 |
 |
Hi, Thanks for your reply. Le 04/10/2013 02:19, Andrew Hanushevsky a écrit : > 1) Your configuration file has to allow reading the file from where it > was placed, Should that line solve this (from our current setup): xrootd.export / > 2) The owner has to be the same the user running xrootd (or at least the > file has to be readable by that user) Fine. > 3) If you use an authorization scheme (I don't think you use the xrootd > scheme), privileges have to be correct. That's a key point. For a "normal" operation, the command comes from the DPM head-node, and afaik the authentication/authorization has been done at that level and some secured transaction between the head and the disk server allow it to be acknowledged on the disk-server (using the dpmxrd-sharedkey). Here I cannot use this mechanism that is currently configured... > I hope this helps. However, I do want to point out that if you adopt a > single file scheme your results will not be valid and likely not even > comparable from protocol to protocol. This is because the file system > will happily cache some portion of the file, and you really don't know > which portion it might be. While reading files of several hundred > gigabytes minimizes this problem it doesn't eliminate it, especially > with the way some protocol implementations handle the file. Testing with a relatively small and cached file or simultaneous access to lots of big files spread on all the server raid array wont test the same thing (memory -> network ability or disk->cache bottlenecks, etc.). I'm not really comparing protocols here (I've read several reports about it and it seems that Xrood is the best, so I want it to work), but more to quantify what we can get in practice with our setup and where are the bottlenecks. Thanks again, Yannick -- Yannick Patois <[log in to unmask]> IPHC - IN2P3 / CNRS - 23 rue du Loess 67037 Strasbourg Tel: 03 88 10 61 83 ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1