Previous | Next --- Slide 20 of 42
Back to Lecture Thumbnails

In my understanding, in the first step, graph A is read-only. We use A to generate B and then use B to update A. So does this mean that to compute a graph we need twice space in memory?


If @monekyking is correct then do programmers have to trade-off between memory and speed when using GraphLab?


The tradeoff between memory and speed is a common tradeoff in numerous parallel systems. For our final project my team is writing a parallel library for the Raspberry Pi's GPU, which has extremely limited memory on the processor (4K for 12 SIMD cores). We have to make a tradeoff between using all of this memory and having a faster implementation or using less but taking more time to do DMA stores back to host memory.