How about multicore par?

[nzb_leecher]

The reason to try out sab40 was for me to play with the new muliticore par. Twice the unpar speed sounds great. Sab40 didnt work here, so no further testing was posible. That and the fact that i cant find that news about multicore anmore on this site.

But how about you, have you installed it? How did you like it? Is it really that fast? I heard one man saying getting 40% more speed with a single core hyperthreading cpu. !!

[switch]

No speed improvements for single core pc's. A dual core pc has about 60% improvements for repairing. This is inpheaux results using the original single core par2 and two different multi-core par2 implementations:

Code: Select all

build   verify   repair   create 
-pc     4:04     10:10    34:20 
-tbb    4:02     4:06     10:59 
-ph     4:08     3:50     10:36

Test data: 70x100Mb rar set, 10% redundancy.
Test rig: Core 2 Duo E6400, 2GB Ram, SATA 3.0 hard drive.

We planned to include the multi-core par2 in beta1 but forgot to include a required dll file (tbb.dll), there were other reported needed dll files but it was not clear whether including the tbb.dll would fix that so we went back to the original par2 for now.

Should be back in for the next beta.

[ferno]

Hi,

I am curious, is this multicore par only available on the win32 version?
I am running on a Atlhon X2 5200 with Mandriva 2008 64bit and am wondering if something needs to be enabled (or installed) for this on Linux or if it is not possible at all.

[nzb_leecher]

Hm from 10 to 4 minutus repair time sound like 250% instead fof 60% to me, not bad 
I was always thinking that repairspeed wasnt cpu related but harddisk related (mostly)

However i hope it didnt slow down the rest of the system when 2 cores are used 100%. Thats the reason i went to a dual core, i got irritated that pc almost stopped when i accidently started a program while some unparring was going on.
With 2 cores i always could work further when 1 core was unparring. Sounds intresting to test however.

[shypike]

Verification is disk bound.
Repairing is mixed:
Calculating the matrices required to reconstruct the data is very CPU bound.
Reading in the files and writing back the corrections is again disk bound.

[sbultez]

looking forward to have it on the OSX build ;-)
64 bits is not really useful, but Multicore certainly is !!

[inpheaux]

If you're interested in trying out multi-core stuff, you can get phpar [the -ph build in the stats above] from this site. Just rename it par2.exe and drop it in your /sabnzbd/win/par2 directory, or compile / install it from scratch on linux.

[nzb_leecher]

when is the next beta with mcpar planned?

[shypike]

no idea. But you can replace par2 yourself!
Just put the proper files in whereever-sab-is\win\par2 .
Just make sure that the program is called par2.exe

[ferno]

If you're interested in trying out multi-core stuff, you can get phpar [the -ph build in the stats above] from this site. Just rename it par2.exe and drop it in your /sabnzbd/win/par2 directory, or compile / install it from scratch on linux.

Hi Tried to compile this on my Mandriva X86_64 but it does not compile.
I tried to google for it the the creator page is the only hit and looking of the page and other tools something tels me that this tool is targeted to windows 32 bit environments.

Anyway,

Would be nice to put my x2 at work for once.

[nzb_leecher]

I gone try this, if i read that site. Original par2 is already twice as slow so with this i should get a nice and free speedboost 


Makes me wonder.
If we use this par.exe but set it to use 1 core max somewhere in xp settings (is that possible in xp or only in vista?). In theory repairing should be twice as fast then with the original SAB par.exe but still leaving 1 core free for other purposes at the same time.

[inpheaux]

I gone try this, if i read that site. Original par2 is already twice as slow so with this i should get a nice and free speedboost 

No, Original Par2 is the baseline. Multi-core Par2 is as-many-times faster than single-threaded Par2 as however many cores your processor has. 2 cores, half the time. Plus, as evidenced from the time-trials listed above, the speed gains are only evident during Par2 creation and repair.

Makes me wonder.
If we use this par.exe but set it to use 1 core max somewhere in xp settings (is that possible in xp or only in vista?). In theory repairing should be twice as fast then with the original SAB par.exe but still leaving 1 core free for other purposes at the same time.

You don't understand how multi-threaded apps work. If you can't use your second core, multi-threaded Par2 will be just as slow as single-threaded Par2.

[nzb_leecher]

Let me explain a little better.

I know how multi threaded apps work, but if i read the text on his site. He did two thing.

1. added mmx support making things faster
2. assembly making it more faster.
3. added multicore support.

The best profit you get when 1,2 and 3 are used. But you can also use only 1 and 2 but no multicore, then also things would get faster with only one core.
It seemed par.exe was always a lot slower then quickpar 091 (nver knew that btw) but with this build they equal again, using more cores par.exe is even quicker.

[Delirium]

I haven't got any figures to support it but it seems the par stage is twice as fast as before with this multi core par.

Is the unrar already multicore?

[ferno]

any one got this compiled and running on Linux? Or i is just win32?