Re: [Xenocide-programming] always PAQ

[Federico A. L. <fl...@ma...>]

> Hi guys,

> The SolidFile class has been temporary modified to accept CRC insteadof 
> String filenames. (With the strings the thing is pretty slower).
I dont have the code here, but the SolidFile is supposed to be supporting 
opening VirtualFiles by Strings and CRC alike. 

> The bad point is the speed of the addFile method, which can be improved.
> We can change the "append to table then sort" algorithm into something 
> quicker, like a mergesort. So we don't need the heavy reordering of the 
> entire table.
Dont worry too much about the addFile it will be used offline so the ones that 
have to be lighting fast are open-read-seek-close... open-read-close looks 
like they are lighting fast, try doing some random seek to see how much it 
takes.

> My implementation for now does not support the developers table, the 
> solidFileManipulator buffering (I discovered that the stdlib files have an 
> internal buffer too.. so we can avoid to build a second buffer) and the 
> defragmentation.
Better if you are using the stdlib internal buffer.

> Soon I'll release the entire packet.
Looking forward for it.

> Benchmark: "lock-release PAQSolidFile, 100 times"
> Elapsed time: 490 milli-seconds

> Benchmark: "add 5000 files (1byte length) with different crc numbers"
> Elapsed time: 2073 milli-seconds

> Benchmark: "deleting the 5000 previous created files"
> Elapsed time: 581 milli-seconds
Dont worry about the time it takes on these.


> Benchmark: "opening 1000 different files"
> Elapsed time: 10 milli-seconds

> Benchmark: "reading 1byte from every previous opened file"
> Elapsed time: 20 milli-seconds

> Benchmark: "closing the 1000 previous opened files"
> Elapsed time: 0 milli-seconds

These are the important ones. And pretty good figures I must say. Good job, as 
a suggestion try to use random accesses and random size reads, so you can 
check for more or less real time conditions.

Greetings
Red Knight