3 ms·

If you investigate what they actually do to achieve those numbers, it's much less simple than just rewriting Cassandra in C++. For example, they use their own T

If you investigate what they actually do to achieve those numbers, it's much less simple than just rewriting Cassandra in C++. For example, they use their own TCP stack and make use of vector intrinsics.

sandGorgon 10y ago
Hey.. "vector intrinsics" looks very cool. Thanks for mentioning that! so what you mean is that, even after throwing facebook scale resources at java.. it is possible for a <10 people team to get 10X performance over java using the features that you mention. That's a huge loss of face for java IMHO
- zigzigzag 10y ago
  You aren't comparing the same program written in two languages. Seastar stuff is written by C++ performance experts who are fanatical about tuning, and does all kinds of unusual far-out things that Cassandra doesn't do to get high performance.
- throwawayish 10y ago
  > That's a huge loss of face for java IMHO Because the JVM JIT, like every other compiler, sucks at developing vectorized algorithms ad-hoc; a task usually carried out by human experts in that?
- echlebek 10y ago
  > so what you mean is that, even after throwing facebook scale resources at java.. it is possible for a <10 people team to get 10X performance over java using the features that you mention. That is not what I mean. Writing a userspace TCP stack isn't a feature of C++.
- pron 10y ago
  Explicit use of vector intrinsics are not the source of any 10x performance boost, nor anything to do with C++ in particular. Again, only a small (but probably not minuscule) portion of the difference has to do with Java vs. C++. The bulk of the difference is due to all sorts of optimizations, most of them could have been done in Java as well. But the ScyllaDB people are more experienced in C++ than in Java, and as they use sequential code anyway, there isn't a big downside for using C++ -- certainly not for them -- so it was the better choice. From what little I know, the reasons why such optimizations weren't done in Cassandra are because 1. the people working on it aren't low-level optimization experts, but more importantly, 2. because the performance was good enough.

sandGorgon 10y ago

Hey.. "vector intrinsics" looks very cool. Thanks for mentioning that! so what you mean is that, even after throwing facebook scale resources at java.. it is possible for a <10 people team to get 10X performance over java using the features that you mention. That's a huge loss of face for java IMHO

zigzigzag 10y ago
You aren't comparing the same program written in two languages. Seastar stuff is written by C++ performance experts who are fanatical about tuning, and does all kinds of unusual far-out things that Cassandra doesn't do to get high performance.
throwawayish 10y ago
> That's a huge loss of face for java IMHO Because the JVM JIT, like every other compiler, sucks at developing vectorized algorithms ad-hoc; a task usually carried out by human experts in that?
echlebek 10y ago
> so what you mean is that, even after throwing facebook scale resources at java.. it is possible for a <10 people team to get 10X performance over java using the features that you mention. That is not what I mean. Writing a userspace TCP stack isn't a feature of C++.
pron 10y ago
Explicit use of vector intrinsics are not the source of any 10x performance boost, nor anything to do with C++ in particular. Again, only a small (but probably not minuscule) portion of the difference has to do with Java vs. C++. The bulk of the difference is due to all sorts of optimizations, most of them could have been done in Java as well. But the ScyllaDB people are more experienced in C++ than in Java, and as they use sequential code anyway, there isn't a big downside for using C++ -- certainly not for them -- so it was the better choice. From what little I know, the reasons why such optimizations weren't done in Cassandra are because 1. the people working on it aren't low-level optimization experts, but more importantly, 2. because the performance was good enough.