Investigating into the source and compiled it many times, I found a weird, but critical bug in the source code, which can lead to discrepancy between the results by Widnwos client and those by the official Linux client.

Eric told me he would fix soon (yeah, fortunately I found a fix), but not yet....I hope he will fix it soon, but I guess that's one of the reasons beta project is down right now (but shutting down of public seti should be the main reason, I hope...but validators of beta and that of public are different; the former is on klaatu, while the latter is on penguin) Whay is beta down? Very weird and many things remain unknown. Such is beta (or boinc)

If the "correct" crunchers are released, some of current pending results will be invalidated. But we need as reliable crunchers as possible.

Now I don't think optimized clients should prevail so we need to check official clients work properly...but by optimization/hacking source code, bugs are found.

beta project is weird...but after all beta is such a thing. It's a destiny.

---

Luckiest in the world. WMD = Weapon of Mass Distraction

Nice catch, Tetsuji! I'm sure it will get taken care of if you've alerted Eric to the situation. Beta is definately on the back burner while the main Seti project is put back into order.

I've been noticing a discrepancy between some optimized app results and standard ones. Sometimes the optimized app will detect a -9 result_overflow "noisy" WU and give up on it, while the standard app will go on to crunch the entire same WU. I don't remember ever seeing a WU with a mixed verdict on when it was a -9 noisy one or not, so it caught my attention.

A few quick examples:
http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=102214
http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=102248

---

Yes, that's exactly I am wondering now. I guess you could find them easily. I'm asking Eric for some problematic WU's because the servers are down right now. If servers were up, I could download them myself. And test (not crunch) these WU's and see how discrepancy happens. But I don't know much about Windows, and don't know how far I can go. But before then I'll try as far as I go.

I found the bug by compiling the client in many ways; standard way with gcc, optimize with gcc, optimize with ICC on linux, compile with VC++, optimize with ICC on Windows...and dumped suspicious data usually hidden inside and compared all of them. Finally I spotted it. Surprisingly the bug was in chirp/fft table generation code instead of the main calculation code. Eric also reproduced it and confirmed my fix, and he is determining which fix he will take (there is one fix, but there are at least two ways to implement it. It's a matter of "double" to "long int" conversion. So the problem is we should "round" or "truncate" the double value; Eric is wondering which to take) But unfortunately if this fixed version is released, some of pending WU's will be invalidated (maybe only a few)....I hope Eric will take care of them in some way like before; he gave credits to those who got invalidated results. At that time, I thought it's unfair, but now I think he was right. We should admit it as a beta project's nature.

In order to make things as simple as possible and to minimize problems to spot them, I will not release my optimized clients which are likely to complicate problems, although I myself will run optimized clients privately to find bugs and adjust optimization. Maybe in some cases only my result will be invalidated because of optimization. But my client will be fast. Sorry about it, but this is a beta project's nature. One reason is this time I found the bug by optimizing and it's likely bugs can be found with optimized clients.

And even if I will be the top participant here, please disregard me from the ranking. There's no function to remove one participant from the ranking list. I'll be using optimized clients on 3 machines and no wonder I am the top. It's unfair if you regard me as the real top.

I am eager to up/dl workunits asap....and wish seti@home servers were really stable...seti@home servers have too many troubles...

---

Luckiest in the world. WMD = Weapon of Mass Distraction

I think TMR could be added to the official lists of Volunteer Testers & Volunteer Devs :-)

Peter

I think TMR could be added to the official lists of Volunteer Testers & Volunteer Devs :-)

Peter

I guess it won't happen. I'll help personally.

---

Luckiest in the world. WMD = Weapon of Mass Distraction

I think TMR could be added to the official lists of Volunteer Testers & Volunteer Devs :-)

Peter

sorry, I've made a mistake. I understood what you mean and sent email to David Anderson.

---

Luckiest in the world. WMD = Weapon of Mass Distraction

sorry, nothing!

---

Luckiest in the world. WMD = Weapon of Mass Distraction