My PII Linux box is now crunching three WUs, a very heavy QMC with deadline 28/1, a small SETI unit with deadline 26/1 and an average Einstein with deadline 1/2 for which I am using 4.24. It is very interesting to see how BOINC 5.10.21 alternates the three WUs,with Einstein shared 300 and the other two 100. My Linux is SuSE 10.1. The PII is a 400 MHz Deschutes and never had a computing error or an invalid result.
Tullio
The point: Those things are not directly Microsoft's fault...
Sure not, but vulnerabilities is, for example ActiveX in IE is a major source of spyware and viruses, its the hole that should never have been opened, wheres the sense in that? Yes, ActiveX would run only on Windows, MS thought if ActiveX became common on Internet people would b ecome even more dependent on Windows. Luckily it has backfired, hope they have learned something. Surely, IE will ask if you want to run an ActiveX but then again you need to be pretty experienced to know whats safe and not.
Some tasks are easy (like defrag as you mentioned), some are not, it takes an experienced user to mess with MSCONF and registry and whatever, all in all Linux is less work and hassle IMO, OS maintenance is a thing of the past for me. I will however set up a Vista rig soon, for gaming and DirectX10, so im not yet finished with paying OS licenses (and i also pay for some Linux software like Cedega). Im not anti-MS anyway, so no problem, in fact i have 11 or 12 MCP certifications i think, my MS knowledge used to be my main source of income.
Linux users who miss the problems you were used to on Windows, you could always get Linux Genuine Advantage :D
Quote:
According to an independent study conducted by some scientists, many users of Linux are running non-Genuine versions of their operating system. This puts them at the disadvantage of having their computers work normally, without periodically phoning home unannounced to see if it's OK for their computer to continue functioning. These users are also missing out on the Advantage of paying ongoing licensing fees to ensure their computer keeps operating properly.
Ahem... pleeeeease, let us not discuss the pros and cons of Linux and Windows here *in general*. As for S5R3, thanks for the nice diagrams! But this is really a compiler comparison , not related to the OS. As we saw with 4.25->4.26 , a near 20% performance difference is even possible for two versions of a compiler by the same vendor.
My expectatation is that a Windows app that will use the SSE vectorization will be much closer to the corresponding Linux version (because the SSE code is handcrafted and can't be messed up by the compiler). Even for the non-SSE version, Bernd has asked Akos for a hand-crafted version of the x87 hot-loop, so after integrating this one there should be a closer platform parity for i686-Linux and Wintel apps.
The app has a second hotloop, tho, so some difference may remain.
I just went back in the S5R3 thread to look what the app performance was at the beginning of S5R3 and found this one:
Quote:
On Host 579944, an AMD Opteron 285 running Linux FC3, I have noticed a gradual decline in the average amount of cobblestones/credit/points awarded each month.
[pre]MONTH CRUNCH TIME TYPE FIXED VALUE AVERAGE[/pre]
Now the same host is completing workunits within runtimes of 27k ... 42k secs. So at least for SSE capable Linux machines, we have now reached the point where performance has roughly doubled since the beginning of S5R3 in late September 2007.
So at least for SSE capable Linux machines, we have now reached the point where performance has roughly doubled since the beginning of S5R3 in late September 2007.
Not to detract from the notion that performance has definitely increased, but these figures need a footnote in regards to the credit increase that happened between then and now. IIRC, the boost was around 7%...
So at least for SSE capable Linux machines, we have now reached the point where performance has roughly doubled since the beginning of S5R3 in late September 2007.
Not to detract from the notion that performance has definitely increased, but these figures need a footnote in regards to the credit increase that happened between then and now. IIRC, the boost was around 7%...
Brian
Well, the credit increase comes atop of this, so credit/h increased even more than by a factor of two. I was just comparing runtime per unit.
Well, I finally got a decently populated first cycle. I charted it, and also charted the "claimed credit/hour" for the same wus. The first point IS a "0" sequence using 4.27, the cyan peak is sequences 101, and 102, so it's close to the 105 peak using 4.24. I imagine those points could be used to compare "speed increase at peak". I wonder if some might not use the newly available tools to "cherry pick" only the best paying wus and abort the rest?? I imagine if there are such individuals, there number are few.
