CS452 - Real-Time Programming - Spring 2011

Lecture 26 - Pathologies

Pubilic Service Announcement

1. Project proposals.
2. Train availability.
3. Second train control demo. `I must have been insane.'

Pathologies

As we go down this list both pathology detection and the time of an edit-compile-test cycle grow without bound.

1. Deadlock

One or more tasks will never run again. For example

1. Task sends to itself (local: rest of system keeps running, task itself will never run)
2. Every task does Receive( ) (global: nothing is running)
3. Cycle of tasks sending around the cycle (local: other tasks keep running)

Kernel can detect such things

• What does it do?

Potential deadlock can be detected at compile time

• cycle in the send graph of all sends that could happen
  • remember implicit sends
• doesn't necessarily occur at run-time
  • that is, it's a necessary but not sufficient condition
• Changes in a critical race can make a potential deadlock reveal itself.

Solutions

• Gating
  • Most common example is initialization, where the send/receive pattern may be different than FOREVER
  • Gate the end of initialization
• Define three types of task
  • Administatrer (A): only receives
  • Worker (W): only sends
  • Client (C): only sends
  • Two A tasks cannot communicate directly; two W/C tasks cannot communicate directly.
  • For W/C tasksSend appears in two flavours
    • C tasks

      FOREVER {
          Send( A, request, result )
          ...
      }

    • W tasks

      FOREVER {
          Send( A, result, request )
          ...
      }

      • The corresponding Receives are also different.
      • Data types of request and result must be compatible!
    • A courier converts one type to another

      FOREVER {
          Send( A1, request, result )
          Send( A2, result, request )
      }

2. Livelock (Deadly Embrace)

Definition

Two or more tasks are READY. For each task, the state of the other tasks prevents progress being made regardless of which task is ACTIVE.

A higher level of coordination is possible.

Two types of livelock exist

1. Ones that are the result of bad coding
   • Find bug and remove it.
2. Ones that are inherent in the application definition
   • Detect livelock and work around it.

Looking for solutions we prefer ones that avoid the central planner

Usually occurs in the context of resource contention

Livelock that's Really Deadlock

• client1 needs resource1 & resource2;
  • obtains resource1 from proprietor1;
  • asks proprietor2 for resource2
• client2 needs resource1 & resource2;
  • obtains resource2 from proprietor2;
  • asks proprietor1 for resource1
• possible code
  • Client 1

    Send( prop1, getres1, ... );
    Send( prop2, getres2, ... );
    // Use the resources and release them

  • Client 2

    Send( prop2, getres2, ... );
    Send( prop1, getres1, ... );
    // Use the resources and release them

  • Proprietor

    FOREVER {
       Receive( &clientTid, req, ... );
       switch ( req-type ) {
       case REQUEST:
          if( available ) { 
             Reply( clientTid, use-it, ... );
             available = false;
          }
          else enqueue( clientTid );
       case RELEASE:
          available = true;
          Reply( clientTid, "thanks", ... );
          if( !empty( Q ) ) {
             available = false;
             Reply( dequeue( ), use-it, ... );
          }
       }
    }

• state:
  • client1, client2: REPLY-BLOCKED - can't release resources
  • proprietor1, proprietor2: SEND-BLOCKED - waiting for release
  • this is a true deadlock, but there are no cycles in the call graph. The dependencies lie elsewhere. (You can find on the internet arguments about terminology just as intense as anything you will ever see in vi vs emacs or Apple vs Microsoft.)

