Discuss newly added features or request new features.
-
hugbug
- Developer & Admin
- Posts: 7645
- Joined: 09 Sep 2008, 11:58
- Location: Germany
Post
by hugbug » 02 Sep 2014, 23:08
The most wanted feature is now available for all NZBGet users:
- doesn't depend on other libraries and works everywhere, on all platforms and all CPUs (with multiple cores), no special compiling steps are required;
- new option ParThreads to set the number of threads for repairing;
- new option ParBuffer to define the memory limit to use during par-repair;
Few test results
All tests made for the following nzb:
- Total size: 3.8 GB, 78 files 50MB each
- Block size: 2 304 000 bytes
- Block count: 1664
- Missing files: 2 (100 MB total, 44 blocks total)
For 100 MB of data which needs to be restored, the optimal buffer size is 100 MB (can be set to a higher value but only 100 MB will be used).
Results:
- Windows PC, Core2 Duo, 2.6 GHz, 4 cores:
- 1 thread, 128MB buffer, Repair time: 0:06:32
- 2 threads, 128MB buffer, Repair time: 0:03:51
- 3 threads, 128MB buffer, Repair time: 0:02:31
- 4 threads, 128MB buffer, Repair time: 0:01:50
- 6 threads, 128MB buffer, Repair time: 0:01:43
- Macbook Core i5, 2.4 GHz, 2 cores + hyper threading (4 logical cores):
- 1 thread, 16MB buffer, Repair time: 0:04:55
- 1 thread, 128MB buffer, Repair time: 0:02:25
- 4 threads, 128MB buffer, Repair time: 0:01:20
- 6 threads, 128MB buffer, Repair time: 0:01:18
- For comparison: par2cmdline+tbb (multicore), 128MB buffer, Repair time: 0:01:16
- Vu+ duo SAT receiver, MIPS CPU with 2 cores 400 MHz:
- 1 thread, 16MB buffer, Repair time: 1:10:28
- 2 threads, 16MB buffer, Repair time: 0:45:39
- 2 threads, 32MB buffer, Repair time: 0:42:58
Measured was only repair time. Verification time isn't included. With quick par-verification NZBGet has a big advantage over par2cmdline.
Implemented in v14-testing-r1116.
-
neilt0
- Posts: 261
- Joined: 10 Sep 2008, 19:11
Post
by neilt0 » 03 Sep 2014, 12:59
I broke down and upgraded unRAID to 6.0 and set up a Docker so I could easily get the latest builds and the multicore repair. It's working great!
Thanks,
Neil.
-
mannibis
- Posts: 60
- Joined: 29 Jul 2014, 15:10
Post
by mannibis » 03 Sep 2014, 15:07
Any timeline on a release for new testing version for OSX?
-
hugbug
- Developer & Admin
- Posts: 7645
- Joined: 09 Sep 2008, 11:58
- Location: Germany
Post
by hugbug » 03 Sep 2014, 15:10
Before I make a testing release I'd like the few most active forum members to test it. If you want, I'll send you the binary for testing.
-
vergessen
- Posts: 23
- Joined: 19 Feb 2014, 15:36
Post
by vergessen » 03 Sep 2014, 15:39
This is working great on my machine. debian with i7. I have tested and broken a couple of files on purpose and I have run into no issues.
-
mannibis
- Posts: 60
- Joined: 29 Jul 2014, 15:10
Post
by mannibis » 03 Sep 2014, 15:56
I'd love to test it out for OSX.
Thanks hugbug
Last edited by
mannibis on 22 Feb 2015, 22:30, edited 2 times in total.
-
salami
- Posts: 68
- Joined: 14 Apr 2014, 11:09
- Location: Switzerland
Post
by salami » 03 Sep 2014, 16:01
Tried a couple of broken jobs on Windows, working perfectly so far! Am I correct that the total of ParBuffer and ArticleCache has to be below ~1900MB in 32bit mode?
-
hugbug
- Developer & Admin
- Posts: 7645
- Joined: 09 Sep 2008, 11:58
- Location: Germany
Post
by hugbug » 03 Sep 2014, 19:06
That's correct about 1900MB. I guess I should add a validation check.
Who is online
Users browsing this forum: No registered users and 17 guests