diff options
Diffstat (limited to 'docs/tutorials/013/page07.html')
| -rw-r--r-- | docs/tutorials/013/page07.html | 257 |
1 files changed, 0 insertions, 257 deletions
diff --git a/docs/tutorials/013/page07.html b/docs/tutorials/013/page07.html deleted file mode 100644 index d2e22b2bdde..00000000000 --- a/docs/tutorials/013/page07.html +++ /dev/null @@ -1,257 +0,0 @@ -<!-- $Id$ --> -<HTML> -<HEAD> - <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1"> - <META NAME="Author" CONTENT="James CE Johnson"> - <TITLE>ACE Tutorial 013</TITLE> -</HEAD> -<BODY TEXT="#000000" BGCOLOR="#FFFFFF" LINK="#000FFF" VLINK="#FF0F0F"> - -<CENTER><B><FONT SIZE=+2>ACE Tutorial 013</FONT></B></CENTER> - -<CENTER><B><FONT SIZE=+2>Multiple thread pools</FONT></B></CENTER> - - -<P> -<HR WIDTH="100%"> -<P> -I've been trying to justify the chain of tasks by talking about a -Work object that implements a state machine. The idea is that your -Work object has to perform a series of discrete steps to complete it's -function. Traditionally, all of those steps would take place in one -thread of execution. That thread would probably be one from a Task -thread pool. -<P> -Suppose, however, that some of those steps spend a lot of time waiting -for disk IO. You could find that all of your thread-pool threads -are just sitting there waiting for the disk. You might then be -tempted to increase the thread pool size to get more work through. -However, if some of the stages are memory intensive, you could run out -of memory if all of the workers get to that state at the same time. -<P> -One solution might be to have different thread pools for each state. -Each pool could have it's size tuned appropriately for the work that -would be done there. That's where the chain of Tasks comes in. - In this tutorial's implementation I've taken the -easy route and set all of the thread pools to the same size but a more -realistic solution would be to set each thread pool in the chain to a -specific size as needed by that state of operation. -<P> -There's not much to this header either so I've combined it with the -cpp file as with task. -<P> -<HR WIDTH="100%"> -<HR width=50%><P><center>work.h</center><HR width=50%> -<PRE> - -<font color=red>// $Id$</font> - -<font color=blue>#ifndef</font> <font color=purple>WORK_H</font> -<font color=blue>#define</font> <font color=purple>WORK_H</font> - -<font color=blue>#include</font> "<A HREF="../../../ace/Log_Msg.h">ace/Log_Msg.h</A>" - -<font color=blue>#if !defined</font> (<font color=purple>ACE_LACKS_PRAGMA_ONCE</font>) -<font color=blue># pragma</font> <font color=purple>once</font> -<font color=blue>#endif</font> <font color=red>/* ACE_LACKS_PRAGMA_ONCE */</font> - -<font color=blue>#include</font> "<A HREF="../../../ace/Synch.h">ace/Synch.h</A>" -<font color=blue>#include</font> "<font color=green>mld.h</font>" - -<font color=red>/* - Our specilized message queue and thread pool will know how to do "<font color=green>work</font>" on - our Unit_Of_Work baseclass. - */</font> -class Unit_Of_Work -{ -public: - Unit_Of_Work (void); - - virtual ~ Unit_Of_Work (void); - - <font color=red>// Display the object instance value</font> - void who_am_i (void); - - <font color=red>// The baseclass can override this to show it's "<font color=green>type name</font>"</font> - virtual void what_am_i (void); - - <font color=red>// This is where you do application level logic. It will be</font> - <font color=red>// called once for each thread pool it passes through. It</font> - <font color=red>// would typically implement a state machine and execute a</font> - <font color=red>// different state on each call.</font> - virtual int process (void); - - <font color=red>// This is called by the last Task in the series (see task.h)</font> - <font color=red>// in case our process() didn't get through all of it's states.