Paul D. Buck wrote:
[...]
As a participant I ask for little other than a fair hearing if I see
problems, and a visible and interested project staff. I can only
think of a couple projects where I feel that is true today. A sad,
sad commentary on the state of the BOINC Community ...
OK, this is pretty much what is already done. However, there are more
things to be done than just CPU and work fetch scheduling.
The sleep has to be 1 second because there ARE things in the loop that have
to run that frequently. For example the messages from the science
applications have to be
jm7
boinc_dev-boun...@ssl.berkeley.edu wrote on 04/29/2009 04:36:26 PM:
Martin m_boinc...@ml1.co.uk
Sent by: boinc_dev-boun...@ssl.berkeley.edu
04/29/2009 04:36 PM
To
BOINC Developers List boinc_dev@ssl.berkeley.edu
cc
Subject
Re: [boinc_dev] Work Scheduling (pt 2, Cooperative
El Jue 30 Abr 2009 12:16:00 john.mcl...@sybase.com escribió:
I have the feeling that nobody ever used the FastCGI scheduler correctly,
that server admins compile it with FastCGI support as if that was all that
needs to be done.
I don't know enough about Fast CGI to have any hope of
El Jue 30 Abr 2009 12:19:05 john.mcl...@sybase.com escribió:
And then there is CPDN trickle-based kill which is completely different.
The
server sends a trickle-down to that host, knowing it has that task. The
application handles the trickle and quits, saying it finished processing.
jm7
boinc_dev-boun...@ssl.berkeley.edu wrote on 04/30/2009 11:35:16 AM:
Martin m_boinc...@ml1.co.uk
Sent by: boinc_dev-boun...@ssl.berkeley.edu
04/30/2009 11:35 AM
To
BOINC Developers List boinc_dev@ssl.berkeley.edu
cc
Subject
Re: [boinc_dev] Work Scheduling (pt 2, Cooperative
El Jue 30 Abr 2009 13:06:36 Martin escribió:
Even if you use multiple separate download servers, you just simply have
message sockets to your WU pool allocator process(es) to report what is
successfully transferred and when. Similarly so for setting what URL is
to point to what WU on the web
On Apr 30, 2009, at 5:51 AM, john.mcl...@sybase.com wrote:
OK, this is pretty much what is already done. However, there are more
things to be done than just CPU and work fetch scheduling.
The sleep has to be 1 second because there ARE things in the loop
that have
to run that frequently.
john.mcl...@sybase.com wrote:
[...]
I may have gotten confused. I thought that someone was claiming that it
***could*** be done without DB access, and be triggered by the file
download.
'Tis I!
Some notes:
It has to be backwards compatible - client versions that do not do this
exist,
Nicolás Alvarez wrote:
BURP uses download servers provided by volunteer users.
A few projects use download servers distributed across the world.
File downloads *must* work on plain old HTTP servers without having to
'call home'.
Then that is an Achilles heel for gaining feedback on what
On Apr 30, 2009, at 8:07 AM, Nicolás Alvarez wrote:
El Jue 30 Abr 2009 10:21:31 john.mcl...@sybase.com escribió:
Rom did an analysis at one time that indicated that the basic
connection
was about 10 times as expensive as the work for a single report or
request.
El Jue 30 Abr 2009 13:39:18 Paul D. Buck escribió:
Take the simplest example,
The first stage of a task is the creation of the Result, which
generates one or more Tasks.
Neither of these are of interest or should be visible to the
participant. So, why are they in the running tasks and
On Apr 30, 2009, at 8:16 AM, john.mcl...@sybase.com wrote:
Maintaining an index is also expensive. Every insert into the table
changes the index. In particular, the link between a WU and a User
would
have to have an index that is updated on every assignment (add) and
every
transfer
El Jue 30 Abr 2009 13:45:38 Martin escribió:
Then that is an Achilles heel for gaining feedback on what was accepted
by the client...
Apache can report what URL downloaded whatever, so that is one means of
feedback for servers under direct project control.
