Hi Kumar,

Ignite need to sync memory and disk time to time, if you write rate is
bigger than disk speed that means Ignite will never able to sync memory and
disk, that's why it starts to throttle your writes.

So ignite added 1.6ms delay to each write, to be more accurate, not for each
writes, but for writes during checkpoint and when throttling is required:
https://cwiki.apache.org/confluence/display/IGNITE/Ignite+Persistent+Store+-+under+the+hood#IgnitePersistentStore-underthehood-Checkpointing
 
and yes on average this delay is 54% of the whole update time(including
delay itself).

You can disable throttling, which I don't recommend, because then, instead
of throttling updates, ignite can completely stop all updates during a
checkpoint.

There are several options you can try, first use different physical disks
for WAL/WAL Archive and Storage.(WAL and WAL archive should on the same
disk). If it won't help then it's time to scale.

So if it's AWS like environment, I would recommend using IO1 disks with high
IOPS to make sure that it can handle your peak load. However, IO1 disks it's
pretty expensive might be it would be more efficient to just add a few more
nodes.

Again, if it's just initial data load, then it can be fine to have
throttling because usually, initial data load creates load significantly
bigger then regular load.

Thanks,
Mike.




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

Reply via email to