On April 17, 2018 9:02:14 AM CDT, [email protected] wrote: >On Tue, Apr 17, 2018 at 08:20:47AM -0400, Russell via talk wrote: >> Currently I have two versions of the same os on the same machine. One >on M.2 Xpoint nvram and one on a standard SSD. I'm playing around with >tweaking before I do a final config. So far the Xpoint direct hw access >appears 3x as fast as the SSD while real world throughput shows up >about twice as fast on the Xpoint, recent INTEL cache fencing >notwithstanding. >> >> dd if=/dev/zero bs=1M count=1024 | md5sum >> 1024+0 records in >> 1024+0 records out >> 1073741824 bytes (1.1 GB, 1.0 GiB) copied, 1.35008 s, 795 MB/s >> cd573cfaace07e7949bc0c46028904ff - >> >> 795 is just under twice as fast as writing to the conventional SSD. > >That command didn't write anything to anywhere.
It wrote a bunch of zeros to a virtual file. Perhaps even touching a tmp file along the way. Even if it didnt touch tmp, it wrote the zeros someplace in order to perform the count. I was just trying to comment on the speeds of the two installs relative to the respective disks the OS runs from. I'm sorry you didn't understand that. Perhaps I should have said running the OS from the two different drives, irrespective of all the other disk writes which may happen when the OS operates normally when calling dd from a GUI. > >It tests how fast md5sum can calculate the checksum of 1GB of zeroes. > >Certainly in no way testing any disk speed. Reasonable test of CPU and >ram speed perhaps. Often tests provide side channel results which are not part of the expected normal metric but quantifiable data arises none the less. My appologies for the misunderstanding. > >-- >Len Sorensen -- Russell --- Talk Mailing List [email protected] https://gtalug.org/mailman/listinfo/talk
