Page 1 of 1

Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 9:28 am
by Quade
Seeing the results of the XP question, It seems that many of you are running Newsbin on relatively low end machines. I dug thought my junk pile and found an old AMD 64 laptop. 2 Ghz single core with an 80 Gb laptop drive. The Ethernet interface was only 100 Mbps. I have a test server on my lan which will deliver files at up to 300-400 Mbps. It's in a Centos Linux server with its built in NNTP server running under VM ware on a 4 core i7 machine. I post problematic file sets to it as people report them so, i can re-test them in Newsbin before release.

On my main work machine, which is a powerful beast, downloads will run at over 200 Mbps, more often than not faster, this includes repairs and unrars. If it's not currently repairing or unraring, download speed might be twice that, 400 Mbps or so. CPU load is probably 15% total across the cores. It typically takes about 15 minutes to run through all the downloads on the server.

I ran the same test on this laptop and it wasn't good. It took all afternoon to complete. Downloads ran at about 60 Mbps for a time, until autopar scanning kicked in. The combination of download and autopar scanning dropped the download speed to 10 Mbps or less. CPU load was never pegged, it might hit 60%. When the Autopar scan, repair or unrar finished download speed would pop back up to 50-60 but, never get much higher than that. The problem on this laptop isn't CPU. It's the disk drive. I disabled autopar so, no par scanning repair or unrar happened and the speed became a fairly consistent 60. The cache line down on the status tab remained at 0/100 meaning Newsbin simply can't shift the data to disk fast enough.

One experiment I did was to pause the download, wait for the cache to build back up again and then unpause it. Doing that, Newsbin flatlines the data transfer at about 60 Mbps until the cache runs out. Then it gets spiky as the lack of places to store the downloaded data stalls the download. This also suggests the real bottleneck is the drive. At 60 Mbps with no autopar, CPU load is 40-50%. I consider that a smidge high even though this machine is really old. There's not much I can do in software about shitty disk IO. The bottom line is I download data and need to offload it to disk at least as fast as I download it and the disk drive in this machine simply can't take it. Key takeaway is on really low end machines like this, you might want to disable autopar and use an external tool to post process the files. You might want to download all the PAR files too.

Key points:

- Newsbin 6.33 worked fine with no crashing under XP patched to SP3
- Multi-core repair DLL worked fine on this single core AMD with no crashing.
- On this low end machine, disk IO dominated everything. Newsbin simply couldn't shift data to disk as fast as it could download and decode it.
- Depending on what you want, high download speed or autopar, you might consider disabling autopar. Autopar worked fine but, the disk IO required by autopar stalled the download. If you just download stuff and come back later and don't really care how fast it downloads, you can leave Autopar active.
- New feature in 6.4 that'll let you disable download during unrar/repair might be useful.
- If you have a high speed internet connection, a fast machine will really maximize your download speed. On slower internet, these issues probably don't matter.

I consider the high/low speed break to be 10 Mbps. If you have a 10 Mbps or less connection, none of this probably impacts you.

Re: Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 11:36 am
by mimauk
Might be a dumb question but -

If I was to add another hard drive to my PC to run NB on, what type would you recommend for the fastest IO? ATA - Sata - Solid State ??

Re: Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 11:43 am
by Quade
Clearly SSD is the fastest but, it's also the most expensive.

SSD if you can afford it.

SATA. The larger the drive typically the faster it is. I'd want 7200 RPM and would avoid "green" drives.

ATA is pretty much dead.

One optimization is to download to one drive and unrar to the other. That should give best unrar performance.

Re: Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 12:10 pm
by mimauk
Thanks for the reply.

Re: Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 6:22 pm
by Quade
I have another data point.

HP1000 Netbook, DuoCore atom running at 1.6 Ghz. Internal 16 GB SSD. Downloads going to a generic external USB drive over a USB2 interface.

This machine truly shocked me. It's maintaining 40 Mbps WHILE unraring and repairing. It hits 80 Mbps when it's not unraring. It absolutely smokes the supposedly more powerful. 2 GHz AMD machine. Smokes it even though the AMD was using an internal hard disk and the netbook is using external USB.

This netbook would be a perfectly viable download slave for someone who uses search, the IOS remote control app or NZB files. It's got no fan, runs cools and uses next to no power.

Re: Performance tests on a one lung machine

PostPosted: Thu May 24, 2012 6:36 pm
by TMM
are there any optimisations I should do when using an SSD as a download drive I do not want it to prematurely killed the drive

Re: Performance tests on a one lung machine

PostPosted: Fri May 25, 2012 9:44 am
by Quade
Another data point. I have two large damaged test sets that simply won't repair on the Netbook but, repaired on my big iron machine and on the AMD. Not sure why at this point.

Edit: I figured out why the repair and unrar FAILS. The download disk is formatted FAT32. The takeaway is that FAT32 is a bad idea when downloading biggish files.


are there any optimisations I should do when using an SSD as a download drive I do not want it to prematurely killed the drive


It might be worth using "High Speed" mode, meaning it might be worth running Newsbin "As Administrator" so, Windows doesn't clear the files before Newsbin writes to them. If you're using XP, this is the default. This clearing can double the amount of IO to the disk.

Re: Performance tests on a one lung machine

PostPosted: Fri May 25, 2012 11:09 am
by Quade
Re-formatting the disk to NTFS fixed the "No Repair/No Unrar" problem. The takeaway is don't use FAT32 any more.

Re: Performance tests on a one lung machine

PostPosted: Fri May 25, 2012 12:26 pm
by itimpi
Quade wrote:Re-formatting the disk to NTFS fixed the "No Repair/No Unrar" problem. The takeaway is don't use FAT32 any more.

Makes sense. FAT32 has a 4GB file size limit and things like a DVD image are typically larger than that.

Re: Performance tests on a one lung machine

PostPosted: Fri May 25, 2012 1:21 pm
by Quade
Exactly, that's what the problem was.