Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming.threads > #2119 > unrolled thread

More on scalability

Back to article view | Back to comp.programming.threads

Contents

  More on scalability aminer <aminer@toto.net> - 2014-03-18 22:29 -0700
    Re: More on scalability Robert Miles <milesrf@Usenet-News.net> - 2014-04-01 12:37 -0500
      Re: More on scalability Lucas Levrel <lucas.levrel@u-pec.fr> - 2014-04-02 11:46 +0200

#2119 — More on scalability

Hello,

"NUMA attempts to address this problem by providing separate memory for 
each processor"

So to  scale  Databases or AVL trees or Red Black trees up to the 
scalability upper limit of my scalable RWLocks you can use a NUMA 
architecture, NUMA architecture is like distributed databases over many 
computers but into a single computer , you can distribute your database 
in each separate memory for each processor on a NUMA architecture. So 
that the reader section of my scalable RWLock will take advantage of 
that and so that your Database scales up to the  scalability upper limit 
of my scalable RWLocks.

So now the problem is solved beautifully.


Amine Moulay Ramdane.

[toc] | [next] | [standalone]

#2133

On 3/19/2014 12:29 AM, aminer wrote:
>
> Hello,
>
> "NUMA attempts to address this problem by providing separate memory 
> for each processor"
>
> So to  scale  Databases or AVL trees or Red Black trees up to the 
> scalability upper limit of my scalable RWLocks you can use a NUMA 
> architecture, NUMA architecture is like distributed databases over 
> many computers but into a single computer , you can distribute your 
> database in each separate memory for each processor on a NUMA 
> architecture. So that the reader section of my scalable RWLock will 
> take advantage of that and so that your Database scales up to the  
> scalability upper limit of my scalable RWLocks.
>
> So now the problem is solved beautifully.
>
>
> Amine Moulay Ramdane.
>
>
If you want such a processor to work well for multi-threaded programs 
using multiple CPU cores, you'll also need to provide a section of 
memory shared among the CPU cores, to allow the cores a reasonable speed 
of communicating with each other.

[toc] | [prev] | [next] | [standalone]

#2134

Le 1 avril 2014, Robert Miles a écrit :

> On 3/19/2014 12:29 AM, aminer wrote:
>>  "NUMA attempts to address this problem by providing separate memory for
>>  each processor"
> If you want such a processor to work well for multi-threaded programs using 
> multiple CPU cores, you'll also need to provide a section of memory shared 
> among the CPU cores, to allow the cores a reasonable speed of communicating 
> with each other.

Correct me if I'm wrong, but NUMA is precisely what you say, and even 
more: all cores in all processors have access to all memory.

All cores in a given processor have uniform access to this processor's 
share of memory. They can also access the other processors' shares of 
memory, but with some penalty. Hence the N in NUMA.

-- 
LL
Eν οιδα οτι ουδεν οιδα (Σωκρατης)

[toc] | [prev] | [standalone]

Back to top | Article view | comp.programming.threads

csiph-web