ES 1.7.2 frequent and long old GC Problem

I'm using Elasticsearch 1.7.2. My cluster is composed by 10 data node. Recently, the cluster emerge frequent and long gc.

The gc time graph is below(red is old gc,green is young gc)

The old generation memory percentage graph is below

The filter eviction count is below

I use the default gc configuration.

My question is

  1. Is there a relation between filter eviction and gc frequency?

  2. Why gc happens when the old gc memory does not reach 75 percent? In the second graph, between 08:15 and 11:00, there are also many old gc.

up up up
any advice?