Simon:
Further work with libvma gave good results with small packets.
Found that receive was faster than transmit - this was due to Turbo Boost being activated on Receive but not transmit. Scaling was in "Power Save" not "Performance".
Richard:
Upgraded GÉANT DTN-1 and DTN-2 to use ConnectX5 NICs and 100GBASE-LR4.
Still see drop in UDP throughput at 7813 Bytes. Also several % packet loss due to NIC rx_out_of_buffer. Fixed by setting RX ring to 4096 or 8192.
Experiment with libvma parameters to improve UDP receive.
Issue with iperf3 TCP not using libvma on receive.
Oleg:
As promised here is some links about xrootd protocol, however xrootd came from HE physic and deeply integrated to the another tools mainly "root framework" (https://root.cern.ch/) and special CMSSW ( CMS experiment SoftWare framework).
The the brief background about xrootd you can find in the slides G.
DONVITO : http://www.desy.de/dvsem/WS1112/donvito_talk.pdf
and xrootd redirection well explained :
https://twiki.cern.ch/twiki/bin/view/Main/RedirectorsSubscription
He has set up an xrootd end point but others are needed for tests.
Julich has the LOFAR LTA (Long Term Archive) running in production. If is running on dCache instance with enabled protocols: GridFTP, SRM, http/https, NFS4.1, based on 8 servers each with 10G network interface and has total 300 TB disk space. The LTA is placed in local DMZ .
LOFAR uses SRM and gridftp, but SRM can use either gridftp or xrootd.
Action: Oleg to put a diagram on LOFAR at Julich on the wiki.
Action: All identify other WP4 nodes to num xrootd.
Action: Introduce Oleg and people at RAL.
