High-units watermark reached but cache does not shrink back to low-units mark

Question

Here is our config:
    <caching-scheme-mapping>
        <cache-mapping>
            <cache-name>Cache2</cache-name>
            <scheme-name>Scheme2</scheme-name>
        </cache-mapping>
    </caching-scheme-mapping>
    <caching-schemes>
        <distributed-scheme>
            <scheme-name>Scheme2</scheme-name>
            <service-name>Service2</service-name>
            <backing-map-scheme>
                <local-scheme>
                    <eviction-policy>LRU</eviction-policy>
                    <high-units>700m</high-units>
                    <unit-calculator>BINARY</unit-calculator> 
                    <expiry-delay>2h</expiry-delay>
                </local-scheme>
            </backing-map-scheme>
            <autostart>true</autostart>
        </distributed-scheme>
We're running on two storage 4 GB JVMs.
We run a load test that will eventually make caches reach 700 MB (734003200 kb) on each JVM/node.
When the high-unit is reached (as advertized in JMX console), prune is triggered (CachePrunes is incremented in JMX console) but cache size (Units) stays around limit, never falls to (default) low-units mark (587202560 kb).
Is this normal behaviour?  I thought when high-units mark was reached, size was brought down to low-units mark.

Answer 1

I think that it is something wrong with kernel and graphic drivers but i'm not able to solve it from the beginning of using of Arch by me.
I'm not so sure this issue is software-related only because in my case it seems the laptop screen enters power-saving mode (10-min timeout) on its own without any events logged (as shown by dmesg) so my next thought was OK, I'm going to get rid of this issue disabling the screen power-saving timeout in the BIOS and set it to never power down or something like that and from then on managing power mode manually; ie: when I want to power down I will just use systemctl suspend/hibernate which work fine ... problem is: there are no related options in my BIOS for such timeout.
The only timeouts I have are in /etc/systemd/logind.conf set to 30min (#IdleActionSec) (commented) by default so these are not in effect because the screens goes blank after 10m which is less than the (default) 30m in logind. Maybe setting loggind to 9m (un-commented) to catch the event before the machine-default one ?
IdleActionSec=9min
IdleAction=suspend
Will check it and report back.