</font> - virtual int fini (void); - -protected: - ACE_Atomic_Op < ACE_Mutex, int >state_; - MLD; -}; - -<font color=red>/* - A fairly trivial work derivative that implements an equally trivial state - machine in process() - */</font> -class Work : public Unit_Of_Work -{ -public: - Work (void); - - Work (int message); - - virtual ~ Work (void); - - void what_am_i (void); - - int process (void); - - int fini (void); - -protected: - int message_; - MLD; -}; - -<font color=blue>#endif</font> -</PRE> -<HR width=50%><P><center>work.cpp</center><HR width=50%> -<PRE> - -<font color=red>// $Id$</font> - -<font color=blue>#include</font> "<font color=green>work.h</font>" - -<font color=red>/* - Initialize the state to zero - */</font> -<font color=#008888>Unit_Of_Work::Unit_Of_Work</font> (void) - : state_ (0) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Unit_Of_Work ctor\n</font>", (void *) this)); -} - -<font color=#008888>Unit_Of_Work::~Unit_Of_Work</font> (void) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Unit_Of_Work dtor\n</font>", (void *) this)); -} - -<font color=red>/* - Display our instance value - */</font> -void <font color=#008888>Unit_Of_Work::who_am_i</font> (void) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Unit_Of_Work instance\n</font>", (void *) this)); -} - -<font color=red>/* - Dispay our type name - */</font> -void <font color=#008888>Unit_Of_Work::what_am_i</font> (void) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x I am a Unit_Of_Work object\n</font>", (void *) this)); -} - -<font color=red>/* - Return failure. You should always derive from Unit_Of_Work... - */</font> -int <font color=#008888>Unit_Of_Work::process</font> (void) -{ - return -1; -} - -<font color=red>/* - ditto - */</font> -int <font color=#008888>Unit_Of_Work::fini</font> (void) -{ - return -1; -} - -<font color=red>/* - Default constructor has no "<font color=green>message number</font>" - */</font> -<font color=#008888>Work::Work</font> (void) - :message_ (-1) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Work ctor\n</font>", (void *) this)); -} - -<font color=red>/* - The useful constructor remembers which message it is and will tell you if - you ask. - */</font> -<font color=#008888>Work::Work</font> (int message) - : message_ (message) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Work ctor for message %d\n</font>", (void *) this, message_)); -} - -<font color=#008888>Work::~Work</font> (void) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Work dtor\n</font>", (void *) this)); -} - -<font color=red>/* - This objects type name is different from the baseclass - */</font> -void <font color=#008888>Work::what_am_i</font> (void) -{ - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x I am a Work object for message %d\n</font>", (void *) this, message_)); -} - -<font color=red>/* - A very simple state machine that just walks through three stages. If it is - called more than that, it will tell you not to bother. - */</font> -int <font color=#008888>Work::process</font> (void) -{ - switch (++state_) - { - case 1: - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Stage One\n</font>", (void *) this)); - break; - case 2: - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Stage Two\n</font>", (void *) this)); - break; - case 3: - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x Stage Three\n</font>", (void *) this)); - break; - default: - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x No work to do in state %d\n</font>", - (void *) this, state_.value ())); - break; - } - return (0); -} - -<font color=red>/* - If you don't have enough subtasks in the chain then the state machine won't - progress to the end. The fini() hook will allow us to recover from that by - executing the remaining states in the final task of the chain. - */</font> -int <font color=#008888>Work::fini</font> (void) -{ - while (state_.value () < 3) - { - ACE_DEBUG ((LM_DEBUG, "<font color=green>(%P|%t) 0x%x <font color=#008888>Work::fini</font>() state %d\n</font>", (void *) this,state_.value())); - if (this->process () == -1) - { - ACE_ERROR_RETURN ((LM_ERROR, "<font color=green>%p\n</font>", "<font color=green>process</font>"), -1); - } - } - return (0); -} -</PRE> -<HR> -<P> -And that is that. For a more complex machine that may want to "jump -states" you would have to set some "state information" (sorry, bad -choice of terminology again) so that process() could decide what to do -at each call. You might also modify Task::svc() so that it will -respect the return value of process() and do something useful with the -information. -<P> -<P><HR WIDTH="100%"> -<CENTER>[<A HREF="../online-tutorials.html">Tutorial Index</A>] [<A HREF="page08.html">Continue This Tutorial</A>]</CENTER> |