I can see the image but unable to read it properly as to what is on X and Y axis. I am assuming the frequency of GC is increasing after sometime of JVM running. This could be because you getting more load at that time. If you have switched on GC logs then you can check if JVM have changed the GC strategy in middle.
The problem seems to be that the system can't recover sufficient memory even after a full garbage collection. This also has the negative effect of promoting objects to tenured and survivor spaces that don't really belong there.
The size of the old generation should settle down over time, most objects are (or at least should be) very short lived. If you find that the old generation keeps growing then there may be a resource leak somewhere. From your graph it seems like some allocation is tipping the system over the edge and into a gc death spiral.
Try to get the an idea of what objects are being created, and, more importantly, how long they are hanging around. This will give you the best idea of where you should be looking.