System Zing Memory

Pause Prevention Memory

For other types of JVMs, when object allocation rates suddenly increase, the garbage collector can have a difficult time keeping up with the increased rate of object creation. The result is that other JVMs need to pause all work in the process to do a garbage collection. This causes Java application pauses. A common work-around to minimize the length of garbage collection pauses is to specify a small Java heap size – the theory being that a smaller Java heap requires less time to collect. Unfortunately, when the rate of work goes up, more live objects are in the heap when the heap fills up and the amount of time required for the garbage collection is proportional to the number of live objects in the Java heap. The pauses are still long.

As part of the configuration of System Zing Memory, the system-config-zing-memory tool sets aside a percentage of the System Zing Memory for Pause Prevention memory. When there is a sudden increase in the object allocation rate, the Zing instance can use the Pause Prevention memory to efficiently keep up with and sustain the throughput during the heavy, unexpected workloads. Using memory from Pause Prevention memory gives the Zing instance’s Continuously Concurrent Compacting Collector additional time to adjust to and keep up with suddenly increased allocation rates. This ensures that the garbage collector does not stop or pause a Zing instance running a Java application.

Having Pause Prevention memory ensures that:

Estimating Zing Memory Allocation Example

The percentages used in the following example are based on the default values used by the system-config-zing-memory tool. The key relationship to remember is that the amount of Reservable memory in the system must always be greater than or equal to the sum of the –Xmx values of all concurrently running Zing instances on the system.

The maximum reservable amount of memory can be found using:

We need to calculate the amount of memory available for System Zing Memory.

 
SystemLinuxMemory_InitiallyConfigured = HostSystemMemory_Total - Reservable_UpperLimit

 
SystemLinuxMemory_UseAboveInitiallyConfigrued =

     SystemLinuxMemory_Used - SystemLinuxMemory_InitiallyConfigured

 
if (SystemLinuxMemory_UseAboveInitiallyConfigrued <= 0)

   ReservableMemory = Reservable_UpperLimit - Reservable_Used

else

   We have less memory available than the value of Reservable_UpperLimit

   ReservableMemory = (Reservable_UpperLimit -

      SystemLinuxMemory_UseAboveInitiallyConfigrued) -

         Reservable_Used

ReservableMemory

The difference between the two formulas arises because the total reservable is fixed and pre-reserved when the reserve-at-config policy is used while there is an upper limit, but no pre-reserved memory, for System Zing Memory when reserve-at-launch policy is used. If using the reserve-at-launch policy, the Host System memory may not be available to use as System Zing Memory because it is already in use as System Linux memory.

When estimating the amount of System Zing Memory to allocate, conscider:

System Zing Memory = Reservable memory Contingency memory Pause Prevention memory

Reservable memory = System Zing Memory – (Contingency memory Pause Prevention memory)

Maximum Zing instance heap memory (-Xmx) = System Zing Memory reserved for the Zing process use.

Unused Reservable memory = Reservable memory – Sum of Java VM Java Heap memory (-Xmx) values for all currently running Zing instances

Generally, to estimate the total amount of System Zing Memory needed, for each Zing instance that will be running concurrently, add the amount of Java heap memory (-Xmx) required; then add the memory required for Pause Prevention and Contingency memory (together, these are 10% of total System Zing Memory).

For example:

If you typically run 5 Zing instances and each will reserve 20 GB of memory using the –Xmx command line option then you need a minimum of 111 GB of System Zing Memory. 100 GB is used by all the Zing instances and 11 GB is used for the Pause Prevention and Contingency memory.

The following is an approximate breakdown of how the system’s memory is used when there is a single Zing instance running.

100% Host System memory is shared between:

For example, if total host system memory is 100 GB:

Every Zing instance running a Java application reserves memory from Reservable memory based on the value of the -Xmx command line option.