There's some stuff that isn't horizontally scalable either intrinsically, or the cost benefit hasn't gotten there yet. Back in the day I heard rumors.that Borg scheduler nodes needed a lot of cores on the local box _and_ a high per thread IPC. I could see Google happily spending $5k for 300 more MHz.
Comments
There's some stuff that isn't horizontally scalable either intrinsically, or the cost benefit hasn't gotten there yet. Back in the day I heard rumors.that Borg scheduler nodes needed a lot of cores on the local box _and_ a high per thread IPC. I could see Google happily spending $5k for 300 more MHz.