Re: Upcoming BRAT changes

From: Christian Holm Christensen (cholm@hehi03.nbi.dk)
Date: Thu Dec 13 2001 - 10:22:00 EST

  • Next message: Flemming Videbaek: "Fw: Geo files"

    Hi Flemming et al, 
    
    On Wed, 12 Dec 2001 22:17:14 -0500
    "Flemming Videbaek" <videbaek@sgs1.hirg.bnl.gov> wrote
    concerning "Re: Upcoming BRAT changes":
    > > It's amassing how much time you have when you're waiting for data from
    > > the quasi-stable CRS, especially when jobs get killed in the middle of
    > > absolutly nothing.  Mind you, processing one 100K run with only the
    > > TMA, SMA, and BB in takes about 10 min.  Seriously, the CRS is not at
    > > all optimal, and I think implementing a schedular on top is a bloody
    > > waste of time - RCF should provide that - and they should provide a
    > > search engine for our mailing-list archives too!  Now, please!
    > 
    > This two items has little to do with each other - so
    > i) Attempting to read 100K events((~5Gb)  in about 10 min means you are
    > pushing the HPSS system to its limit regardsless of the crs
    > implementation .
    
    Ok, I was a bit wage.  What I meant to say, is to process one sequence
    with only TMA, SMA and BB modules, takes about 6-7 min.  For a 100K
    trigger 6 events, you have around 10 sequences (which is more like
    10Gb - the number depends on other triggers as well), and with some 30
    nodes in CRS you can get the whole job done in around 10 min. 
    
    What I was aluding to in terms of a schedular was the proposal from
    Tony Chan that we implement another schedular on top of the one in the
    CRS already - apparently some of the other experiments has done that.
    
    > If you go through the numbers there is no way ANY set of nodes can
    > be fed off tape at kept busy. If we do little analysis do not expect
    > this to get done quickly. The stability is another matter, but as I
    > understand it has less to do with the farm as with the socalled DCE
    > problems related to one arm of IBM not being willing to share
    > information with another !! 
    
    Hmm. 
    
    > ii) The list servers are NOT maintained by RCF, but by ITD, and do
    > not expect a response from them. You can possible blame Brahms, but
    > we also have to prioritize what is done- and itmes like DAQ,
    > backups, and systems maintance is high on that list. 
    
    Ok, so the list server is from ITD, but the web-archives is on RCF
    web-pages.  The thing is that it's really simple to set up a search
    engine on some machine - just install HtDig and add the web-addresses
    you want to index, and you're rolling.  I don't see why the RCF or ITD
    couldn't do that for us. 
    
    Another possiblity is to make an arrangement with Google - they
    provide free customised search engines for academia. 
    
    I really think a search engine would be a nice thing.  And I think we
    should not spend time doing it, but have ITD or RCF do it. 
    
    Yours, 
    
    Christian Holm Christensen -------------------------------------------
    Address: Sankt Hansgade 23, 1. th.           Phone:  (+45) 35 35 96 91 
             DK-2200 Copenhagen N                Cell:   (+45) 28 82 16 23
             Denmark                             Office: (+45) 353  25 305 
    Email:   cholm@nbi.dk                        Web:    www.nbi.dk/~cholm
    



    This archive was generated by hypermail 2b30 : Thu Dec 13 2001 - 10:22:38 EST