Elas­tic­i­ty with­out in­con­sis­ten­cy: Atomikos and the LogCloud (Un­pub­lished)

So you want elas­tic scal­ing. What about trans­ac­tion re­cov­ery?

It's the cloud era, and you want to build elas­tic ap­pli­ca­tions with dis­trib­uted trans­ac­tions. You would typ­i­cal­ly need a lo­cal disk to per­sist trans­ac­tion logs and per­form re­cov­ery. This would then de­feat the pur­pose of elas­tic scal­ing since by de­f­i­n­i­tion nodes can dis­ap­pear. In ad­di­tion, the cloud plat­form where you are de­ploy­ing to might not even sup­port disk ac­cess in the first place. So you would have to make this tough choice: ei­ther you get elas­tic­i­ty, or con­sis­ten­cy but not both.

We now have a so­lu­tion and it's called The LogCloud: a ded­i­cat­ed log­ging and re­cov­ery node that al­lows you to add / re­move process­es dy­nam­i­cal­ly in your ap­pli­ca­tion clus­ter. Trans­ac­tion pro­cess­ing will nev­er be the same since you can now have elas­tic­i­ty with­out in­con­sis­ten­cy!