Hi Paul I'm almost certain your "Error in data.frame" error is due to a bug I fixed a while ago. You need to upgrade to the latest release of puma (1.4.3), but for this you will first need R version 2.6.x installed (you should really go for 2.6.2). My guess is that this won't solve your SUSE problem, but its certainly the first thing to try. Let me know how you get on. Hope that helps! Richard. Paul Geeleher wrote: > Ya I found the 4x increase rather suspicious myself. > > I can indeed now ssh to localhost on the SUSE box. In R pretty much > the same thing as before happens, except that it doesn't prompt for a > password. I'll email you a screenshot actually. > > At the moment I'm just using the Estrogen data set to test this so the > script is as simple as: > > library(puma) > library(pumadata) > library(snow) > > data(eset_estrogen_mmgmos) > cl <- makeCluster(c("localhost", "localhost"), type="SOCK") > eset_estrogen_comb <- pumaComb(eset_estrogen_mmgmos, cl=cl) > > The output of session info on Ubuntu, after running the above, is as follows: > > > R version 2.5.1 (2007-06-27) > i486-pc-linux-gnu > > locale: > LC_CTYPE=en_IE.UTF-8;LC_NUMERIC=C;LC_TIME=en_IE.UTF-8;LC_COLLATE=en_ IE.UTF-8;LC_MONETARY=en_IE.UTF-8;LC_MESSAGES=en_IE.UTF-8;LC_PAPER=en_I E.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_IE.UTF -8;LC_IDENTIFICATION=C > > attached base packages: > [1] "tools" "stats" "graphics" "grDevices" "utils" "datasets" > [7] "methods" "base" > > other attached packages: > snow pumadata puma ROCR gplots gdata gtools annotate > "0.2-9" "1.0.0" "1.2.3" "1.0-2" "2.3.2" "2.4.0" "2.4.0" "1.14.1" > limma affy affyio Biobase > "2.10.5" "1.14.2" "1.4.1" "1.14.1" > > > > And on SUSE, which only makes it as far as the "cl <- > makeCluster(c("localhost", "localhost"), type="SOCK")" line before > going blank: > > R version 2.5.1 (2007-06-27) > i686-pc-linux-gnu > > locale: > LC_CTYPE=en_GB.UTF-8;LC_NUMERIC=C;LC_TIME=en_GB.UTF-8;LC_COLLATE=en_ GB.UTF-8;LC_MONETARY=en_GB.UTF-8;LC_MESSAGES=en_GB.UTF-8;LC_PAPER=en_G B.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_GB.UTF -8;LC_IDENTIFICATION=C > > attached base packages: > [1] "tools" "stats" "graphics" "grDevices" "utils" "datasets" > [7] "methods" "base" > > other attached packages: > snow pumadata puma ROCR gplots gdata gtools annotate > "0.2-9" "1.0.0" "1.2.3" "1.0-2" "2.3.2" "2.4.0" "2.4.0" "1.14.1" > limma affy affyio Biobase > "2.10.5" "1.14.2" "1.4.1" "1.14.1" > > > -Paul > > > On Wed, Feb 27, 2008 at 8:52 PM, Richard Pearson > <richard.pearson at="" postgrad.manchester.ac.uk=""> wrote: > >> Paul >> >> Can you now ssh without a password from the unix command line on your >> SUSE box? >> >> Could you send me the script you ran and the output of sessionInfo()? It >> sounds a bit suspicious getting over a 4-fold speed increase if you're >> just using a 2-node cluster! >> >> Thanks >> >> Richard. >> >> >> >> >> Paul Geeleher wrote: >> >>> Well Richard, >>> >>> That does seemed to have solved the problem on my Ubuntu desktop. The >>> same problem is still however occurring with SUSE so I'm guessing it >>> might be some kind of problem with the SSH server on SUSE? >>> >>> To my amazement, running pumacomb parallel (on the estrogen dataset) >>> cut the estimated time of execution from over an hour, to just 15 >>> minutes on my Pentuim D 2.8 GHz. >>> >>> Unfortunately though pumacomb() has thrown up an error at the very end >>> of its execution and has failed to output any data: >>> >>> " Error in data.frame(numeric(n), row.names = nms) : >>> duplicate row.names: " >>> >>> Any ideas what might be causing this? I've run it twice now with the >>> same result. It runs fine on the single core, without using the >>> cluster, although about 4 or 5 times slower. >>> >>> -Paul >>> >>> >>> >>> On Wed, Feb 27, 2008 at 12:58 PM, Richard Pearson >>> <richard.pearson at="" postgrad.manchester.ac.uk=""> wrote: >>> >>> >>>> Hi Paul >>>> >>>> makeCluster is a function from the snow package, and I also can't find >>>> much information about setting up "socket" clusters. However, I think >>>> the fundamental problem here is that you can't ssh to localhost without >>>> a password. The following instructions (taken from >>>> http://linuxproblem.org/art_9.html) should allow you password- free ssh >>>> access to localhost. Could you try this and see if this solves your >>>> problem? Post back to list to let me know how you get on - if you're >>>> still having problems I'll investigate further: >>>> >>>> ssh-keygen -t rsa >>>> (press enter 3 times for each of the questions asked) >>>> cat .ssh/id_rsa.pub | ssh <username>@localhost 'cat >> .ssh/authorized_keys' >>>> (enter your password) >>>> >>>> You should then be able to "ssh localhost" without having to enter a >>>> password. >>>> >>>> Best wishes >>>> >>>> Richard >>>> >>>> >>>> >>>> >>>> Paul Geeleher wrote: >>>> >>>>> Dear Members, >>>>> >>>>> I've followed the instructions as per the Puma Vignette in an attempt >>>>> to get the pumacomb() function to run on both cores of my dual core >>>>> notebook. >>>>> >>>>> I run the command "cl <- makeCluster(c('localhost', 'localhost'), >>>>> type='SOCK')", to create the cluster. >>>>> >>>>> This appears to open an SSH connection and prompts for a password. If >>>>> I type in the wrong password it re-prompts for the password, if I use >>>>> the correct one it seems to accept it but nothing happens. I can type >>>>> stuff on the screen but there is no response and I seem unable to >>>>> issue any subsequent commands. >>>>> >>>>> I've tried this in Ubuntu and OpenSUSE to the same result. Obviously >>>>> I'm missing something here? There seems to be virtually no >>>>> documentation out there other than the package vignette. >>>>> >>>>> Thanks, >>>>> >>>>> -Paul. >>>>> >>>>> >>>> >>>> >>>>> _______________________________________________ >>>>> >>>>> Bioconductor mailing list >>>>> Bioconductor at stat.math.ethz.ch >>>>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>>>> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >>>>> >>>>> >>>>> >>>> >>> > >

Paul, glad you got this working. The speed up you've seen looks similar to what I've seen running 2 cores rather than 1. Yes, I would highly recommend using the same versions of all packages on all machines - in fact, I think the general rule with Bioconductor is to always use the latest release versions of everything. Richard. Paul Geeleher wrote: > Yes Richard it seems that updating R and Puma has succeeded on Ubuntu. > Still no idea what's going on with SUSE, but my gut instinct is that > it could be an SSH problem. > > Running on both cores of the Pentium D has reduces execution time for > the Estrogen dataset from about and hour and a half to around an hour, > which makes a lot more sense that what I was seeing previously. > > My next step is to try to get this working between two networked > machines and maybe scale up from there, as the datasets we'll be > analyzing seem to take up to 10 hours on a single core. I'm presuming > both machines should be running the same version of R and Puma etc? > > Anyway I'll give it a go and see what happens. > > -Paul. > > > On Thu, Feb 28, 2008 at 5:46 PM, Richard Pearson > <richard.pearson at="" postgrad.manchester.ac.uk=""> wrote: > >> Hi Atro >> >> The parallel functionality of pumaComb was really developed in order to >> run the function much more quickly on large clusters. I wanted this to >> be as generic as possible, which is why I've made use of the >> functionality from the snow package (this means, for example, the >> function can be run utilising either MPI or a simple network of >> workstations connected via socket connections). When testing this on my >> laptop I realised the function could be sped up by running on a simple >> 2-node network on the same machine, so thought I'd include some >> instructions for how to do this in the vignette. I'm sure there are far >> better solutions to maximising the speed on a single machine (e.g. your >> suggestion of simply starting another process in the same session, or >> the suggestion from Martin Morgan on this thread), which I hope to chase >> up when I get the time. >> >> Best wishes >> >> Richard. >> >> >> Atro Tossavainen wrote: >> >> makeCluster is a function from the snow package, and I also can't find >> >> much information about setting up "socket" clusters. However, I think >> >> the fundamental problem here is that you can't ssh to localhost without >> >> a password. >> >> >> > >> > Of course this begs the question of why does one need to ssh to >> > localhost at all rather than just start another process in the >> > session that is already logged in...? >> > >> > >> >> >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at stat.math.ethz.ch >> https://stat.ethz.ch/mailman/listinfo/bioconductor >> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >> >> > >