Pascal Jasmin wrote:
> The timings you should be comparing to are a database engine's 2k data
> append query on a 2MB table.  Even if its done in place, there's reliance
> on OS's append files facilities, and even if they are faster than "erase
> old file, create new file , add contents (old data + new data) to new
> file, rename new file to old file", its not necessarily as fast as disk
> access on smaller files.

I understand your example, but I believe (or hope?) that the jmf situation
is different.  First, the file has been allocated.  Second, I am assured
by Roger and Eric that append is done "in place".  I assume that this
means we are not relying on an OS append at all, but rather modification
of an existing file whose size will never change, and the initial segment
of the file will never be rewritten.  Of course, this may still mean that
disk blocks have to be collected and written, but I would expect linear
performance rather than the quadratic performance I seem to be observing.

John


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to