what is things?

For my reference 
NOTE: MAY NOT CORRECT.
HPX - Modern C++ Parallel runtime System and supports distributive computing across the cluster, based on Parallex execution model 
uses

Local control objects - Main part of hpx-runtime system because it provides the facilities for things to express in hpx and also to 
represent synchronization between million of user-level threads(hpx-threads) it mainly supports the future and continuation via .then()
and composition between futures like when_all, when_any, dataflow etc.

AGAS-Active Global Address Space - Single common address space over the cluster of nodes(include GPU not sure) instead of partitioned 
global address space, where each node have their own address space and the runtime system which uses PGAS, collects the whole node address
space under the hood, this increases the necessary to keep the some sort of table for making address translation while 
int a = 12;
int * ptr = &a;
move(a 2  loc2)
will the ptr hold the address of valid location ? 

Parcel port layers - parcels are the active messages which are sent to localities, Mostly it is the pack contains targets(execution site-id,
function to execute and its arguments, possibly continuation if appicable) they are sent to execution sites without any sychronization
between sites HPX currently uses OPENMP, MPI as backend it is also possible to use some other parcelport such as LIBFABRIC.

Components are objects in HPX which can be created on the execution site's memory. They follow curiously recursive template pattern.
They are created on execution site memory and their client objects can be created in the calling  execution site memory space.
All these have address tags attached with active global address space, so the client can be created on wide number of execution sites,
it way more natural,Isn't ?

Threading subsystem - HPX doesn't have pre-emption does it sounds weird ? Wait HPX really mean it, Hpx-threads are really small at most
run for microseconds less than 10,so threads have very very small less execution time and smaller context switch time. Everything is 
thread all piece of code in HPX is a hpx-thread, that is piece of code is wrapped inside a thread and gets into task queue and
gets scheduled once if precondition are met( whether all I/O ops are done and all data dependency are notified). HPX threads got 
scheduled in one OS thread per one Core, and the thread affinity can be made through hpx-bind & it's friend's like(scattered, balanced)
one queue steals from other queue's if it is empty but it not steal from other execution sites.

Parallel Algorithms
HPX implements in parallel algo's specified in the standard c++, with execution policy and executors. Parallelism TS is a good read to go
with this.

Serialization - Bits sent across wire ! No idea what mechanism HPX is using !
HPX.Compute   -