![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Since bigadv is now available for Linux, I figured we needed a good comparison. If you are currently using Windows, should you switch to Linux? The rumor mill is in full force - I have heard it is faster, and I have heard that it is slower. I wanted to put the rumor to rest and get some real data to go with.
The machine:
There is no sense pussyfooting around - if you are going to test, test big! My test machine is, you guessed it, a dual hex SR-2. This particular one is running at 3.618 GHz with 6 x 1 GB Kingston HyperX PC1600 running at 1608, 8-8-8-20-1T. It is actually my slowest dual hex machine, but it is also the most consistent. It has been churning out 130K - 135K ppd for several months very consistently without a reboot or crash. It also was the first box I switched over to the new 6.34 Windows core, so I have numbers from a couple of the new P6901 units running the Windows client.
Windows 7 Ultimate x64 was what was originally on this machine. The first 6901 it ran under Windows averaged 12:56/frame. The second averaged 12:59/frame. The variance was around +/- 10 seconds, so I am pretty confident that these times are very representative of what I would continue to see under Windows. The times are also right in line with what I was seeing on P6900s and P2686s, which also makes me think that these times would hold over time.
I installed Ubuntu Server 10.10 on this machine last night. My recent history with various flavors of Linux has not been good at all, but since I was feeling adventurous and wanted to get some data to the masses, I went ahead and installed it. The install itself went without issue – basic server with SSH and samba, nothing special. On first boot, I saw a problem – the pciehp daemon was logging errors, two every few seconds, displaying to the main screen and filling it up. On a hunch, I SSH’ed to it from a different machine and no longer saw the errors. They were still being logged and displayed on the main terminal, but were at least out of site from my remote terminal. Since I could actually work on the machine and everything seemed to function correctly, I went ahead and configured samba and set up F@H. Googling the errors being logged was not too helpful. A big thanks to Dsee from our IRC channel for helping me troubleshoot. While he was unable to solve the problem, he got me on the right track. At this point it was getting late and I wanted to see some numbers, so I fired up the client. It downloaded an A5 core and a P6901 work unit and started without issue. Success!
The first few frames were slower by around 15 seconds. Running top showed me that this pciehpd process would jump up and grab some CPU time every so often – not at all good on faster bigadv machines. Overnight the frame times settled in at 13:14 average, not bad, but slower than Windows. Since I was still not convinced that the pciehpd daemon and all of the errors it was logging weren’t affecting my frame times, I started troubleshooting again this morning. Several hours and a complete kernel recompile later, I was no longer logging the errors. I started F@H back up and now top was showing virtually nothing using CPU time except the F@H core. This is about as good as it is going to get.
The results: 10 frames after the kernel recompile, frame times are at 12:58. This is virtually identical to my Windows times. I don’t plan on changing anything on this machine any time soon, so we will see if the frame times hold up. It looks like they will – all of the frame times so far are within 5 seconds of each other, so it look pretty consistent.
Summary – you aren’t going to gain anything switching to Linux, but you also aren’t going to lose anything. I was hoping for a more conclusive result, but I am happy with the fact that we now have a valid non-Windows choice for folding boxen.
The machine:
There is no sense pussyfooting around - if you are going to test, test big! My test machine is, you guessed it, a dual hex SR-2. This particular one is running at 3.618 GHz with 6 x 1 GB Kingston HyperX PC1600 running at 1608, 8-8-8-20-1T. It is actually my slowest dual hex machine, but it is also the most consistent. It has been churning out 130K - 135K ppd for several months very consistently without a reboot or crash. It also was the first box I switched over to the new 6.34 Windows core, so I have numbers from a couple of the new P6901 units running the Windows client.
Windows 7 Ultimate x64 was what was originally on this machine. The first 6901 it ran under Windows averaged 12:56/frame. The second averaged 12:59/frame. The variance was around +/- 10 seconds, so I am pretty confident that these times are very representative of what I would continue to see under Windows. The times are also right in line with what I was seeing on P6900s and P2686s, which also makes me think that these times would hold over time.
I installed Ubuntu Server 10.10 on this machine last night. My recent history with various flavors of Linux has not been good at all, but since I was feeling adventurous and wanted to get some data to the masses, I went ahead and installed it. The install itself went without issue – basic server with SSH and samba, nothing special. On first boot, I saw a problem – the pciehp daemon was logging errors, two every few seconds, displaying to the main screen and filling it up. On a hunch, I SSH’ed to it from a different machine and no longer saw the errors. They were still being logged and displayed on the main terminal, but were at least out of site from my remote terminal. Since I could actually work on the machine and everything seemed to function correctly, I went ahead and configured samba and set up F@H. Googling the errors being logged was not too helpful. A big thanks to Dsee from our IRC channel for helping me troubleshoot. While he was unable to solve the problem, he got me on the right track. At this point it was getting late and I wanted to see some numbers, so I fired up the client. It downloaded an A5 core and a P6901 work unit and started without issue. Success!
The first few frames were slower by around 15 seconds. Running top showed me that this pciehpd process would jump up and grab some CPU time every so often – not at all good on faster bigadv machines. Overnight the frame times settled in at 13:14 average, not bad, but slower than Windows. Since I was still not convinced that the pciehpd daemon and all of the errors it was logging weren’t affecting my frame times, I started troubleshooting again this morning. Several hours and a complete kernel recompile later, I was no longer logging the errors. I started F@H back up and now top was showing virtually nothing using CPU time except the F@H core. This is about as good as it is going to get.
The results: 10 frames after the kernel recompile, frame times are at 12:58. This is virtually identical to my Windows times. I don’t plan on changing anything on this machine any time soon, so we will see if the frame times hold up. It looks like they will – all of the frame times so far are within 5 seconds of each other, so it look pretty consistent.
Summary – you aren’t going to gain anything switching to Linux, but you also aren’t going to lose anything. I was hoping for a more conclusive result, but I am happy with the fact that we now have a valid non-Windows choice for folding boxen.
