[JVM] G1 garbage collector

Tips ： When the article is finished , Directories can be generated automatically , How to generate it, please refer to the help document on the right

---

Why not use CMS?

     Previous introduction CMS The purpose is to minimize the response pause time , Is a milestone in the garbage collector ,CMS It is divided into Background and Foreground, Background Is normal mode , The whole recovery process is divided into 4 Step ： Initial marker , Concurrent Tags , Re label , Concurrent elimination , In order to avoid the large amount of data for full scan , After the concurrent marking process Eden Area capacity reaches 2M, There is a pre marking phase , When the occupied space reaches Eden Of 50% Terminate this phase and proceed to retag ;Foreground Error environment entered for concurrency failure , The execution status is single thread .
     Even though CMS The response pause time is very good , But throughput is ignored , In single thread or dual thread state CMS Is also in poor condition , Because when two threads are concurrent , It is divided into a business thread and a garbage collection thread , Also exist CPU The scheduling of results in frequent context switching , So the efficiency loss problem .
secondly , In the terminable pre marking phase , If it doesn't happen minor GC,CMS The limit time of pause is 5s; And enter foreground Also said CMS Need to carry out Full GC;
Sum up ：CMS The drawback is that his concurrency fails , Unpredictable pauses and low throughput .

---

G1 summary

    G1 Purpose ：Garbage First, That is, the garbage priority principle , That is, the focus of space . At the same time, the pause time and throughput are taken into account .
G1 Redefine memory space , No longer divided by physical space Old and Young, Instead, the memory is logically divided into several Region, Default in memory Region by 2048 individual , Every Region The size is 1M~32M

Objects are divided into ： Free object （Empty）, The new generation （Eden, Survivor）,Old Zones and large objects
Region For their own role is not fixed , The next time the space is allocated, the role will start from Old Turn into E Or other

JDK11 Then there is a special kind of partition , It is called archive partition , Close the archive partition and open the archive partition

    TLAB： Thread allocates memory locally , stay G1 When allocating object memory in , because G1 Is a concurrent garbage collector , At the same time, high concurrency , This involves the problem of data concurrency security, which requires a global lock to ensure data security , But it will bring about a decline in efficiency .
    TLAB yes Eden A part of an area on an area （1%）, When allocating memory, first judge TLAB Can I assign , If not , Will choose TLAB External Region Distribution of areas , If there is still insufficient space to proceed Full GC.

Other concepts

RSet： Application set , yes G1 The realization of middle pair and memory set , analogy CMS Card table in .
     stay G1 in , Suppose there is A Region, Rset Will record other Region perform A The record of ;
Q： What are the advantages of doing so ？
A： Suppose if there is O The zone object points to Y District （ It feels like the opposite is true ）, When recycling the new generation , Also need to O The region is traversed once , Indirectly, it becomes a full scan , And by Rset You can record ： Set here Y The object of the zone is A, therefore Rset The record in is O Zone reference , All you need to do is scan Rset You can know all the references , No full scan is required .

The part that has not been studied in depth ： Sparse table , Coarse grained bitmap , Fine grained bitmap

---