Read IOPS are higher than write when all we are doing is write

classic Classic list List threaded Threaded
4 messages Options
krkumar24061975@gmail.com krkumar24061975@gmail.com
Reply | Threaded
Open this post in threaded view
|

Read IOPS are higher than write when all we are doing is write

Hi - I have a simple cache object which has a signature of <Long, byte[]> and I am writing events through streamer. Though it's just a write operation throughout the test, After sometime, read IOPS have gone significantly higher and they are actually greater than write IOPS. So my question is what ignite reading and why it's reading anything when all I am doing is writing data to ignite. Ignite native persistence is enabled and the disk that I am using is gp2 at 16000 IOPS in AWS environment. Any pointers??

Here is the visual snapshot of the IOPS on one of the servers:

read-iops.png

Thanx and Regards,
KR Kumar
vdpyatkov vdpyatkov
Reply | Threaded
Open this post in threaded view
|

Re: Read IOPS are higher than write when all we are doing is write

Hi,

After region capacity exhausted (DataRegionConfiguration#setMaxSize) a page replacement is started.
There is a read-load from reading the pages from a disc, when you update some key or try to write into evicted page was not fully filled.

On Fri, Sep 4, 2020 at 5:55 PM KR Kumar <[hidden email]> wrote:
Hi - I have a simple cache object which has a signature of <Long, byte[]> and I am writing events through streamer. Though it's just a write operation throughout the test, After sometime, read IOPS have gone significantly higher and they are actually greater than write IOPS. So my question is what ignite reading and why it's reading anything when all I am doing is writing data to ignite. Ignite native persistence is enabled and the disk that I am using is gp2 at 16000 IOPS in AWS environment. Any pointers??

Here is the visual snapshot of the IOPS on one of the servers:

read-iops.png

Thanx and Regards,
KR Kumar


--
Vladislav Pyatkov
krkumar24061975@gmail.com krkumar24061975@gmail.com
Reply | Threaded
Open this post in threaded view
|

Re: Read IOPS are higher than write when all we are doing is write

Hi Vlad - Thanks for the response.

In this test, we write once and no reads/updates throughout the process.
Would that still result into reading of pages. Also, are you telling that
Ignite writes/evicts half filled pages to disk and read them back later when
you have to append a key/value to the page??

Thanx and Regards,
KR Kumar



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
vdpyatkov vdpyatkov
Reply | Threaded
Open this post in threaded view
|

Re: Read IOPS are higher than write when all we are doing is write

Yes, it can be. More over, indexes pages also will be able to be evicted and after it will be required to return them for updating when insert a new entry.

On Sat, Sep 5, 2020 at 6:39 PM [hidden email] <[hidden email]> wrote:
Hi Vlad - Thanks for the response.

In this test, we write once and no reads/updates throughout the process.
Would that still result into reading of pages. Also, are you telling that
Ignite writes/evicts half filled pages to disk and read them back later when
you have to append a key/value to the page??

Thanx and Regards,
KR Kumar



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


--
Vladislav Pyatkov