Well its here now, just installed without problem. Note that you need to close Internet Explorer (and of course the current client) before installing.

Too early to call it bug free as yet (but I haven't had any major issues with any dev versions yet), but many users will enjoy some of the improvements made to the visual interface.

We can track usage and reports through this thread now ;-)

---

Kolch - Crunching for the BOINC@Australia team since July 2004.
Search for your own intelligence...

New messages as well, I see. Also still needs the old client to be uninstalled. (I was using 4.30)

29/04/2005 17:32:55||schedule_cpus: must schedule
Right, go schedule then. :)

29/04/2005 17:32:55||request_reschedule_cpus: files downloaded
There's a whole new message lot here for Paul to decode.

---

I'm guessing right now isnt the best time to update since the upload and download stuff takes awhile to work, i got 4 WU's downloaded, 2 more waiting, and i have 6 waiting to upload...so people should wait for them to catch up, which i am guessing is happening right now. A lot of backed up work units waiting to be sent and recieved.

---

Join the team, SETI.USA We are growing and could use your help to overcome SETI.Germany...www.setiusa.net

Looks like they have set up the code that makes the client work on those WU's that have the earliest report deadline.

---

Kolch - Crunching for the BOINC@Australia team since July 2004.
Search for your own intelligence...

Yes, I noticed that. Had to suspend all other projects running for a couple of minutes to force one E@H unit to run again that had only 10 minutes left. Looks like that's still not being done then, ignoring all other projects' timelines if one project stops with only a little of x-amount time to go.

---

I have been running 3 projects on v4.35 for 2 weeks now without any problems. IMO this version is quite safe to use.

---

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

Running 4.35 fine here too.
Win XP SP2
Seems ok with BOINCView 1.0.4
I did uninstall 4.25 first - don't know if that was necessary.

---

> Looks like they have set up the code that makes the client work on those WU's
> that have the earliest report deadline.
>

any idea if it runs the WU with the earlist report deadline until it's finished, without using the resource share we decided?

Because if that's how it works, then projects with short deadlines will have an advantage over projects like ClimatePrediction that would never run.

G

---

Military intelligence is a contradiction in terms (Groucho Marx)

at

From the developer's mailing list:


Goal

To make a per client effective scheduling system that can complete all work by its deadline.

Known factors :-

1.. boinc uptime (based on a average on a 24 hour deadline)
2.. computers speed
3.. internet connection
4.. resource share
5.. work unit deadline
6.. project debt
7.. Projects can have many applications
8.. Needs to be able to deal with many projects
9.. Needs to deal with projects that require ASAP workunits
10.. Project queue length
11.. Projects entering pre-deadline mode need to be treated as ASAP candidates in this case put into deadline mode

Short term Scheduler

The short term scheduler decides what work unit is placed on the cpu at what time. The scheduler will normally use a round robin. The scheduler will check to see if work units are reaching there pre-deadline. If a project is placed within its pre-deadline. It will be entered into "deadline mode". (Explained in detail below). Work units downloaded can have two possible prority settings. 1 meaning process as soon as possible and 2 meaning normal.

What is deadline mode?

Deadline mode is when the computer finds a workunit within a 20% (this value might need to be changed) margin of its final deadline. If a client is in deadline mode no work is to be downloaded from other projects unless the client is on dialup and does not have enough work to cover processing till the next connection time.

- How does this affect multi-cpu machines

In the case of a multi cpu machine all cpu's keep and house a separate queue therefore if only one cpu is in deadline mode the other(s) don't take notice of the deadline mode status unless this one cpu is snowed under then one or all of the other processors can change and help the cpu that is in deadline status solve its deadline problem.

Queue Manager (Previously known as Long Term Scheduler)

The Queue Manager is the part of the client that forms connections between the client and project servers. The queue manager can run in two modes "im connected permently" or I connect every n days. On the case of im connected every n days the client will avoid downloading work units that fall before the n day period.

- When will Boinc know when to download more workunits?

There is a list of logic instructions the client will check through:

1)Will the client meet all client deadlines

2)Is there proritised work units in progress causing other workunits to be pushed closer to there deadlines

3) Is a workunit in deadline mode

4) Project with the highest long term debt has sufficient work on the
client.

5) CPU scheduler is booked. Idea given by someone else, but I did not
see it. Add up the shares of CPU time needed to get each WU done by
deadline. For example a WU that has a deadline in 10 days and needs 1 day
of crunching to complete would require a share of 0.1. A WU that has a
deadline of 2 days and requires 1 day of crunching would have a share of
0.5.

6) A server returns a response that indicates that there would be time if
the client were somewhat less booked.

7)Is the client on dialup and are we on a day when a connection is present and data can be transferred

What extra data will be sent to the server respective of what currently is sent?


86400=7200; // 1 day deadline 2 hours of slack time.
172800=7200; // 2 day deadline still 2 hours of slack time.
432000=122400; // 5 day deadline 34 hours of slack time.
1209600=327600; // 2 week deadline 91 hours of slack time.
315360000=20656800 // 1 year deadline 5738 hours of slack time.

0
18000 // changed data to seconds

What changes will be made to the server?

On the server the client needs to be able to enter how often the machine is connected and how much work the client wishes to download.

