Trow­el is writ­ten around AnyEvent, an event loop pro­gram­ming frame­work. That means the pro­gram is de­v­id­ed in units that can con­tin­ue to run with­out wait­ing for each oth­er. This is con­ve­nient when we're us­ing net­work con­nec­tions that are apt to hang - like twit­ter - so the parts that have noth­ing to do with the net­work con­nec­tion can keep go­ing in­de­pen­dent of the state of the net­work con­nec­tion (in the­o­ry). The down­side is that you nev­er know in what or­der the var­i­ous units in­ter­act and to pre­vent race con­di­tions in which two or more units try to change the same cen­tral data at the same time the in­ter­ac­tion has to be atom­ic. For Trow­el this means one unit push­es the col­lect­ed tweets onto an ar­ray while the oth­er unit pops them of to dis­play them, so at my lev­el of ab­strac­tion the units don't have to ex­plic­it­ly wait for each oth­er to fin­ish chang­ing the con­tents of the ar­ray. I as­sume there's one proces­sor that is chang­ing val­ues in mem­o­ry and in one proces­sor cy­cle the push and the pop can't be done si­mul­ta­ne­ous­ly.

With lin­ux on Cell Broad­band Engine hard­ware it's pos­si­ble to do awe­some things. But the trick is that the Cell con­sists of a rather slow proces­sor and about 6 su­per fast spe­cialised proces­sor units. To cre­ate pro­grams that uti­lize all the proces­sors it is re­quired to write them about the same as with an event loop. The pro­gram would re­quire mul­ti­ple threads that try to keep all the spe­cial­ized proces­sors busy, and the hard part is to man­age the data that is used in all those threads. be­cause if 6 units want to pop or push the same ar­ray at the same time you're sud­den­ly look­ing at one of those that will have to wait up to 5 cy­cles, and a big­ger chance that it should be af­fect­ed by what the oth­er 5 units have been do­ing.

Get­ting bored of some repet­i­tive task in the Gimp I de­cid­ed to write a script that would do the same task with the push of a but­ton. The chal­lange was that it had to be done in Scheme. Which is a pain. Just like with the Cell you're con­front­ed with how the un­der­lay­ing hard­ware re­al­ly works and not how your brain works. After a while your brain starts to work like the hard­ware though and it gets eas­i­er, but it shows that writ­ing op­ti­mized code is still hard. And why a lot of code is still slow on to­day's blaz­ing­ly fast hard­ware...