xfs_and_dpx

In the gkrellm screen shot below, you can see what happens in the second run after capturing about 2.5 minutes of DPX files. Data continues streaming into the network interface at about 40 MB /sec,  but there is a pause in writing data to the 16-drive RAID array.  Beginning with the pause, CPU utilization for smbd goes from about 9 percent to about 16-17 percent.  The additional use is noted in the color blue, which is supposedly user versus system (which is orange) In this case, the xfs filesystem was created with an internal journal, so all activity is going to /dev/sdb. Here you can see another example where data is still coming into the network interace at about 40 MB/sec, but suddenly there is no writing to the disks for about 4 seconds. In this case, the xfs filesystem was formatted with an external log device (/dev/sdc).  When the data writing stops completely, the log writing continues, but at a much lower rate. The length of the gap -- when there is no data writing to disk -- seems to depend on the Linux vm settings.  The smaller I make /proc/sys/vm/dirty_ratio and /proc/sys/vm/dirty_background_ratio,  the shorter the gap is.