When a server sends a priority workunit what differences does this make to the client?

The work unit is completed as soon as possible and when the next possible internet connection is made the client attempts to report the data.

How does the server decide what work units to send me?

The server takes the RPC sent to it and reads through it, it checked the provided slack time and uses your computers benchmarks to validate the workunit to see if it can fit within the slack time provided. If the server thinks the client can complete the workunit within time it will be sent. Otherwise if no work can be sent a RPC reply will be returned with no work available message.

---End of proposal---


There was a lot of discussion on the list about specific details, but as far as I know, the basic proposal did not change ...

---

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

uhuh...

29/04/2005 22:24:22||Too many projects have work.
29/04/2005 22:24:22||New work fetch policy: no work fetch allowed.

That I have 9 LHC units that say they want to run in an hour and 25 minutes, but actually run in 35 minutes and less, isn't accumulated for.

That those units, as well as most units in my list, have a deadline of 12 to 13 May 2005, isn't accounted for.
Tomorrow Pirates may have units for me, but I won't be able to crunch them, as I am bound by 5 projects all of a sudden. If one LHC unit is still lurking around by then, I can say goodbye to helping out with Pirates.

Who came up with this scheme? I am managing fine to run 7 projects on my PC. I don't miss deadlines. So who came up with this scheme?

---

Hi All,
I have developed a problem when updating my projects. They don't update and they say the following:
4/29/2005 9:29:29 PM||request_reschedule_cpus: process exited
4/29/2005 9:29:29 PM||schedule_cpus: must schedule
4/29/2005 9:29:30 PM||request_reschedule_cpus: project op
4/29/2005 9:29:30 PM||schedule_cpus: must schedule
4/29/2005 9:29:53 PM||Suspending network activity - user request
4/29/2005 9:32:57 PM||request_reschedule_cpus: project op
4/29/2005 9:32:57 PM||schedule_cpus: must schedule
4/29/2005 9:32:57 PM|climateprediction.net|Pausing result 19ii_000079597_0 (removed from memory)
4/29/2005 9:32:58 PM|Einstein@Home|Restarting result H1_0269.9__0270.1_0.1_T06_Run2_2 using einstein version 4.79
4/29/2005 9:33:01 PM||request_reschedule_cpus: process exited
4/29/2005 9:33:01 PM||schedule_cpus: must schedule

Can Someone help please and let me know what this means. I have uninstalled 4.35, but I am still getting the same responses on 4.32 which was working fine before.

Respectfully,

Rocky

Windos XP SP2 512mb Ram

---

www.boincsynergy.com

Depends on which project exited. Was it CPDN?

---

> Depends on which project exited. Was it CPDN?

It Is all of them, CPDN, Protien, LHC, and Seti

Respectfully,

Rocky
>

Time for a Boinczilla entry? ;)

---

Seems that this is very similar to when a new version was released not that long ago, I think it was 4.30, this version was hastily replaced with the "true" release.

---

Well, after going from 4.32 to 4.35 is changed my CPDN due time to 1901 and then politely suggested I consider aborting it as it was some 130,000 days overdue.

I installed 4.35 yesterday. Since then Boinc has only crunched one Einstein WU after the next. I see in the log:

4/30/2005 1:10:57 AM|Einstein@Home|Computation for result H1_0682.9__0683.0_0.1_T17_Run2_3 finished
4/30/2005 1:10:57 AM||schedule_cpus: must schedule
4/30/2005 1:10:57 AM||earliest deadline: 1115209786.000000 H1_0682.9__0683.2_0.1_T19_Run2_0
4/30/2005 1:10:58 AM|Einstein@Home|Starting result H1_0682.9__0683.2_0.1_T19_Run2_0 using einstein version 4.79

So, It worked continuously on Einstein since 2:20 EDT (1420 UTC). At the times it is supposed to switch, I get the following message:

4/30/2005 3:10:57 AM||schedule_cpus: time 7200.063183
4/30/2005 3:10:57 AM||earliest deadline: 1115209786.000000 H1_0682.9__0683.2_0.1_T19_Run2_0


the Einstein WU only have a 7 day deadline and happen to be the nearest to the deadline. But if it gets a new Einstein WU then the new WU deadline will also be before some of the others I already have downloaded for LHC and Seti.

Does this new version only run the WU nearest the deadline? If so, then Einstein's going to get a higher Percentage of the CPU cycles and get more WUs done.

I shall wait and watch

tony

> Does this new version only run the WU nearest the deadline? If so, then
> Einstein's going to get a higher Percentage of the CPU cycles and get more WUs
> done.
>

In the short-term you will most likely crunch all einstein-wu in nearest-deadline-mode, but when long-term debt for einstein is negative you'll not ask for more einstein-wu before other projects has catched-up, or you're running out of work and other projects doesn't supply any.

If no work is so close to deadline the client switches to nearest-deadline-mode, it will work just like before in highest-debt-mode, switching depending on your preference-settings "Switch between applications every N minutes".


BTW, for anyone wanting to cache for more than the default 5 projects at once, stop the client and manually edit global_prefs.xml adding a line:

<max_projects_on_client>nn</max_projects_on_client>

> Note that you need to close Internet Explorer

Why? What does BOINC have to do with IE???