Hi there,
I have a question for dynamic snitch, specifically in reading.
For dynamic snitch, it is wrapped with other snitches. For reading, dynamic
snitch plays a very important role, as mentioned in this article:
http://www.datastax.com/dev/blog/dynamic-snitching-in-cassandra-past-present-and-future
But I do feel confused by the update interval in it. The default dynamic
update interval is 100 ms. Does it means every 100 ms it'll calculate the score
based on the history latency. But we know that for each read request, the
process/service time is much less than 100ms. So during this 100ms, there's
tons of read requests being processed, where each request may have a latency
record.
So my questions are: 1. For the history latency, is it based on all
latency information of the past 100 ms, or only a latest few samples? 2.
After the score calculation, does it means for the lowest score/node, it will
be choose as the replica to read from in the next 100ms. Then after next 100ms,
the score will be recalculated and repeat the whole process?
Any comment would be appreciated. Thanks in advance!
Jun
