WCG won't run without the protection against creating multiple hosts. We have too many users who have the client installed on machines that are re-imaged nightly or weekly which cause a lot of abandoned results if the hosts are not linked with previous entries.
We need to update our server so we will work on that (option 2) In the meantime, I can throw a hack onto our server if needed for a specific user. I haven't been following on this thread but if the user sends an email to [email protected], then I can help them. thanks, Kevin Reed . . . . . . . . . . . . . . . . . . . . . . . . . i b m i n t e r a c t i v e 71 S. Wacker Dr From: David Anderson <[email protected]> To: Rod Walker <[email protected]>, BOINC Developers Mailing List <[email protected]> Cc: Kevin Reed/Chicago/i...@ibmus, Mark Silberstein <[email protected]> Date: 06/29/2010 11:56 AM Subject: Re: [boinc_dev] --allow_multiple_clients ignored I think I figured this out. It turns out that the --allow_multiple_clients option assumes a corresponding server option that tells the scheduler to allow multiple concurrent requests from a given host. I implemented this scheme for a particular project, superl...@technion. No other BOINC projects (including WCG) use this server option, which is why Rod gets the "Another instance" errors. I have fixed this as follows: the client now tells the server that it was run with --allow_multiple_clients, and the server handles it accordingly, with no special server flags. Now, there are 2 ways to make things work with WCG: 1) WCG uses its current server software and sets the <multiple_clients_per_host> flag in its config file. No client changes needed. The drawback: WCG loses its protection against buggy clients creating duplicate host records. 2) WCG updates its server software with changes in [21839], and Rod uses clients built from the SVN trunk. Kevin, any thoughts? -- David PS to Mark Silberstein: this change means that when up upgrade your server software to [21839] or later, you'll need to start using a new client. On 28-Jun-2010 1:10 PM, Rod Walker wrote: > There's no problem with multiple clients per host, as long as: >> 1) you use the --allow_multiple_clients option >> 2) each client has its own data directory, >> and you don't copy client_state.xml into this directory >> (client_state.xml contains host IDs, and each client must >> have a separate host ID). >> Are you doing both of these things? > Yes. I do not provide client_state.xml and this file is not in the job > run directory after boinc has run. Choosing 2 clients which ran on the > same host at the same time... > 13-Jun-2010 16:55:22 [---] Data directory: /tmp/ri32buz/boinc_TN5146 > 13-Jun-2010 16:56:08 [World Community Grid] Started download of > wcg_hcc1_img_6.0 > 8_i686-pc-linux-gnu > > and then ran to completion. The other > > 13-Jun-2010 16:55:27 [---] Data directory: /tmp/ri32buz/boinc_CL5241 > 13-Jun-2010 16:56:12 [World Community Grid] Message from project server: > Not sen > ding work - last request too recent: 5 sec > > Both are started with args > /boinc --attach_project www.worldcommunitygrid. > org 43ec69412c60d394929d1d31f2495411 --fetch_minimal_work > --exit_when_idle --no_ > gui_rpc --no_priority_change --allow_multiple_clients > > but in the log I only see Config lines > 13-Jun-2010 16:55:27 [---] Config: run apps at regular priority > 13-Jun-2010 16:55:27 [---] Config: report completed tasks immediately > 13-Jun-2010 16:55:27 [---] Config: fetch minimal work > > Should a line for --allow_multiple_clients show up here? I do not see > that in the code. Recall that I built this from the svn head in order to > get the --fetch_minimal_work functionality. > The other error I see, in a different log, is > 13-Jun-2010 16:56:14 [World Community Grid] Message from project server: > Another > scheduler instance is running for this host > > I do NOT see this error though > log_message_error("Another instance of BOINC is running."); > so I guess config.allow_multiple_clients is set. The problem seems to be > on the project side, like the project uses the hostname rather than a > hostID. > >> I don't know what WLCG and WN stand for. > WLCG is an acronym factory. World LHC Computing Grid, i.e. the computing > part of the atom smasher in Geneva. WN=Worker node (host in a cluster). > > This security info should be well received. For sure, IBM do not want to > get sued when boincers go bad. > > Cheers, > Rod. > >> >> BOINC projects are not peer-reviewed in general. However: >> >> 1) IBM World Community Grid has only applications that are >> reviewed by IBM security experts, and that do some type of >> good-of-humanity science (biomedicine, environment). >> If you attach only to WCG, you have very strong security. >> >> 2) The projects listed on http://boinc.berkeley.edu/projects.php >> are known to me, and can be trusted. >> >> -- David >
<<inline: graycol.gif>>
_______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
