virtual XRootD Workshop

Sören Fleischer wrote a ticket about this here: https://github.com/xrootd/xrootd/issues/1278. According to Andy, this issue can be solved by moving the ”oss.localroot” directive to the stanza that contains ”all.role server”

S. Atay is asking for a configuration option that allows an XCache Server to use more threads

S. Atay describes an issue with Root’s TNetXNGFile compatibility. The problem seems to lie in URL compatibility/parsing between ROOT’s TNetXNGFile and XRootD caching server. It will be good to have some insight where this might be rising? ROOT or XRootD?

Max Fischer: With xrootd being used more dynamically/opportunistically (Caches, Cloud, ?) we can no longer rely on every point-to-point connection being possible. This is not always obvious, because it can be triggered by faulty network configuration, IPv6/IPv4 routing differences, internal xrootd redirection, authentication, etc.
Andy: Agreed, that's a tough one given the distributed nature...

We would like to have a mechanism to monitor the state of the XRootD-HTTP(s) dataservers, and based on this available monitoring data, run replication mechanisms.
Additionally, we would like to discuss options to extend such a mechanism to dataservers that are not XRootD, but do speak xrootd-protocols

Andy: Well, good news is that xrootd should be fully integrated with HPSS. SRM replacement? CERN did that using xrootd's "prepare" plugin but it's a CERN thing that might not work for you. Another thing to discuss.

What we want is an aggregation of different HTTP(s) storages into a global namespace. Parts of that namespace should also be cached for a configurable time to reduce the amount of lookup operations.

It would be good to discuss XRootD's potential regarding QoS and GSI/FAIR needs

XRootd delivering files via HTTPs has significantly reduced throughput compared to HTTP

Disk-Cache-On-The-Fly is a prototypical setup that has similarities to XCache with the direct cache access directive. While XCache with dca has many advantages over dcotf, there might be parts of the setup that could be useful in the main XRootD-codebase.

Concurrent writes seem to put an XRootD-dataserver into an undefined state. Discussing why this happens and how to avoid this is the goal.

A "how to configure" and a short explanation of VOMS credential forwarding would be interesting, if time permits.