Databases
StackOverflow2013
Postsdbo
PO
 

Note that there are some explanatory texts on larger screens.

plurals
  1. PO
    primarykey
    data
    text
    <p>Are you sure it's complete code? I don't see where you're creating threads or where BuildKCData is being called from.</p> <p>You ought to have a memory barrier after pthread_kill(), even though I doubt it makes a difference in this case.</p> <p><strong>EDIT:</strong> You're confusing in-order execution and cache-consistency.</p> <p><strong>Cache consistency:</strong> x86 (currently) guarantees that aligned 4-byte accesses are atomic, so <code>a[0]=123</code> in thread A and <code>a[1]=456</code> in thread B will work — thread C will eventually see "123,456". There are various cache-consistency protocols around, but I believe it's approximately a MRSW lock.</p> <p><strong>Out-of-order execution:</strong> x86 doesn't guarantee ordering of reads (and possibly writes; there was a debate about whether sfence was needed in the linux kernel). This lets the CPU prefetch data more effectively, but it means <code>a[0]=123,a[1]</code> in thread A and <code>a[1]=456,a[0]</code> in thread B could both return 0, because the fetch of a[1] can happen before the load of a[0]. There are two general ways to fix this:</p> <ul> <li>Only access shared data when you're holding the lock. In particular, do not read shared data outside of the lock. Whether this means a lock for each entry or a lock for the entire array is up to you, and what you think lock contention is likely to be like (tip: it's not usually very big).</li> <li>Stick a memory barrier between things that need to be in order. This is difficult to get right (pthread doesn't even have memory barriers; <code>pthread_barrier</code> more like a sync point).</li> </ul> <p>While memory barriers are a recent trend, locking is <em>far</em> easier to get right (<em>I am holding the lock, therefore nobody else is allowed to change the data under my feet</em>). Memory barriers are all the rage in some circles, but there's a lot more to get right (<em>I hope this read is atomic</em>, <em>I hope other threads write atomically</em>, <em>I hope other threads use a barrier</em>, and <em>oh yeah, I need to use a barrier too</em>).</p> <p>And if locking's too slow, reducing contention will be much more effective than replacing locks with barriers and hoping you got it right.</p>
    singulars
    1. This table or related slice is empty.
    1. POa simple boss-worker model using pthreads
      singulars
      1. PTQuestion
    1. PTAnswer
    1. UStc.
    1. UStc.
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. VO
      singulars
      1. PO
      1. This table or related slice is empty.
      1. VTUpMod
    1. COWhat do you mean by "memory barrier"? I'm not sure what you're getting at -- if you tell me what you mean I'd be happy to try it. BuildKCData is being called, it's in the while loop that ends with comp_START++. Reading from top to bottom, it's the last while loop before the std::cout<<" .. sorting .. "; call.
      singulars
      1. PO
      1. USRyan Budney
    2. COOh, now I see what you mean. There's some kind of formatting issue. The code is there, but I think angle brackets are treated strangely in a code snippet. I'll see if I can fix that.
      singulars
      1. PO
      1. USRyan Budney
    3. COIt seemed the angle brackets for things like template calls, less than symbols and such were causing the code formatting to go haywire. I've more or less fixed it but you'll have to imagine things like "leq" mean "<" etc.
      singulars
      1. PO
      1. USRyan Budney
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload