Re: [Dataspace-sabul] AMD64 Question.

[Yunhong Gu <yu...@la...>]

Hi, Steven,

Thanks for the information, it is detailed enough for me to trace the 
error.

For the "File size limit exceeded" error, it seems that the OS you are 
using on unet1 does not support file size exceeding 2^31. UDT support 
64-bit file size and we have tested it to tranfser a 100GB file. (At 
least the new version works fine with this, see below.)

For the MTU error, it needs to be set at both side, and UDT will choose a 
smaller value at the connection set up phase; so the appclient side also 
needs to set UDT_MTU as 9000, otherwise 1500, the smaller value, will 
still be used.

For the file corruption error, it may be caused by a bug in UDT that we 
already fixed. You can find a new version of UDT release (v1.2) at
sourceforge.net/projects/dataspace.

The project of lambdaftp from which you download UDT is obsoleted, please 
go to the dataspace project webpage to look for new release of UDT in 
future.

Please let me know if you have any further questions.

Thanks,
Gu



On Tue, 24 Feb 2004, Steven M. Carter wrote:

> On Mon, 23 Feb 2004, Yunhong Gu wrote:
> 
> > 
> > Hi, Steven,
> > 
> > I never tried UDT on an Opteron machine to test, but it should NOT matter 
> > that UDT is running on Xeon or Opteron, or any other CPU. There are 
> > assembly code in UDT for Intel CPUs; on other CPU it simply uses the 
> > system calls provided by OS. And since you have use it successeully for 
> > memory-memory transfer, it seems the problem is related to file 
> > operations.
> > 
> > Can you let me know more details of the problem? Does the file transfer 
> > always fail (no matter what the file size is)? Does it work fine on other 
> > machines (e.g, between two Xeon machines) with the same OS?
> 
> The memory to memory tests seems to work when unet2 (Opteron) sends to
> unet1 (Xeon), but not the other way.  The same is true with small file
> transfers.  Large file transfers die with "File size limit exceeded".
> 
> Also, setting UDT_MTU in appserver.cpp doesn't seem to work.  It still 
> uses an MTU of 1500.
> 
> My notes are at:
> http://www.ccs.ornl.gov/~scarter/net100/udpblaster/udt.html
> 
> Thanks,
> 
> Steven.
> 
> 

-- 
Yunhong Gu, PhD Candidate

Laboratory for Advanced Computing
University of Illinois at Chicago
SEO 700, M/C 249, 851 S Morgan St
Chicago, IL 60607-7045
T (312) 996-0305
F (312) 355-0373