Ubuntu Server vs Arch in VMWare

Cob

Cob

Cob

Soldato
Joined
30 Jul 2006
Posts
18,488
Location
Antrim town
This may not interest anybody, but it follows on from this discussion in the Linux forum.

I'm not going to go into the 'why's again, but basically we wanted to try something different to Ubuntu.

x64 Arch linux was the recommended distro ('Faster than a scalded cat' apparently, something I can confirm after installing Gnome on it for a couple of days back at the start of Nov), but we were unable to get the 32bit libraries to install on it. These are a necessity for linuxSMP folding, because although the client is x64, some parts of the package are still 32bit and need these libraries to run.

And now, 6 weeks later, I came across this Arch VMWare install guide over at EOC, which includes a working 32bit-libs installation method.


Now that I've been folding on an Arch VM for a day or so, I thought I'd post my results for the likes of Billy and anyone else that would like to try something different.

Comparison benchmarks are taken from an uninterupted 12 hours folding of 2605 work units (midnight to midday). No WU's were finished during this time. The only extra load on my machine during that time was my daily NOD32 scan at 9am.

The VMs used are a fully updated x64 Ubuntu Server VM (Gutsy) and a fully updated x64 Arch VM, both of which have only a pair of linuxSMP clients, the 32bit-libs and Samba installed. They are running on VMWare Server v1.0.4.

System is-

x64 Vista Home Premium SP1
Q6600 @3.5ghz (389x9)
3gb memory @972mhz 5-4-4-12
VMware has 1108mb allocated to it, with each VM being allocated 512mb.
Both VM's are set-up on the same 20gb partition on my storage drive (300gb SATA Maxtor DiamondMax 10)

Project : 2605
Core : SMP Gromacs
Frames : 100
Credit : 1760


Ubuntu Server-

Code: 
-- CPU1 --

 Min. Time / Frame : 20mn 22s  - 1244.39 ppd
 Avg. Time / Frame : 20mn 27s  - 1239.32 ppd
 Cur. Time / Frame : 20mn 34s  - 1232.29 ppd
 R3F. Time / Frame : 20mn 29s  - 1237.30 ppd
 Eff. Time / Frame : 20mn 36s  - 1230.29 ppd
Code: 
-- CPU2 --

 Min. Time / Frame : 20mn 16s  - 1250.53 ppd
 Avg. Time / Frame : 20mn 25s  - 1241.34 ppd
 Cur. Time / Frame : 20mn 33s  - 1233.28 ppd
 R3F. Time / Frame : 20mn 27s  - 1239.32 ppd
 Eff. Time / Frame : 20mn 43s  - 1223.36 ppd

Arch

Code: 
-- CPU5 --

 Min. Time / Frame : 20mn 21s  - 1245.41 ppd
 Avg. Time / Frame : 20mn 27s  - 1239.32 ppd
 Cur. Time / Frame : 20mn 35s  - 1231.29 ppd
 R3F. Time / Frame : 20mn 31s  - 1235.29 ppd
 Eff. Time / Frame : 13mn 36s  - 1863.53 ppd

Code: 
-- CPU6 --

 Min. Time / Frame : 20mn 19s  - 1247.45 ppd
 Avg. Time / Frame : 20mn 27s  - 1239.32 ppd
 Cur. Time / Frame : 20mn 34s  - 1232.29 ppd
 R3F. Time / Frame : 20mn 30s  - 1236.29 ppd
 Eff. Time / Frame : 13mn 34s  - 1868.11 ppd


So, as you can see, other than the 'Eff. Time' (what is that?), there is **** all of a difference :D
 
Last edited:
Eff. Time is effective time, this includes time when the client isn't being run.
If you were only folding part time on a machine than Eff. Time would give the actual ppd you're getting from that machine running as it was.


edit: Nice work on the testing btw - would have been nice to find some huge benefit but at least there's no need for others to try it (unless just for giggles :D)
 
After a rather wordy and well written post I was kind of disappointed that there was no improvement(!)

At least its been tested. Im sure someone must know what "Eff. time" is.

Edit - the answer beat me :D
 
Well the main aim of getting the client to actually run on Arch was accomplished.

And a better comparision for those who run native linux as their main OS would be between Ubuntu Desktop and an Arch install running from a GUI to see how the extra background processes affected the results. But I'll leave that up to them. In theory Arch should be a bit faster in those circumstances, tho a Ubuntu veteran will probably know how to tune a Ubuntu install so that they have only the necessary processes running etc.

I've now moved both my VM's over to Arch. I won't gain any extra ppd, but I like the fact that it is a rolling release, and therefore I won't be needing to reinstall a new version every 6 months to keep up with the Jones'.
 
I tried out the Windows SMP client again the other night and it's suddenly become faster than Ubuntu under VMWare on my system :confused: Always used to be a good 100ppd slower, but now I'm reliably getting 2030ppd overnight, folding the 2653s under Windows where I could only manage 1900ppd under Ubuntu. Not sure I understand it as I'm still using the same hardware (E6300 at 3.2, 3GB DDR2, DS3, Vista Ultimate)
 
Thanks for that, Cob. I'm having a few issues with Gutty Gibbon (like it won't change screen resolution and the time-keeping is even worse than FF's) but it's nice to know there's no performance advantage in changing.

I added an Ubuntu SMP client under VMWare to my second Q6600 (2.7GHz) and am now getting a combined 3300 PPD instead of 2500 for just the Windows SMP client. The cores are also running about 4C warmer.
 
You must log in or register to reply here.
Back
Top Bottom