My PII Linux box is now
)
My PII Linux box is now crunching three WUs, a very heavy QMC with deadline 28/1, a small SETI unit with deadline 26/1 and an average Einstein with deadline 1/2 for which I am using 4.24. It is very interesting to see how BOINC 5.10.21 alternates the three WUs,with Einstein shared 300 and the other two 100. My Linux is SuSE 10.1. The PII is a 400 MHz Deschutes and never had a computing error or an invalid result.
Tullio
RE: The point: Those things
)
Sure not, but vulnerabilities is, for example ActiveX in IE is a major source of spyware and viruses, its the hole that should never have been opened, wheres the sense in that? Yes, ActiveX would run only on Windows, MS thought if ActiveX became common on Internet people would b ecome even more dependent on Windows. Luckily it has backfired, hope they have learned something. Surely, IE will ask if you want to run an ActiveX but then again you need to be pretty experienced to know whats safe and not.
Some tasks are easy (like defrag as you mentioned), some are not, it takes an experienced user to mess with MSCONF and registry and whatever, all in all Linux is less work and hassle IMO, OS maintenance is a thing of the past for me. I will however set up a Vista rig soon, for gaming and DirectX10, so im not yet finished with paying OS licenses (and i also pay for some Linux software like Cedega). Im not anti-MS anyway, so no problem, in fact i have 11 or 12 MCP certifications i think, my MS knowledge used to be my main source of income.
Linux users who miss the problems you were used to on Windows, you could always get Linux Genuine Advantage :D
Team Philippines
RE: Linux users who miss
)
Or even Linux Genuine Advantage. Which OS messed up that link, then? ;-)
RE: RE: Linux users who
)
I dont see a messed up link, your browser must have been playing tricks on you =)
Or i edited it, its correct also in your quote.
Team Philippines
Ahem... pleeeeease, let us
)
Ahem... pleeeeease, let us not discuss the pros and cons of Linux and Windows here *in general*. As for S5R3, thanks for the nice diagrams! But this is really a compiler comparison , not related to the OS. As we saw with 4.25->4.26 , a near 20% performance difference is even possible for two versions of a compiler by the same vendor.
My expectatation is that a Windows app that will use the SSE vectorization will be much closer to the corresponding Linux version (because the SSE code is handcrafted and can't be messed up by the compiler). Even for the non-SSE version, Bernd has asked Akos for a hand-crafted version of the x87 hot-loop, so after integrating this one there should be a closer platform parity for i686-Linux and Wintel apps.
The app has a second hotloop, tho, so some difference may remain.
Bikeman
RE: The app has a second
)
My personal viewpoint is that 5% is max variation that is acceptable to me. 1-3% should be the target range, IMHO...
I just went back in the S5R3
)
I just went back in the S5R3 thread to look what the app performance was at the beginning of S5R3 and found this one:
Now the same host is completing workunits within runtimes of 27k ... 42k secs. So at least for SSE capable Linux machines, we have now reached the point where performance has roughly doubled since the beginning of S5R3 in late September 2007.
CU
Bikeman
RE: So at least for SSE
)
Not to detract from the notion that performance has definitely increased, but these figures need a footnote in regards to the credit increase that happened between then and now. IIRC, the boost was around 7%...
Brian
RE: RE: So at least for
)
Well, the credit increase comes atop of this, so credit/h increased even more than by a factor of two. I was just comparing runtime per unit.
CU
Bikeman
Well, I finally got a
)
Well, I finally got a decently populated first cycle. I charted it, and also charted the "claimed credit/hour" for the same wus. The first point IS a "0" sequence using 4.27, the cyan peak is sequences 101, and 102, so it's close to the 105 peak using 4.24. I imagine those points could be used to compare "speed increase at peak". I wonder if some might not use the newly available tools to "cherry pick" only the best paying wus and abort the rest?? I imagine if there are such individuals, there number are few.