On February 15, Darren Williams wrote:
> We are currently benchmarking Linux scalablity, so to avoid
> disk latency issues on filesystem benchmarks we have been
> generating each filesystem of interest on ramdisk then
> running osdl-re-aim-7 with disk only tests. With xfs I start
> seeing the looping messages at about 50 clients, ( using a 2M
> FILESIZE and 128M DISKDIR in the reaim.config ) with a 8way
> configuration. (This is just for your records)
> What effect can this have on the filesystem transactions,
> does it slow writes down due to the out of order logs. We
> are currently using the O_SYNC flag on open.
What this message is saying is log buffer writes are coming in as fast
as the callbacks on them are being processed.
I put this this message in the log i/o completion callback about 5
years ago, when fixing out-of-order callback problems. On some
systems, buffer completion callbacks are run from an interrupt thread.
I had changed xlog_state_do_callback to potentially be an infinite
loop, and I was concerned about the overall impact on the system if it
were hold the interrupt thread for a long time and possibly hold off
The only thing "wrong" is that your CPU and your I/O system are really
fast :-) You can safely delete the printing of this message if the
message printing process is interfering with your benchmark.