[mnet-devel] other explanations of performance problems

[Zooko]

6.  The current design has it reading a bunch word (4 bytes) from the file for 
each block generated.  Unfortunately, those words are evenly distributed 
throughout the file.  For example, if M is 64, then it reads the first word of 
the file, then skips the next 63 words, then reads 1 more, etc., to the end of 
the file.  Then for the next block it reads the 2nd word, skips 63, reads the 
next, etc.

So we could easily be killed on "disk access scales superlinearly with size of 
file" [*] this way.

We should test this (I could do it using my tmpfs partition -- an in-ram 
filesystem), and if it's the problem, one fix is to sacrifice systematicity in 
return for the performance of using sequential words instead of evenly spaced 
out words.

--Z

[*] It's actually not superlinear -- it maxes out with M = 254 or so, which 
mreans the file is of size 4 GB or so, but probably it is way too slow before 
it reaches the ceiling.


-------------------------------------------------------
This SF.Net email sponsored by: Free pre-built ASP.NET sites including
Data Reports, E-commerce, Portals, and Forums are available now.
Download today and enter to win an XBOX or Visual Studio .NET.
http://aspnet.click-url.com/go/psa00100003ave/direct;at.aspnet_072303_01/01
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel