sblantipodi wrote:
my workloads does not fall in that category but previously, doing intensive task with small plan was possible, now is not possible.
(...)
for this reason, newer "middle/small" linodes are a big downgrade over previous linode and I think that most linoders uses small linodes.
I suppose there's no way to know definitively, but my strong suspicion is that most of the users of the smaller Linodes are actually more likely to be just like you than not, with workloads that do not fall into the category of those harmed by the CPU change. Therefore, I would disagree with your generalization that the newer Linodes are a big downgrade for most users; instead I suspect the opposite. Certainly my own experience bears that out.
Will there be users who were previously able to take sufficient advantage of the additional cores for this to be a downgrade even with the per-processor improvement? Sure. But it works the other way too - there will be users for whom the reduction in cores (and corresponding reduction in contention) see a benefit for their workloads. I have no way to judge which side is more prevalent though I have my suspicion that more fall into the second.
While it's mentioned occasionally in threads like these, contention (and processor scheduling overhead) is not something I think is emphasized enough, at least in terms of the impact I see in my own use. To be honest, I was extremely happy to see the core count changes, given my experience with the last upgrade, and applaud Linode for restructuring things the way they did.
When Linode expanded from 4 cores to 8 cores in the prior upgrade, my workload suffered noticeably (something I hadn't expected and actually got burned by when I took advantage of the upgrade). On my (largely smaller) Linodes steal rates rose a lot, and even with the bump in CPU type, my processing was not as effective. I ended up separating services into multiple Linodes to maintain my prior performance.
Even worse was the uncertainty as to whether going to higher plans would be a viable solution as load rose, given that even the entry plan already had access to the same core count as the higher plans. While I'm sure others saw improvement with that upgrade, for myself, I'm firmly in the camp that expanding the plans from 4 to 8 cores was a net negative.
So I'm quite happy with the recent change. To date, steal is vastly improved. It's hard to judge since there's so much else changing (and the ssds have such a large impact too), but CPU behavior appears more consistent and I appear to more readily be able to take advantage of the cores I do have access to. How effectively the CPU can be used is just as important as how many cores I have, to my mind. Certainly as an entire package, the new plans are night and day better than the old for my own workload.
So for me, I see the 8->2 core change as a solid improvement, and not a downgrade at all.
Of course, as with the other anecdotes quoted here, this is my own experience, and not necessarily a general result. However, it's experience that directly contradicts your statement about the new plan being a big downgrade - in general - for smaller Linodes (as I primarily use 2048s), so it's at least one counterpoint to that assumption.
-- David