Sabnzbd 0.4.2 Mac Testing Results and Observations
Posted: July 18th, 2008, 5:10 am
For about a week and a half now, I have been getting dog slow speeds from Usenet. Having explored all links from my provider, to my isp, to my router firmware, to my own computer, I decided to attempt to undertake a thorough test of download speeds from different NZB downloaders, specifically Sabnzbd.
Background
-------------
I am on a 50Mbit duplexed last mile fiber connection. I can typically see real world usage between 45-50Mbit/sec. From Usenet, that means I can typically download at around 6MByte/sec. Up until last week that was average. I don't point this out to brag, but just to give you an idea of my testing platform.
Here is a speed test from around the same time I conducted these other tests:
For my provider I use Giganews. My account allows me 10 connections and I connect on port 119. My router is a WRT300n V 1.1 and for most tests I was running DDWRT v24-Mega firmware unless otherwise noted. It is from this firmware's web interface that I am grabbing screenshots of my speed test results. Please note the scale on each screengrab - they are relative.
My computer is a 1.83Ghz Intel Core Duo Macbook Pro Running a squeaky clean install of OS 10.5.4.
Control
--------
In the past I have had the most consistent speed results using Unison. As Raf pointed out, Sabnzbd can't really be expected to be as fast as Unison as it is written in python. Still, I WAS seeing similar results before.
Testing
--------
My first test was using the latest and great version of sabnzbd 4.2.
Next, I adjusted the cache as suggested by the wiki to allow for less disk access and supposedly higher speeds.
For reference, I tried an old version of Sabnzbd. Version 0.4 beta 5
Here is .4b5 and 0.4.2 running one after the other. 0.4.2 is on the left
At this point I reflashed to the factory firmware on my router. I couldn't grab the same type of a screenshot anymore, so I grabbed this one from my desktop. This is 0.4.2 again.
Observations
---------------
I thought that the newer versions of sab must be slower, but the data shows they are actually much faster - as the second to last graph shows. This doesn't explain why I was able to blow all of these speeds out of the water two weeks ago, but clearly something has changed.
Unison downloads about twice as fast and sabnzbd now.
Increasing the memory cache for sabnzbd doesn't seem to work and actually appears to make things slower. I set the cache to 70 as suggested by the wiki, but will test more settings later.
Sabnzbd's connection seems much more sporadic. It tends to have short bursts of high speed split by period of just a little activity. Unison seems to keep a little bit more consistent speed.
Switching my router firmware appears to have doubled the speed of sabnzbd.
Conclusions
-------------
I don't think that sabnzbd is inherently that much slower than Unison. I think their download speeds should be about equal, and that is what I normally saw. I think it's more likely that Unison and Sab simply work differently, and something about my network is effecting sab more than Unison - but as you can see in the grpah, even Unison is effected a little bit (should be 5-10Mb faster).
I don't think the tests on my platform are consistent enough to draw any conclusions. I want to see more tests from more people. Anyone up to doing some comparison testing?
Background
-------------
I am on a 50Mbit duplexed last mile fiber connection. I can typically see real world usage between 45-50Mbit/sec. From Usenet, that means I can typically download at around 6MByte/sec. Up until last week that was average. I don't point this out to brag, but just to give you an idea of my testing platform.
Here is a speed test from around the same time I conducted these other tests:
For my provider I use Giganews. My account allows me 10 connections and I connect on port 119. My router is a WRT300n V 1.1 and for most tests I was running DDWRT v24-Mega firmware unless otherwise noted. It is from this firmware's web interface that I am grabbing screenshots of my speed test results. Please note the scale on each screengrab - they are relative.
My computer is a 1.83Ghz Intel Core Duo Macbook Pro Running a squeaky clean install of OS 10.5.4.
Control
--------
In the past I have had the most consistent speed results using Unison. As Raf pointed out, Sabnzbd can't really be expected to be as fast as Unison as it is written in python. Still, I WAS seeing similar results before.
Testing
--------
My first test was using the latest and great version of sabnzbd 4.2.
Next, I adjusted the cache as suggested by the wiki to allow for less disk access and supposedly higher speeds.
For reference, I tried an old version of Sabnzbd. Version 0.4 beta 5
Here is .4b5 and 0.4.2 running one after the other. 0.4.2 is on the left
At this point I reflashed to the factory firmware on my router. I couldn't grab the same type of a screenshot anymore, so I grabbed this one from my desktop. This is 0.4.2 again.
Observations
---------------
I thought that the newer versions of sab must be slower, but the data shows they are actually much faster - as the second to last graph shows. This doesn't explain why I was able to blow all of these speeds out of the water two weeks ago, but clearly something has changed.
Unison downloads about twice as fast and sabnzbd now.
Increasing the memory cache for sabnzbd doesn't seem to work and actually appears to make things slower. I set the cache to 70 as suggested by the wiki, but will test more settings later.
Sabnzbd's connection seems much more sporadic. It tends to have short bursts of high speed split by period of just a little activity. Unison seems to keep a little bit more consistent speed.
Switching my router firmware appears to have doubled the speed of sabnzbd.
Conclusions
-------------
I don't think that sabnzbd is inherently that much slower than Unison. I think their download speeds should be about equal, and that is what I normally saw. I think it's more likely that Unison and Sab simply work differently, and something about my network is effecting sab more than Unison - but as you can see in the grpah, even Unison is effected a little bit (should be 5-10Mb faster).
I don't think the tests on my platform are consistent enough to draw any conclusions. I want to see more tests from more people. Anyone up to doing some comparison testing?
