Message boards : Number crunching : Initial Replication
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

Profile Irondog

Send message
Joined: 30 Aug 05
Posts: 11
Credit: 7,169,523
RAC: 117
Message 17728 - Posted: 5 Aug 2007, 22:52:58 UTC - in response to Message 17726.  

I've also added four machines to this project. If you’re still upset about wasting time, feel free to leave the project as you've mentioned. I'll pick up your slack.
I can always tell when I am dealing with someone of lesser intelligence when they resort to overworked cliches like the above. You have demonstrated your ability to repeat rote and very little creativity or ability to analyse the situation and generate new ideas unfetered by dogma. Now take your hackneyed "I can cover for you" attitude and ram it up your nose.

After re-reading what I said, I'll admit that was a stupid comment on my part. And reading your reply, I'd have to say our intelligence about the same. <g>

ID: 17728 · Report as offensive     Reply Quote
Profile [B^S] ShanerX

Send message
Joined: 14 Jul 05
Posts: 41
Credit: 1,788,341
RAC: 0
Message 17731 - Posted: 6 Aug 2007, 0:09:27 UTC

I think Neasan gave up on this thread a long time ago. You get about half way through and it became a battle of words, and worse. The original point is lost - whether it was an opinion, a valid question, feedback, or sheer foolishness. I've seen nothing but great responsiveness from the moderators here in the past few months (much better than in the many months of waiting during migration) - the way to get a thread acknowledged and acted on is not the way this one is going. I raise you 5.72 cents and offer the following:

* create a 'Wish List' or 'Suggestions' or ‘To Do’ main category in the message boards; LHC has the least amount of categories I’ve seen. Not a top priority, but it will help differentiate the ‘must look at’ from the ‘hey, I’m here and crunching’ posts

* questions, suggestions, or feedback like the IR5 for LHC could be reposted there and given time to be digested by persons more knowledgeable about the exact reason(s)

* let this thread die, it's outlived it usefulness

* if you need to 'show-off' your vocabulary - I'd love help wording flame-mail to my insurance company, lawyer, gas company, etc. and could better use your help there!

* if this thread continues … I won’t ‘add’ you as a friend on MySpace

ID: 17731 · Report as offensive     Reply Quote
Profile caspr

Send message
Joined: 26 Apr 06
Posts: 89
Credit: 309,235
RAC: 0
Message 17737 - Posted: 6 Aug 2007, 3:21:36 UTC

Off hand I'd say everyone has made it clear they have an in closing I'd just like to say, NOW CHILDREN,...PLAY NICE!!
A clear conscience is usually the sign of a bad memory

ID: 17737 · Report as offensive     Reply Quote
Profile Keck_Komputers

Send message
Joined: 1 Sep 04
Posts: 275
Credit: 2,652,452
RAC: 0
Message 17738 - Posted: 6 Aug 2007, 4:46:43 UTC

The initial studies when the IR was decieded upon showed an almost aysomatic(sp?) curve in turnaround time when increasing the IR. The point of diminishing returns was between 5 and 6 copies for a quorum of 3, so 5 was chosen. Once the design and testing phase of the collider is finished a different IR will most likely be desireable, until then the reasons for this decision are still valid.

Basically that means any work with the sixtrack application should continue to have the same IR and quorum. Work with other applications will most likely have different needs and therefore different IR and quorum.

BOINCing since 2002/12/8
ID: 17738 · Report as offensive     Reply Quote
Profile Irondog

Send message
Joined: 30 Aug 05
Posts: 11
Credit: 7,169,523
RAC: 117
Message 17739 - Posted: 6 Aug 2007, 5:25:39 UTC - in response to Message 17735.  
Last modified: 6 Aug 2007, 6:11:10 UTC

There is no need to create a new batch or delete the WU from the current batch and put it in the next batch. On all the projects I've ever crunched, if you return a result that has errored on your system then a replacement copy of the WU gets generated and put in the outbound queue within a few seconds. It gets sent out whenever another host requests work, could be only a matter of minutes if lots of hosts are requesting work. Some projects put the replacement at the head of the queue, some put it at the tail. If it goes to the tail then no problem, other WUs get sent and completed a little sooner, the replacement waits. It doesn't matter which one gets done first, it does matter how soon they ALL get done. Ya have to look at the entire forest not just individual trees.

Thanks for explaining this. I can now see how setting the IR to 5 is wasteful and can agree that once the project is stable, the admins should look into this.

As for your personal attack against me... I'm not going to drop back to that level by responding. We're done.

ID: 17739 · Report as offensive     Reply Quote

Send message
Joined: 4 Dec 06
Posts: 33
Credit: 75,491
RAC: 0
Message 17742 - Posted: 6 Aug 2007, 15:31:43 UTC

and here we have it: Flame Fest 2007 Part II

I sincerely hope your pointless e-arguments inflate your e-ego to the point where you feel better about yourself in the real world. At least something good will come of this nonsense.
ID: 17742 · Report as offensive     Reply Quote

Send message
Joined: 4 Oct 06
Posts: 38
Credit: 24,908
RAC: 0
Message 17745 - Posted: 6 Aug 2007, 20:01:56 UTC
Last modified: 6 Aug 2007, 20:07:31 UTC

My thoughts: The initial replication of five is a legitimate setting provided the servers don't always try to keep five results out there. The problem with the algorithm as it stands is that the servers will replicate a WU regardless of how many results are still being worked on.

