On Sunday 29 June 2003 04:31, Zack Gilburd wrote: > On Saturday 28 June 2003 08:36 am, Dhruba Bandopadhyay wrote: > > Hello, > > > > Previously hdparm (-Tt) 5.3 gave me results of 752/41. Today I > > upgraded to hdparm 5.4 and I get the following results! > > > > $ hdparm -Tt /dev/hda > > > > /dev/hda: > > Timing buffer-cache reads: 2912 MB in 2.00 seconds = 1454.55 > > MB/sec Timing buffered disk reads: 120 MB in 3.01 seconds = > > 39.91 MB/sec > > > > Is it telling the truth or am I seeing things? If this is true > > then my hard drive has become twice as fast within the time it took > > to perform the upgrade. > > > > What is going on here? > > I can almost garuntee you that your disks are not doing buffer-cache > reads at 1.42 gigabytes per second (as the output reads). Your disk > reads look normal, though.
Buffer reads are not from disks, but from memory buffers. And many machines can move memory content from place A to place B at that pace. For example my memory bandwidth execeeds the bandwidth of my FSB, which is 610 MHz and thus provides healthy theoretical bandwidth of 4.8GB/s. The memory have theoretical bandwidth of about 6.2 GB/s as I have DUAL channel DDR memory clocked as 390 MHz. The fact that the result changed to twice as fast in the later version of the hdparm is most likely a case of new test code, which does less CPU work per transfered (read copied) byte. -- [EMAIL PROTECTED] mailing list