Archives of the TeradataForum
Message Posted: Mon, 07 Mar 2005 @ 17:02:39 GMT
| Subj: | | Re: FastExport performance vs. Fastload |
| |
| From: | | Victor Sokovin |
| | The data transfer is way below the connection capacity. The physical throughput is 1Gb/s (or 1000Mb/sec) and my average fexport is 2000
blocks per minute. Even assuming the max block size 64k it is way below the capacity. | |
Yes, but who said FastExport would use all the network capacity you might happen to have? I don't know whether you had a chance to measure the
speed of a simple ftp connection but perhaps even that basic process would not show anything close to 1000Mb/sec.
| | As fare as disk - I tried two types of processing: | |
| | 1) Writing to SAN attachment made up of small disks of 36G. | |
| | 2) Writing to named pipes to avoid I/O. | |
| | In case 2) I created named pipe in C++ program. This program was reading data as it comes from fexport and did not write anywhere just
discard, to avoid any I/O on the disks. | |
What kind of volumes are you testing this on? You seem to call 36 GB a small disk but cannot see any difference between writing or not writing
data to the disk. Strange.
I personally would not introduce pipes as they add complexity to the test.
Did you have a chance to inspect the FastExport log file? It usually contains the block size, number of blocks exported and the timings per
phase. Perhaps you could post that data to add some more context to the question?
Regards,
Victor
| 
