[Soekris] net5501 and linux results

[Jan Ceuleers]

Hans,

Hans Harder wrote:
> Jan,
> 
> never looked at it, seems it was default on.
> When I test it with or without readahead, it makes no difference.
> 
> Hans

I would not expect to see a performance difference depending on whether 
read-ahead was on or off if the test method is hdparm -tT, because I 
believe that it only tests sequential reads anyway.

Any difference would only be discernable in a test that involves 
significant seeking.

I'm just wondering whether, on a solid state device, it makes sense to 
invest in read-ahead in order to avoid the situation where a process is 
accessing a file sequentially, but another process is causing the 
"heads" to be moved to another "cylinder" before the first process has a 
chance to access the next sector which would have been possible without 
moving the heads in the first place. Such a speculative investment aimed 
at avoiding needless seeks probably makes less sense on a solid state 
device than on a physical disk.

But the above is theory; I'd be interested in real-world opinions.

Cheers, Jan