An article about TLAB

(From: https://blogs.oracle.com/jonthecollector/entry/the_real_thing)

 

A Thread Local Allocation Buffer (TLAB) is a region of Eden
that is used for allocation by a single thread.  It enables
a thread to do object allocation using thread local top and
limit pointers, which is faster than doing an atomic operation
on a top pointer that is shared across threads.


A thread acquires a TLAB at it's first object allocation
after a GC scavenge.  The size of the TLAB is computed via
a somewhat complex process discribed below.  The TLAB is
released when it is full (or nearly so), or the next GC scavenge
occurs. TLABs are allocated only in Eden, never from From-Space
or the OldGen.

Flags               default description

          
UseTLAB     true   Use thread-local object allocation
ResizeTLAB  false  Dynamically resize tlab size for threads
TLABSize           (below) Default (or starting) size of a TLAB (in bytes)
TLABWasteTargetPercent   1    Percentage of Eden that can be wasted
PrintTLAB              false  Print various TLAB related information

AggressiveHeap settings:

  TLABSize          256Kb
  ResizeTLAB        true     (Corrected 2007/05/09)

Minor flags

MinTLABSize             2\*K   Minimum allowed TLAB size (in bytes)
TLABAllocationWeight     35   Weight for exponential averaging of allocation
TLABRefillWasteFraction  64   Max TLAB waste at a refill (internal fragmentation)
TLABWasteIncrement        4   Increment allowed waste at slow allocation
ZeroTLAB              false   Zero out the newly created TLAB

These flags are for tuning the current implementation of
TLABs and maybe disappear or change their initial value in a
future release of the jvm.

If it is not specified on the command-line (or specified as zero)
via the -XX:TLABSize flag, the initial size of a TLAB is computed as:

  init_size = size_of_eden / (allocating_thread_count \* target_refills_per_epoch)

where:

  a) Allocating_thread_count is the expected number of threads
    which will be actively allocating during the next epoch
    (an epoch is the mutator time between GC scavenges.)
    At jvm startup this is defined to be one.  It is then
    recomputed at each GC scavenge from the number of threads
    that did at least one allocation of a tlab during the
    latest epoch. It's then exponentially averaged over the
    past epochs.
 
  b) Target_refills_per_epoch is the desired number of tlab
    allocations per thread during an epoch.  It is computed from
    the value of TLABWasteTargetPercent which is the percentage of
    Eden allowed to be wasted due to TLAB fragmentation.
 
    From a mutator thread's perspective a GC scavenge can occur
    unexpectedly at any time.  So, on average, only half of a
    thread's current TLAB will be allocated when a GC scavenge
    occurs.
 
      TLABWasteTargetPercent = 0.5 \* (1/target_refills_per_epoch) \* 100
 
    Solving for target_refills_per_epoch:
 
      target_refills_per_epoch = ( 0.5 \* 100) / TLABWasteTargetPercent
 
    With the default value of 1 for TLABWasteTargetPercent
 
      target_refills_per_epoch =  50


When TLABResize is true (which it is by default) the tlab size
  is recomputed for each thread that did an allocation in the latest
  epoch.  Threads that did not allocate in the latest epoch do not
  have their TLABs resized.  The resize goal is to get the number
  of refills closer to the ideal: target_refills_per_epoch (default
  value is 50).  For each thread, the number of refills in the latest
  epoch is exponentially averaged with values from previous
  epochs.  If this average refill number is greater than
  target refills_per_epoch, then the tlab size is increased.  If
  the average is less, the tlab size is decreased.
 
  The computation is (approximately):
 
     new_size = (old_size \* avg_refills_per_epoch) / target_refills_per_epoch
 
  It's actually computed from the fraction of the latest epoch's
  eden size used by this thread, because the next epoch may use a
  resized eden.

To experiment with a specific TLAB size, two -XX flags need
  to be set, one to define the initial size, and one to disable
  the resizing:

     -XX:TLABSize= -XX:-ResizeTLAB

  The minimum size of a tlab is set with -XX:MinTLABSize which
  defaults to 2K bytes.  The maximum size is the maximum size
  of an integer Java array, which is used to fill the unallocated
  portion of a TLAB when a GC scavenge occurs.


