Dear Don:

Thank you very much for the detailed report.

We'll follow up separately on the code patches.

On the performance front: I know nothing about your configuration, but I can 
imagine that you might be able to do even better. Would you be interested in 
having us work with you on identifying the bottleneck(s)?

Regards -- Ian.


On Aug 13, 2010, at 1:04 PM, Don Gilbert wrote:

> 
> Dear Globus folks,
> (maybe this should go to -dev rather than -user)
> 
> We have been testing GridFTP + UDT (gt5.0.2) for the Bio-mirror.net  project,
> which daily mirrors 100's of GB of biology data around the world. The results
> so far are very good. There have been some installation problems, and
> I have code patches if you want them.  A few of these I think would be
> usefully added to a future GridFTP release.
> 
> - Don Gilbert
> 
> Summary of GridFTP (gt5.0.2) patches
>  -- UDT change udt4/src/channel.cpp for Solaris10:
>       add UNIX to ifdef BSD, OSX for setsockopt max buffer
>   -- UDT globus configure{.ac}: update for *solaris* and *darwin* flags
>       (always compiles with -DLINUX otherwise, also fails to configure
>       on Linux-Powerpc, were it tries assembly for wrong processor)
> 
>   -- gridftp server, anonymous ftp directories:  limit to anonymous  user 
> home dir
>       in globus_i_gfs_control.c:globus_l_gfs_get_full_path() : add  anon_user 
> path restrictions
>       (I tried chroot() first but that was overly complex to decipher).
> 
>   -- gridftp server: log_transfers, UDT client IP was 0.0.0.0 always
>       globus_i_gfs_data.c:globus_l_gfs_data_end_transfer_kickout() :  small 
> patch
> 
>   -- globus-url-copy:  set dest file timestamp to match source file  time
>         (essential for -sync, usual FTP practice)
>       globus_url_copy.c: add globus_l_set_dest_timestamp() and add  int 
> src_mtime
>         in globus_l_guc_src_dst_pair_t and globus_l_guc_transfer_t  structures
> 
> Here are our results to date, with testing still in progress.
>     GridFTP TCP and UDT times for 113 GB transfer
>     from Bio-mirror.net (Indiana USA)
> 
>        Ping  Time(min)  TCP/  Distance
> Site    RTT  TCP   UDT  UDT    Km      Network Route
> --------------------------------------------------------------
> NCSA    10   139   138   1    200  Indiana - U of Illinois - NCSA
>              13.9  13.9            Megabytes/sec
> Purdue  17   125   125   1    500  In. - Chicago - Purdue, Indiana
>              15.3  15.3            Mb/sec
> ORNL    25   361   120   3   1200  In. - Chi. - Nashv., Tennesee - ORNL
>               5.3  15.9            Mb/sec
> TACC    37   616   120   5   2000  In. - Chi. - Huston, Texas - TACC
>               3.1  15.9            Mb/sec
> SDSC    65    --    --   -   3300  In. - Chi. - LA, California - SDSC
>     ...
> CSTNET 274    --    --  14  12000  In. - Michigan - Kreonet, Korea -  
> Beijing, China
>              0.48  6.67            Mb/sec (from preliminary 3GB test)
> --------------------------------------------------------------
>   Transfer times (minutes), and below speed in Megabytes/second,
>   for TCP and UDT, and the TCP/UDT ratio.
>   NCSA, Purdue, ORNL, TACC, SDSC are Teragrid.org sites in USA.
>   Land/sea line distance is given in Km.
>   RTT is network distance as average round trip ping time in ms.
>   TCP and UDT transfers were run simultaneously from each site.
> 
> -- d.gilbert--bioinformatics--indiana-u--bloomington-in-47405
> -- gilbe...@indiana.edu -- http://marmot.bio.indiana.edu/
> 

Reply via email to