Ignore:
File:
1 edited

Legend:

Unmodified
Added
Removed

  • doc/proposals/concurrency/text/future.tex

    rb3ffb61 rd67cdb7  
    55% ======================================================================
    66
    7 \section{Flexible Scheduling} \label{futur:sched}
    8 An important part of concurrency is scheduling. Different scheduling algorithm can affact peformance (both in terms of average and variation). However, no single scheduler is optimal for all workloads and therefore there is value in being able to change the scheduler for given programs. One solution is to offer various tweaking options to users, allowing the scheduler to be adjusted the to requirements of the workload. However, in order to be truly flexible, it would be interesting to allow users to add arbitrary data and arbirary scheduling algorithms to the scheduler. For example, a web server could attach Type-of-Service information to threads and have a ``ToS aware'' scheduling algorithm tailored to this specific web server. This path of flexible schedulers will be explored for \CFA.
    9 
    10 \section{Non-Blocking IO} \label{futur:nbio}
    11 While most of the parallelism tools
    12 However, many modern workloads are not bound on computation but on IO operations, an common case being webservers and XaaS (anything as a service). These type of workloads often require significant engineering around amortising costs of blocking IO operations. While improving throughtput of these operations is outside what \CFA can do as a language, it can help users to make better use of the CPU time otherwise spent waiting on IO operations. The current trend is to use asynchronous programming using tools like callbacks and/or futurs and promises\cite. However, while these are valid solutions, they lead to code that is harder to read and maintain because it is much less linear
    13 
    14 \section{Other concurrency tools} \label{futur:tools}
    15 While monitors offer a flexible and powerful concurent core for \CFA, other concurrency tools are also necessary for a complete multi-paradigm concurrency package. Example of such tools can include simple locks and condition variables, futures and promises\cite{promises}, and executors. These additional features are useful when monitors offer a level of abstraction which is indaquate for certain tasks.
    16 
    17 \section{Implicit threading} \label{futur:implcit}
    18 Simpler applications can benefit greatly from having implicit parallelism. That is, parallelism that does not rely on the user to write concurrency. This type of parallelism can be achieved both at the language level and at the library level. The cannonical example of implcit parallelism is parallel for loops, which are the simplest example of a divide and conquer algorithm\cite{uC++book}. Listing \ref{lst:parfor} shows three different code examples that accomplish pointwise sums of large arrays. Note that none of these example explicitly declare any concurrency or parallelism objects.
    19 
    20 \begin{figure}
    21 \begin{center}
    22 \begin{tabular}[t]{|c|c|c|}
    23 Sequential & Library Parallel & Language Parallel \\
    24 \begin{cfacode}[tabsize=3]
    25 void big_sum(
    26         int* a, int* b,
    27         int* o,
    28         size_t len)
    29 {
    30         for(
    31                 int i = 0;
    32                 i < len;
    33                 ++i )
    34         {
    35                 o[i]=a[i]+b[i];
    36         }
    37 }
     7Concurrency and parallelism is still a very active field that strongly benefits from hardware advances. As such certain features that aren't necessarily mature enough in their current state could become relevant in the lifetime of \CFA.
     8\section{Non-Blocking IO}
    389
    3910
     11\section{Other concurrency tools}
    4012
    4113
    42 
    43 int* a[10000];
    44 int* b[10000];
    45 int* c[10000];
    46 //... fill in a & b
    47 big_sum(a,b,c,10000);
    48 \end{cfacode} &\begin{cfacode}[tabsize=3]
    49 void big_sum(
    50         int* a, int* b,
    51         int* o,
    52         size_t len)
    53 {
    54         range ar(a, a+len);
    55         range br(b, b+len);
    56         range or(o, o+len);
    57         parfor( ai, bi, oi,
    58         [](     int* ai,
    59                 int* bi,
    60                 int* oi)
    61         {
    62                 oi=ai+bi;
    63         });
    64 }
     14\section{Implicit threading}
     15% Finally, simpler applications can benefit greatly from having implicit parallelism. That is, parallelism that does not rely on the user to write concurrency. This type of parallelism can be achieved both at the language level and at the system level.
     16%
     17% \begin{center}
     18% \begin{tabular}[t]{|c|c|c|}
     19% Sequential & System Parallel & Language Parallel \\
     20% \begin{lstlisting}
     21% void big_sum(int* a, int* b,
     22%                int* out,
     23%                size_t length)
     24% {
     25%       for(int i = 0; i < length; ++i ) {
     26%               out[i] = a[i] + b[i];
     27%       }
     28% }
     29%
     30%
     31%
     32%
     33%
     34% int* a[10000];
     35% int* b[10000];
     36% int* c[10000];
     37% //... fill in a and b ...
     38% big_sum(a, b, c, 10000);
     39% \end{lstlisting} &\begin{lstlisting}
     40% void big_sum(int* a, int* b,
     41%                int* out,
     42%                size_t length)
     43% {
     44%       range ar(a, a + length);
     45%       range br(b, b + length);
     46%       range or(out, out + length);
     47%       parfor( ai, bi, oi,
     48%       [](int* ai, int* bi, int* oi) {
     49%               oi = ai + bi;
     50%       });
     51% }
     52%
     53% int* a[10000];
     54% int* b[10000];
     55% int* c[10000];
     56% //... fill in a and b ...
     57% big_sum(a, b, c, 10000);
     58% \end{lstlisting}&\begin{lstlisting}
     59% void big_sum(int* a, int* b,
     60%                int* out,
     61%                size_t length)
     62% {
     63%       for (ai, bi, oi) in (a, b, out) {
     64%               oi = ai + bi;
     65%       }
     66% }
     67%
     68%
     69%
     70%
     71%
     72% int* a[10000];
     73% int* b[10000];
     74% int* c[10000];
     75% //... fill in a and b ...
     76% big_sum(a, b, c, 10000);
     77% \end{lstlisting}
     78% \end{tabular}
     79% \end{center}
     80%
    6581
    6682
    67 int* a[10000];
    68 int* b[10000];
    69 int* c[10000];
    70 //... fill in a & b
    71 big_sum(a,b,c,10000);
    72 \end{cfacode}&\begin{cfacode}[tabsize=3]
    73 void big_sum(
    74         int* a, int* b,
    75         int* o,
    76         size_t len)
    77 {
    78         parfor (ai,bi,oi)
    79             in (a, b, o )
    80         {
    81                 oi = ai + bi;
    82         }
    83 }
     83\section{Multiple Paradigms}
    8484
    8585
    86 
    87 
    88 
    89 
    90 
    91 int* a[10000];
    92 int* b[10000];
    93 int* c[10000];
    94 //... fill in a & b
    95 big_sum(a,b,c,10000);
    96 \end{cfacode}
    97 \end{tabular}
    98 \end{center}
    99 \caption{For loop to sum numbers: Sequential, using library parallelism and language parallelism.}
    100 \label{lst:parfor}
    101 \end{figure}
    102 
    103 Implicit parallelism is a general solution and therefore has its limitations. However, it is a quick and simple approach to parallelism which may very well be sufficient for smaller applications and reduces the amount of boiler-plate that is needed to start benefiting from parallelism in modern CPUs.
    104 
    105 
     86\section{Transactions}
Note: See TracChangeset for help on using the changeset viewer.