Hacker News new | comments | show | ask | jobs | submit login

"Is this really something that needs testing?"

Yes. Running tests to validate your assumptions is a big part of robust software engineering. In this case the results were unsurprising but not uninteresting.




I'll do some testing this weekend by mounting an SSHFS as my swap partition and removing all but a 256MB RAM module then opening several dozen instances of FireFox. My calculations show that this will have terrible performance but I want to ensure that my software engineering is robust.

Testing your assumptions is something that you're supposed to do when you hit a wall, not when you're driving through a field.


If you've been using computers for long, you should already know VM thrashing can murder a system, to the point of making it unresponsive and needing a reboot. So this is an assumption you have backed by direct experience. And of course, probing it won't tell you anything.

But do you have enough information about how Redis accesses memory under the benchmark in question, combined with the OS page replacement strategy, combined with the characteristics of SSDs, to know the results beforehand? You can guess, for sure; but do you know?

If we all follow your approach, we'll never be surprised unless we get stuck; and if we know what we think we know as well as you seem to think we know it, we shouldn't get stuck in the first place. We should have assumed that we would have gotten stuck, and avoided it.

The article has relatively low value in terms of information content, but the mindset is to be commended. It should have given the author better intuitions about the 3 factors mentioned above. Modern, non-budget systems very seldom thrash; there's a younger generation coming along who've never experienced systems frozen in that way.


I think that there is more to the pool of information than the points you rightly outlined. Other key pieces of data are that Redis is designed to be extremely fast and that it has a strong reliance on the speed of the memory it is operating on. Those facts, coupled with the statistics sniglom mentioned above, very strongly indicate that performance will be terrible.

An analogy I can think of is testing if a stock Ford Fiesta can reach the speed of sound. You know what the engine is capable of, the environment it is operating in and the tires it is running on - you simply don't need to floor the accelerator to come to a conclusion.

That saying about picking ones battles comes to mind. The mindset is certainly of a sharp character, but what good is a knife without a hand to guide it? The map is not the territory but it does save a lot of time if used strategically.


The difference is that SSD backed key value stores are something people are actually interested in (Google LevelDB, Tokyo Cabinet, Twitter fatcache), and Firefox with SSHFS swap partition is not.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact

Search: