Go to previous page / next page of Tier3 site log

25. 02. 2009 Worker Node to Fileserver write throughput

We currently have three Thumpers active in dcache. T. Punz had a number of jobs running which showed some I/O wait when writing in parallel to the dCache. The max write throughput was around 300 MByte/s =~ 2.5 Gbit/s, and generated by ca. 25 gridftp connections (we do not know the number of streams). The SE showed a symmetric load of 30 MByte/s read/write from the worker nodes plus 20-30 MByte/s read from the UI (dcap, ca. 4 processes).

6 worker nodes wrote with ca. 40 MByte/s =~ 335 Mbit/s, 2 with ca. 25 MByte/s. All showed some I/O wait over the period of the intensive writing, and little CPU usage. Based on the load graphs for the machines, it seems that each machine ran 6 to 8 of these jobs.

-- DerekFeichtinger - 25 Feb 2009

Go to previous page / next page of Tier3 site log * wn-load-day-2009020935.png:

* wn-cpu-day-2009020935.png:

* wn_netw-day-2009020935.png:

* fs-netw-day-2009020935.png:

* wan-movers-day-2009020935.png:

* se-netw.day-20090225.png: