tag:blogger.com,1999:blog-7506175133404439580.comments2019-10-12T06:45:45.338+01:00Droplets FormingUnknownnoreply@blogger.comBlogger3125tag:blogger.com,1999:blog-7506175133404439580.post-50158299074915639092008-06-13T10:50:00.000+01:002008-06-13T10:50:00.000+01:00One process was what I was aiming for - hence the ...One process was what I was aiming for - hence the 'one thread' as the first in the initial version. <BR/><BR/>I suspect that seeking forward would be faster - that way the read head is doing the least complicated thing (although I am very hazy on what is fast or slow for a hard disk). <BR/><BR/>Randomly seeking for a single record would add more overheads because of the seek time and disk buffering. It needs to have time to get to the full streaming speed, then skip past enough of the file to offset the new seek time. If the seek time is 10ms, and the disk reads at some 50MB/s, then the seek time will cost you something like 0.5MB. <BR/><BR/>Then there is the buffering delay - usually disks have a buffer of 8MB or so, so it makes sense to pull this much data at once, then skip ahead. Hopefully with command queueing and speed matching, the hard disk's built in cleverness would be fully employed. <BR/><BR/>Experimentation definitely a good thing. <BR/><BR/>PhilPhil Hhttps://www.blogger.com/profile/01941670161404784895noreply@blogger.comtag:blogger.com,1999:blog-7506175133404439580.post-13582717550408402692008-06-13T06:26:00.000+01:002008-06-13T06:26:00.000+01:00Nice one!But each parallel access to the file is g...Nice one!<BR/>But each parallel access to the file is going to hit I/O so why not have one process randomly sampling the file.<BR/><BR/>Do we need to always seek forward in the file or can we randomly seek for a record to process?<BR/><BR/>Needs some experimentation methinks!<BR/><BR/>- Paddy.Paddy3118https://www.blogger.com/profile/06899509753521482267noreply@blogger.comtag:blogger.com,1999:blog-7506175133404439580.post-83810258167869612622007-10-07T23:34:00.000+01:002007-10-07T23:34:00.000+01:00http://arstechnica.com/journals/linux.ars/2007/09/...http://arstechnica.com/journals/linux.ars/2007/09/10/nokia-opens-the-hildon-input-method-frameworkPhil Hhttps://www.blogger.com/profile/01941670161404784895noreply@blogger.com