Diagnostic Printing Options

-XX:+PrintTLAB

  Prints at each scavenge one line for each thread (starts with "TLAB: gc thread: "
  without the "'s) and one summary line.

Thread example:

TLAB: gc thread: 0x0004ac00 [id:  2] size: 61KB slow allocs: 5  refill waste: 980B
         alloc: 0.99996     3072KB refills: 50 waste  0.1% gc: 0B slow: 4144B fast: 0B

  The tag "gc" indicates that this information was printed at a GC scavenge,
  after the tlabs have been filled. The "gc" tag doesn't mean a thread is a gc
  thread.
 
  Fields:
 
  thread:
 
     The address of the gc thread structure and it's system thread id.
 
  size:
 
     The size of the tlab in kilobytes.
 
  slow allocs:
 
     The number of allocations too large for remaining space in the TLAB.
     The allocation was done directly in eden space.
 
  refill waste: (in HeapWord units)
 
     The name is truncated in the dump, and should be: refill_waste_limit
     and is used to limit the amount of wasted space from internal
     fragmentation.  If the remaining space in the TLAB is larger than
     this amount, and an allocation is requested that is too large to
     be allocated in the TLAB, then the allocation is done directly in
     Eden and the TLAB is not retired.  If the remaining space is less
     than refill_waste_limit then the TLAB is retired, a new TLAB is
     allocated, and the object allocation is attempted in the new TLAB.

     After each allocation outside of the TLAB, the refill_waste_limit
     is incremented by TLABWasteIncrement to prevent an allocation of
     a size slightly less than refill_waste_limit from continually
     being allocated outside of the TLAB.

  alloc: [fraction]  [sizeInBytes]

     Expected amount of eden allocated by this thread computed as
     a fraction of eden and number of heap words.

  refills:

     Number of tlab refills.    

  waste [percent] gc: [bytes] slow: [bytes] fast: [bytes]

     Percentage of eden allocated to this thread that was wasted.
     Waste is the sum of three components:
        gc:   unused space in the current TLAB when stopped for a scavenge.
        slow: sum of unused space in TLABs when they're retired to allocate a new one.
        fast: the client system can allocate a TLAB with a fast allocator.
              This is the amount of waste via that method.

Summary example:

TLAB totals: thrds: 1  refills: 50 max: 50 slow allocs: 5 max 5 waste:  0.1%
             gc: 0B max: 0B slow: 4144B max: 4144B fast: 0B max: 0B

  thrds:

     Number of threads that did an allocation.
             
  refills: [tt] max: [mm]

     Total number of TLAB refills by all threads, and
     maximun number of TLAB refills by a single thread.

  slow allocs: [ss] max [mm]

     Total number of allocations done outside of a TLAB, and
     maximum number by a single thread.

  waste [percent] gc: [bytes] slow: [bytes] max: [mm] fast: [bytes] max: [mm]

     Percentage of eden that was wasted across all threads.
     Waste is the sum of three components:
        gc:   unused space in the current TLABs when scavenge starts.
        slow: sum of unused space in TLABs when they're retired to allocate a new ones.
        fast: the client system can allocate a TLAB with a fast allocator.
              This is the amount of waste via that method.

     For "slow" and "fast", the maximum value by a single thread is printed.


More detail with addition of Verbose flag.

-XX:+PrintTLAB -XX:+Verbose

  Using both: -XX:+PrintTLAB -XX:+Verbose will print the
  new tlab sizes for each thread when it is resized.  Resizing is
  only done at GC scavenges.

  Example:

TLAB new size: thread: 0x001eac00 [id: 19] refills 50  alloc: 0.402570 size: 19684 -> 18996

  New size 18996.  Previous size was: 19684.

  refills:

     Number of tlab refills for this thread.

  alloc:

     The expected fraction of eden this thread will use.