[Go-essp-tech] Progress and Problems with P2P

Wed Feb 1 17:54:26 MST 2012

On Feb 1, 2012, at 4:20 PM, Sébastien Denvil wrote:

> Hi Jennifer, all
> 
> see below:
> 
> Le 01/02/2012 17:08, Jennifer Adams a écrit :
>> Hi, Everyone --
>> I have been working with Luca and Gavin and Estani on testing the P2P system. I'm happy to report that the many significant parts of my workflow for downloading data can now be fully automated:
>> 1. Search for available data sets that meet my desired requirements (e.g. decadal1980/atmos/mon/Amon, all models, all members, selected variables)
>> 2. Compare search results to a list of what I've already got
>> 3. Build script to download and run wget scripts for datasets I still need
>> All this using shell scripts and without touching a browser!
>> 
>> There are still a few wrinkles, however. The worst of these is that not all data nodes are authenticating certificates properly.
>> 
>> These data nodes are working:
>> bmbf-ipcc-ar5.dkrz.de
>> cmip-dn.badc.rl.ac.uk
>> dias-esg-nd.tkl.iis.u-tokyo.ac.jp
>> esgdata.gfdl.noaa.gov
>> esg.cnrm-game-meteo.fr
>> norstore-trd-bio1.hpc.ntnu.no
>> pcmdi11.llnl.gov
>> vesg.ipsl.fr
> 
> To help you diagnose the source of your PKI issues :
> 
>> These data nodes are not:
>> dap.cccma.uvic.ca
> 
> It works for us as I write. PCMDI openid if it matters.
I believe it does matter which OpenID you use, and PCMDI is the one that works. Gavin told me that he didn't have many problems with downloads from these nodes either. I think it makes a difference whether you are wgetting from a data node or from someplace else. Gavin seems to have the keys to all the castles.

> 
>> esg.nccs.nasa.gov
> We have never been able to download from there using PKI.
I got some email directly from Ellen Salmon at GSFC who explained that they have been working for months to get all the upgraded ESGF software packages to work properly, but there is hope that it will be working soon (~weeks). I know this has been a chronic problem for data node administrators.

> 
>> bcccsm.cma.gov.cn
> We have been able to download from there using PKI recently.
This one fails 95% of the time for me. Estani says BCC is a special case.

> 
>> tds.ucar.edu
> We have been able to download from there using PKI recently.
This may another case of it working for requests from other data nodes. 

> 
>> pcmdi9.llnl.gov
> We used pcmdi3 up to know (it worked), will give a try to pcmdi9
Gavin has been tweaking this one lately. I haven't tested it in the past few days. 

> 
>> esg-datanode.jpl.nasa.gov
> Did not tried yet.
> 
>> 
>> P2P wget scripts to download data from the second set of data nodes always fail completely (for me, a "pure client" user). I know that Luca and Gavin and Estani are working to fix these problems, but here's some encouragement to the data node administrators to help them resolve this issue. Wgets that rely on the quickly-expiring authorization tokens ought to be deprecated as soon as possible -- they are the biggest, most irritating source of errors in the whole system (my opinion only).
>> 
>> The second problem is that the wget scripts don't often succeed in getting all the files they are configured to grab. Of the 64 scripts that I ran this morning, 55 were incomplete (that's 86%). The errors in log files are all "ERROR 403: Forbidden." I was running the 64 scripts at one time, but each one was grabbing its list of files in order -- I am not parallelizing the wgets in the way that some other users have described. When I rerun the wget scripts from the incomplete runs, some of them finish the job on the 2nd try, others will take as many as 8 tries before all the files are in.
>> 
>> These are the data nodes that are not allowing me to complete the download (for the decadal1980 example mentioned above).
>> esg.cnrm-game-meteo.fr
>> vesg.ipsl.fr
>> dias-esg-nd.tkl.iis.u-tokyo.ac.jp
>> cmip-dn.badc.rl.ac.uk
>> bmbf-ipcc-ar5.dkrz.de
> 
> I notice that those node seems to be far away from your home location.
> May I suggest you to give those options to wget ; from our experiences with synchro-data they appear to improve substantially the success rate :
> wget --timeout=20 --tries=10
I will definitely give those options a try. 

> 
> A lot has been said about this issue.
I'm sorry I didn't go searching through the forum archives before writing about wget failures. I didn't join this list until fairly recently, when I started to actively work on downloading CMIP5 data. I was subscribed to plain go-essp (not tech) and there was no traffic at all. Here is where the the players are!! I'm glad I'm here … lately I've been following the threads in this forum with more interest than in my own GrADS forum. 

> Difficult to attribute properly the responsibility of those failure to such or such components/parties. Give a try to the parameter above.
> 
> If you send me the IP adress you used we could investigate the vesg.ipsl.fr case.
The boxes I have been using are inside a firewall and do not have public IP addresses. I did a test by pinging my public GDS from there and in the log file it came up as cola.gmu.edu. 

> 
>> I suspect there are some throttling settings in place to limit the number of wgets that a data node will allow at any particular time. I think my use of the data nodes is reasonable, and these throttles are set too high.
> 
> It's quiet hard to set this up properly using a tomcat free version. Some commercial tomcat version offers easiest way to achieve that. Up to know (on our node : vesg.ipsl.fr) we haven't done anything specific regarding that.
Huh! Well, if you think it would be worthwhile, I would be willing to try to slam your data node with a barrage of wgets from one of our servers that has a resolvable IP and we could see what it takes to receive the Forbidden error on a default tomcat implementation. 
--Jennifer

> 
> regards.
> Sébastien
>> The dreaded "Forbidden" errors may be related to another problem, but are still a nuisance no matter what the reason. YAO (Yet Another Obstacle). Please put this on the high priority list of things to fix, right behind the certificate authentication issue.
>> 
>> Respectfully submitted,
>> Jennifer
>> 
>> 
>> 
>> 
>> _______________________________________________
>> GO-ESSP-TECH mailing list
>> GO-ESSP-TECH at ucar.edu
>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
> 
> 
> -- 
> Sébastien Denvil
> IPSL, Pôle de modélisation du climat
> UPMC, Case 101, 4 place Jussieu,
> 75252 Paris Cedex 5
> 
> Tour 45-55 2ème étage Bureau 209
> Tel: 33 1 44 27 21 10
> Fax: 33 1 44 27 39 02
> 
> 
> _______________________________________________
> GO-ESSP-TECH mailing list
> GO-ESSP-TECH at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech

--
Jennifer M. Adams
IGES/COLA
4041 Powder Mill Road, Suite 302
Calverton, MD 20705
jma at cola.iges.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/go-essp-tech/attachments/20120201/25b28c5b/attachment-0001.html 