openalto / ietf-hackathon

3 stars 7 forks source link

FTS setup and eval #53

Open giralt opened 2 years ago

fno2010 commented 2 years ago

I have made a fork of hackathon docker environment with some predifined configurations and utility scripts to handle this: https://github.com/fno2010/ietf-hackathon/blob/fts/docs/fts_usage.md

Current version fixed the previous "Connection refused" issue. But when I try to generate the setting to demonstrate the potential race condition of FTS, the following non-deterministic error occurs when the number of concurrent transfers grows:

TRANSFER [116] Error on XrdCl::CopyProcess::Run(): [ERROR] Operation expired

It seems that we get the same issue as this one: https://github.com/scikit-hep/uproot5/discussions/355

fno2010 commented 2 years ago

The limit to the maximum number of parallel I/O requests can be the root cause.

[root@11bdaf907fc3 rucio]# XrdSecPROTOCOL=gsi XRD_REQUESTTIMEOUT=10 xrdfs xrd1 query config pio_max
5
fno2010 commented 2 years ago

Change XRootD client.conf may fix this issue somehow. But need more tests to verify.

fts-optimizer-evolution

Note: The first throughput reduction of the vlink xrd1 -> xrd3 occurred when new transfers were submitted to the vlink xrd1 -> xrd2. But the active decision was still increased in some time intervals because of the jitter.