail.com>:

> When we are writing the data to ignite thru streamer ( key, value) and
> Ignite
> JDBC ( into couple of SQL tables) we get very high throughput when read
> IOPS
> are low and Write IOPS are high and we get very low throughput when Reads
> and Writes are competing. So my question when I am writing to the Ignite
> which is mostly write intensive operation, why are the read IOPS are going
> up. I understand you much be reading some ( meta data and indexes ) but it
> gets to a point where it starts competing with writes and grabs almost
> half.
> I am attaching the graph which gives you some insights into the write/read
> pattern. I am curious what is causing the read IOPS go up and how can we
> avoid if possible ? Let me know what you guys think??
>
> In the diagram attached take a look blue ( write  ) and yellow ( read )
> IOPS
>
> <
> http://apache-ignite-users.70518.x6.nabble.com/file/t2578/read-write-grapha.png>
>
>
>
> Thanx and Regards,
> KR Kumar
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>

Understanding Write/Read IOPS during Ignite data load through Streamer and JDBC SQL

[krkumar24061...@gmail.com]

When we are writing the data to ignite thru streamer ( key, value) and Ignite
JDBC ( into couple of SQL tables) we get very high throughput when read IOPS
are low and Write IOPS are high and we get very low throughput when Reads
and Writes are competing. So my question when I am writing to the Ignite
which is mostly write intensive operation, why are the read IOPS are going
up. I understand you much be reading some ( meta data and indexes ) but it
gets to a point where it starts competing with writes and grabs almost half.
I am attaching the graph which gives you some insights into the write/read
pattern. I am curious what is causing the read IOPS go up and how can we
avoid if possible ? Let me know what you guys think??

In the diagram attached take a look blue ( write  ) and yellow ( read ) IOPS


 


Thanx and Regards,
KR Kumar



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/