Solutions

1. Make a single compound resourse, BUT
   • all clients may not need both
   • some resources simply cannot be compounded
2. Impose a global order on resource requests that all clients must follow.
   • unsafe against malicious or incompetent programmers
   • some resources don't admit strong enough ordering
3. Create a mega-server that handles all resource requests
   • clients request all at once
   • client might not know that A is needed until processing with B is well-advanced

Real Livelock

Proprietor1 & proprietor2 fail the requests

• Proprietor

  FOREVER {
     Receive( &clientTid, req, ... );
     switch ( req-type ) {
     case REQUEST:
        if( available ) { 
           Reply( clientTid, use-it, ... );
           available = false;
        }
        else Reply( clientTid, "sorry", ...);
     case RELEASE:
        available = true;
        Reply( clientTid, "thanks", ... );
     }
  }

• Polling is the most likely result. Typical client code.

  while ( Send( prop1, getr1, ... ) != GotIt ) ; 
  while ( Send( prop2, getr2, ... ) != GotIt ) ; 
  // Use the resources

Livelock that's Really a Critical Race

We could try to make the clients a little more considerate

While ( no resources ) {
   Send( prop1, get1es1, result );
   while ( result == "sorry" ) {
   if ( result == "sorry" ) {
      Delay( ... );
      Send( prop1, getres1, result );
   }
   Send( prop2, getres2, result );
   if ( result == "sorry" ) {
      Send( prop1, relres1, ... );
      Delay( ... );
   } else {
      break; 
   }
}

Inherent Livelock

Remember the example where two trains come face to face, each waiting for the other to move. They will wait facing one another until the demo is over, probably polling.

What's hard about solving this problem?

• Neither engineer knows what the other engineer is trying to do.

In real life,

• the engineers would communicate, but
• in your software that's neither easy nor desirable

What's most easy for you to do is to programme each engineer with

1. detection, e.g.,
   • Delay some time
   • Request again
   • If turned down, work around
2. work around, e.g.,
   • Recommence working on goal as though track is blocked.

3. Critical Races

Example

1. Two tasks, A & B, at the same priority
2. A is doing a lot of debugging IO
3. B always reserves a section of track before A, and all is fine.
4. Debugging IO is removed
5. A reserves the section before B can get it, and execution collapses.
6. Lower priority of A to the same level as C.
7. Now C executes faster and gets a resource before D .
8. You shuffle priorities forever, eventually reverting, to put back in the debugging IO.

Definition

The order in which computation is done is an important factor in determining whether or not it is successful.

Critical races, like Livelock can be

• internal to the program, like the one above, or
• external to the program but inherent in the application

Symptoms

1. Small changes in priorities change execution unpredictably, and drastically.
2. Debugging output changes execution drastically.
3. Changes in train speeds change execution drastically.
   • Example from two terms ago

`Drastically' means chaos in both senses of the term

1. Sense one: a small change in the initial conditions produces an exponentially growing change in the system
2. Sense two: exercise for the reader.

Solutions

1. Explicit synchronization
   • but you then have to know the orders in which things are permitted to occur
2. Gating is a technique of global synchronization
   • which can be provided by a detective/coordinator

4. Performance

The hardest problem to solve

• You just don't know what is possible
• Ask a question like:
  • Is my kernel code at the limit of what is possible in terms of performance?
  • We can compare the performance on message passing, etc., because two kernels are pretty much the same.
    • Compare a lot of kernels and you should be able to find a lower limit
  • Can't do the same thing for train applications

Priority

The hardest thing to get right

• Sizing stacks used to be harder, but now we have lots of memory
  • But last term, ...
• NP-hard for the human brain
• Practical method starts with all priorities the same, then adjusts
• symptoms of good priority assignment
  • The higher priority, the more likely the ready queue is to be empty
  • The shorter the run time in practice the higher the priority

Problems with priority

1. Priority inversion
2. One resource, many clients
3. Tasks try to do too much

Congestion

1. Too many tasks
   • blocked tasks don't count,
   • lowest priority tasks almost don't count

Layered abstraction are costly

e.g. Notifier -> SerialServer -> InputAccumulater -> Parser -> TrackServer

Hardware

1. Turn on optimization, but be careful
   • There are places where you have done register allocation by hand
2. Turn on caches
   • locking is possible

Size & align calibration tables by size & alignment of cache lines

• linker command script
• I think that this is stretching it.

Return to:

• Bill Cowan's lecture notes for CS452 in s11
• Bill Cowan's Spring 2011 CS452 page
• Bill Cowan's CS452 page
• Bill Cowan's teaching pagelu