Upgrade Boinc client instructions? tx36 to 5.4.9?
27 Jun 2006 17:13:43 UTC
Topic 191471
(moderation:
As Einstein S4 ends, and S5 and current SETI use credit methods obsoleting the calibration in Trux's client, I imagine I am not alone in wanting to do a clean replacement of tx36 on three machines with the latest distributed BOINC client, 5.4.9.
Would any kind reader of this thread who knows well how to do this, or has found a good posting, please let us know?
I see the client offered for download at:
BOINC download page but am unclear on whether the previous BOINC client must first be uninstalled, nor on just what degree of preparatory tidying up is appropriate. While I do see BOINC listed in Windows Add/Remove programs, I don't know whether using this uninstall is a needed preparation for upgrade, nor whether it even works in the case of tx36.
Any wisdom or relevant experience would help me, and, I suspect, some others.
{N.B., I am speaking of the client, boinc.exe, not the science ap(s)}
Upgrade Boinc client instructions? tx36 to 5.4.9?
)
it worked fine here by uninstalling the old version through Windows Add/Remove programs and then install the new version to the same directory.
i personaly took the lazy
)
i personaly took the lazy route, i just installd the new official cliant over the trx one, and deleted the truxoft_prefs.xml file tho i didnt need to. meh was just lazy :)
but from my understanding to be on the safe side you could back up the "projects" folder with in the boinc folder, then go in to add/remove programs and uninstall boinc, then see whats left in the boinc folder... wouldnt mess with anything tho. then install the official cliant and check to be sure the projects folder is there.
ofcourse i used my firewall to block all internet (pullin the network cable from the modem woulda done the same thing) and startid up boinc.. just in case :)
seeing without seeing is something the blind learn to do, and seeing beyond vision can be a gift.
BTW, you don't have to ditch
)
BTW, you don't have to ditch 5.3.12tx36 as a matter of requirement for S5. Just select "no new tasks", wait for your S4 units to complete, and send them upon completion. Then in your account preferences, set your cache (connect to network every: ) to a low value, .05 days for example.
You can "turn-off" calibration in the Trux client. Close BOINC. Open Truxoft_prefs.xml w/ notepad and remove Einstein from calibration or off if it is the only project.
Afterwards, restart BOINC and select "allow tasks". With the low cache setting, you should only DL 1 or 2 WUs. This will allow the DCF to correct itself after the first S5 WU is completed and returned. After the first couple of WUs are uploaded, you can increase your cache to your preferred level.
If you still want to do the 5.4.9 upgrade, here's what's worked for me.
I recently did this exact upgrade when S@H migrated to Enhanced.
Suspend project.
Suspend Network Activity
Close BOINC
Go to your BOINC folder, remove/delete Truxoft_prefs.xml file
Trash boinc.exe as well
unzip and extract 5.4.9 into Boinc folder, click overwrite "yes", if asked.
Launch BOINC, reconnect network, resume project.
Done.
Just my personal preference, but I try to avoid uninstalls/program removals unless absolutley necessary. Have had ugly issues w/ Windows in the past w/ uninstalls.
Hope this helps!
Some observations after all
)
Some observations after all the activity this past weekend..When einstein was going down there some weeks back, I hooked to Rosetta and ran apps using 5.3.12tx36...was getting credits in the 110 range..so, I installed 5.4.9, ran a Rosetta and was immediately rewarded with 1/2 the credit.
So, back to reinstalling 5.3.12tx36 and now the credits went back to 110+ for the same time.
Now, I had not done a Seti since 2000, so I got a unit and ran it with the 5.3.12tx36, whoa...it recalibrated 34,675 seconds to 15,515 and gave 58 credits which hurt everyone in the quorum. Now, I had to abort the Einstein WU's because we were asked to, so I aborted one, detached and reattached.
Whoa..I get a WU with a time of completion of over 28 hours.
So, I will try the Einstein unit with the 5.3.12tx36 and see what happens, since I like having Rosetta as a backup project and it works better with 5.3.12tx36, I will run this Einstein WU on it and then try a 5.4.9 on one.
I am keeping both apps in reserve until this shakes out for me..
RE: it worked fine here by
)
Thanks. I did just that (after first stopping boincmgr of course) on one machine.
Getting a little braver, on a second machine I just stopped boincmgr and did the new install without touching the old at all, let alone the uninstall.
Both seem OK, though I have an extremely long-queue quorum partner to wait for on the validation confirmation.
RE: Some observations after
)
The same thing happened to me. The Trux client tweaks the benchmark values in addition to the optional calibration feature. I only had calibration active for Seti (non-enhanced) and Einstein S4 (running Akos's S41.08 app), but Rosetta still overclaimed credit (vs. the official client) by a factor 2+.
Seti Enhanced awards credit based on FpOps, not time/benchmark. The Trux .tx36 client will claim the correct amount of credit regardless of calibration. However, the calibrated wu times will throw off the projects Duration Correction Factor and could adversely affect BOINC's ability to manage your wu cache.
Probably a "long" work unit. They take ~12 hours on my dual Athlon MP 2000+ box.
You won't see any difference in credit claimed/granted on Einstein S5 wu's, as the wu credit is determined on the server side. Nor will you see any difference in crunching speed, as this has almost notihing to do with the BOINC core client. You will get an artificial credit bonus on Rosetta with Trux .tx36, but that could disappear after the CASP7 study, as there has been discussion about going to FpOps or fixed wu credit there as well.
Regards,
-- Tony
Tony.....Appreciate your
)
Tony.....Appreciate your enlightening comments...Stan
Glad to hear that the long Einstein WU's will probably take less time
than it is showing...
RE: Tony.....Appreciate
)
Just a couple of clarifications:
FWIW, I didn't see a significant change in what the Trux' CC produced for benchmarks as compared to a straight 5.2.13 (or going to 5.4.9 for that matter) when I first started using it. BYMMV.
It does the calibration by adjusting two of the variables used in the orignal BM based claimed credit calculation, of which CPU time is one (don't have any old SAH results left, didn't need to use it here even running Akos S4 on my slugs, and can't remember the other one).
Also, it takes a few results to be run through the mill before it "hits the mark" with the cal, which explains what you saw on your test run.
Finally, you can set which projects you want to calibrate with a parameter in the Truxoft_prefs file.
Alinator
RE: RE: Tony.....Apprecia
)
The .tx36 client almost doubled the integer benchmark on my Athlon MP 2000+ box, and also increased the floating point benchmark a little as well. The amount of tweaking could vary by CPU speed, L1/L2 cache, etc. As you said, YMMV.
According to the S4 results still in my results list, the other calibtated value is "Mfpops", which I presume is the floating point benchmark.
Right, except that he shouldn't be calibrating Rosetta at all (no optimized apps), and it doesn't matter for Seti Enhanced and Einstein S5.
Also note that out of the box, the parameter is set to "all" and should be adjusted if one is running projects that don't use optimized apps or don't use time*benchmark credit.
Regards,
-- Tony
[EDIT: Fixed formatting]
Yep, since SAH was the target
)
Yep, since SAH was the target project when Trux did this work, it probably should have been set to only calibrate on it instead of all by default, but that's water under the bridge now. ;-)
Alinator