Another means could be that that
jm7
boinc_dev-boun...@ssl.berkeley.edu wrote on 04/30/2009 12:45:38 PM:
Martin m_boinc...@ml1.co.uk
Sent by: boinc_dev-boun...@ssl.berkeley.edu
04/30/2009 12:45 PM
To
boinc_dev@ssl.berkeley.edu
cc
Subject
Re: [boinc_dev] Work Scheduling (pt 2, Cooperative Scheduling?)
Nicolás
El Jue 30 Abr 2009 13:55:46 john.mcl...@sybase.com escribió:
At present, how do the download servers know to delete
downloaded/expired WUs if they can't 'phone home'?
The Validator deletes the files.
The validator compares files and chooses a canonical result. The assimilator
does
El Jue 30 Abr 2009 13:58:27 john.mcl...@sybase.com escribió:
Once a day if there was no CPU request, or with the CPU request if there is
one. This is a substantially lower count than if every project is asked
for a work proposal for every work request.
You're adding more every words than we
The assimilator has to have DB access to do its job. It determines which
files to be deleted and passes the information on to the file deleter ( and
I do not know the mechanism - it is quite possibly through the DB as the
two are completely asynchronous).
jm7
boinc_dev-boun...@ssl.berkeley.edu
Not until the next state is acted on - which happens after the state is
fully determined. Is there a bug someplace, yes, but merely setting the
flag does NOT actually change the state until the flag is acted upon.
jm7
boinc_dev-boun...@ssl.berkeley.edu wrote on 04/30/2009 01:05:01 PM:
Paul D.
On Apr 30, 2009, at 9:58 AM, john.mcl...@sybase.com wrote:
jm7
Paul D. Buck p.d.b...@comcast.net wrote on 04/30/2009 12:51:30 PM:
Paul D. Buck p.d.b...@comcast.net
04/30/2009 12:51 PM
To
Nicolás Alvarez nicolas.alva...@gmail.com
cc
BOINC Developers Mailing List
Some projects have real world deadlines that have to be met.
I have no problem with this being a goal for a project.
But in my opinion, to expect it to happen consistently is wishful
thinking. Therefore, it should be the __duty__ of project
administrators to plan for less than 100%
El Jue 30 Abr 2009 13:32:01 Mikus Grinbergs escribió:
[ And what if the
result that's returned is useless because of a bug in the project's
application? ]
And what if their output files are binary, a bug in the project's application
causes invalid results, and sample_bitwise_validator marks
On 30 Apr 2009 at 9:21, john.mcl...@sybase.com wrote:
Talk about spamming the servers. Ask every one for an offer (requires DB
access). Accept some subset of those (requires DB access).
Rom did an analysis at one time that indicated that the basic connection
was about 10 times as
On Apr 30, 2009, at 8:07 AM, Nicolás Alvarez wrote:
El Jue 30 Abr 2009 10:21:31 john.mcl...@sybase.com escribió:
Rom did an analysis at one time that indicated that the basic
connection
was about 10 times as expensive as the work for a single report or
request.
Sorry, I was reporting from memory without the actual data in front of me.
Still, a 3X reduction is important.
jm7
Josef W. Segur
jse...@westelcom
El Jue 30 Abr 2009 15:26:40 john.mcl...@sybase.com escribió:
The two are at different layers in the protocol stack.
No news to me.
Using HTTP, we're *forced* to use individual requests, each sending complete
information, and the server has to do eg. user/host lookup on every request.
Using a
OK, you want to switch to something other than http at the application
level. (possibly hand rolled) at least for the updates. That might be
some interesting design anyway.
The obvious questions:
1) What about version compatibility?
A) Old client to new server
B) New client to
El Jue 30 Abr 2009 15:35:10 Nicolás Alvarez escribió:
El Jue 30 Abr 2009 15:26:40 john.mcl...@sybase.com escribió:
The two are at different layers in the protocol stack.
No news to me.
Using HTTP, we're *forced* to use individual requests, each sending
complete information, and the server
The problem with persistent connections is the overall memory consumed in just
maintaining the connection.
Even if you striped down the scheduler to a self contained daemon you still
will have to deal with the basics of maintaining connection state for which
ever clients are connected and
29 matches
Mail list logo