From what I have experienced, a WU's life might go something like this: a WU is replicated five times and is distributed. Two results are returned, a third errors out. The servers see the error, replicates the WU, adds it to the queue, and ultimately distributes it. Now we have six results for the one WU. A fourth result errors out, gets replicated, added to the queue but hasn't been distributed yet. Seven results now. The 5th result comes back good, the WU met quorum, and the three good results are validated and granted credit. The 7th result gets killed (marked as "not needed") before it is distributed. The WU has been completed but a sixth result is still being crunched. If it comes back good, it will be granted its due credit. If it errors, oh well, no more replications since quorum has been met.

Now, I think that in order for the IR=5 to be a time saver, the servers would have to wait until additional results are needed, TAKING INTO ACCOUNT THE IN PROGRESS RESULTS, to meet quorum before they replicate a WU again. If an error occurs a replication would only be created if there where less than the number of results needed for quorum (either Success or In Progress), so in the above scenario, no additional replications would have been created since 3 of the original 5 met quorum. Ideally, with an IR=5 the WU should be able to handle two errors/past deadline before any additional work is created, since the other three have a chance to meet the quorum. But, alas, I do not see this happening here.
Don't get distracted by shiny objects.
ID: 17745 · Report as offensive     Reply Quote
Profile anarchic teapot

Send message
Joined: 15 Feb 06
Posts: 67
Credit: 460,896
RAC: 18
Message 17746 - Posted: 6 Aug 2007, 22:31:32 UTC

Girls, girls! Please!

Has everyone quite finished telling the guys actually planning and running this project how to do their jobs?

Let it ride, will you?
Plague of Mice
Intel Core i3-9100 CPU@3.60 GHz, but it's doing its bit just the same.
ID: 17746 · Report as offensive     Reply Quote
Betting Slip

Send message
Joined: 17 Sep 04
Posts: 41
Credit: 27,497
RAC: 0
Message 17750 - Posted: 7 Aug 2007, 13:14:44 UTC - in response to Message 17746.  

Girls, girls! Please!

Has everyone quite finished telling the guys actually planning and running this project how to do their jobs?

Let it ride, will you?

Why is it that everytime you try to discuss a relevant topic on BOINC boards someone pops up to tell you to forget it or, as above, "let it ride"?

I have not heard anybody trying to tell the LHC how to run the project. We are discussing the apparent wasting of computational resources at the expense of other BOINC projects. This discussion has become lively, which, is good, isn't it?

Maybe you're one of those who don't question anything, accept the status quo, don't rock the boat.
ID: 17750 · Report as offensive     Reply Quote
Profile Neasan
Volunteer moderator
Volunteer tester

Send message
Joined: 30 Nov 06
Posts: 234
Credit: 11,078
RAC: 0
Message 18200 - Posted: 16 Oct 2007, 12:58:09 UTC

IR = 5

That is the decision that has been made by the project, with input from all sides
ID: 18200 · Report as offensive     Reply Quote

Send message
Joined: 2 May 07
Posts: 39
Credit: 142,918
RAC: 0
Message 18204 - Posted: 16 Oct 2007, 14:02:58 UTC

My 2 cents worth...

1: I doubt that the engineers assembling this beast really wait much longer, if at all with the IR set to 5 rather then 3. Why? Because we have more hungry systems then work to do. The work gets done much faster then the "due by" date. I'm sure they are not just "sitting around" waiting for us to get done; I'm sure they have scheduled work as needed and don't just wait for us. [Yes I know you think that we waste 40% that could be used by other projects - see point #4]

2: We work for "them", in this case being LHC@home. If "they" wanted to set the IR to what ever, then let them set it. Yes voice your opinion, but don't keep hammering it when you don't get your way.

3: From my understanding the internal and international coordination efforts probably totally swamp this piddly 3 vs 5 IR issue. You need to keep some perspective.

4: Act locally. If you are really that concerned with efficiency and waste what are YOU doing about it? Are you using solar/battery power to run your systems? Have you gotten rid of all other applications on your systems and removed any non essential tasks so that back ground task switching does not use CPU time? Are your running RAID 5 or 10 to speed disk access? Do you have enough real memory so that you never swap out to disk? Are you using Dual Quad CPU's? Are your systems dedicated? Have you installed a virus/spam gateway device (like IronPort)and removed antivirus/antispam software from your systems that don't access the internet except for BOINC?

5: Help ME - ]:8) Send me $10K US and I'll match it and install Solar Power. Send me $500 or more and I'll match it and replace my W98 system with some fast modern system(s). You could probably help the project out more by sending THEM the money!

6: Get other people to join the project or at least BOINC.

ID: 18204 · Report as offensive     Reply Quote

Send message
Joined: 16 Jul 05
Posts: 84
Credit: 1,875,851
RAC: 0
Message 18212 - Posted: 16 Oct 2007, 18:34:20 UTC

With BOINC 5.10 it should be possible to abort the already send workunits when three canocial results are back!?
Linux Users Everywhere @ BOINC
ID: 18212 · Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : Initial Replication

©2024 CERN