First: thanks for your work on Protocol Buffers. I've used it fairly extensively for RPC communications between C++/Java clients and a Java service. It made things so much easier to get native objects in each language using a well-defined protocol.
One thing that bugged me about Protobuf is that it provided a skeletal mechanism for RPC (e.g. RpcController/RpcChannel) but later deprecated the use of that mechanism in favor of code-generating plugins. Since Cap'n Proto is billed as an "RPC system", do you have plans to include a more fleshed-out version of RPC than was provided in Protocol buffers? Having abstract classes for event-handling and transport mechanisms is a good idea for extensibility but it sure would make it easier for your users if there was at least one default implementation of each.
I imagine that Google has standard implementations of these things internally but balked at trying to support them for multiple languages as an open source project.
Yes, in fact, the next release of Cap'n Proto (v0.4) is slated to include RPC support. There are some hints on what it might look like in the docs already:
The reason Google never released an RPC system together with protobufs is because Google's RPC implementation simply had too many dependencies on other Google infrastructure, and wasn't appropriate for use outside of Google datacenters. There were a few attempts to untangle the mess and produce something that could be released, but it never happened.
The public release had support for generating generic stubs, as you mentioned, but it was later decided that these stubs were actually a poor basis for implementing an RPC system. In their attempt to be generic, their interface ended up being rather awkward. We later decided that it made more sense to support code generator plugins, so that someone implementing an RPC system could provide a plugin that generates code ideal for that particular system. The generic interfaces were then deprecated.
Cap'n Proto also supports code generation plugins. But, as I said, we will soon also have an "official" RPC layer as well -- and it will hopefully be somewhat plugable itself, so that you can use a different underlying transport with the same generated interface code. Anyway, this will all become clearer with the next release, so stay tuned!
I'm not going to lie; it took me a little while to wrap my head around those stubs before implementing a TCP transport and semaphore triggers to unblock outstanding RPC function calls. However, it seemed much easier to do that than write a plugin for protoc to generate code that did roughly the same thing.
I'm currently considering RPC implementations for a personal project I'm working on. Right now I may end up trying Thrift since it seems to support RPC out of the box, but my ultimate goal is to have a WebSockets transport which Thrift doesn't provide. I may end up contributing to Cap'n Proto if it looks like the effort required to get RPC up and running has at least some parity with the effort required to extend Thrift for my needs.
It's clear from your planned use of futures and shared memory that your goal for Cap'n Proto is to make it the go-to library for communication in parallel computing. I'm definitely eager to see Cap'n Proto succeed in that endeavor. JSON is great for readability but it really isn't going to cut the cake when efficiency matters!
I hope you haven't forgotten to see if ICE might work for you, before you go looking at these "new" things (Not assuming you haven't, but it seems ICE fell a little out of hype from the day protobuf launched. I'm not convinced that fall from hype was entirely justified):
I look forward to hearing from you, should you decide to contribute. :) A web socket transport for Cap'n Proto would make a lot of sense, particularly if paired with a Javascript implementation, which one or two people have claimed they might create. I expect it will be easy to hook this in as a transport without disturbing much of the RPC implementation.
One random idea that just hit me if you're thinking about RPC layers anyways. Make sure that Cap'n Proto plays well with 0MQ. They probably do already, but a published example or two demonstrating it would not be a bad thing.
You can certainly send Cap'n Proto messages over 0MQ (or nanomsg) pretty easily -- Cap'n Proto gives you bytes, 0MQ takes bytes. Done deal.
However, supporting Cap'n Proto's planned RPC system on top of 0MQ may not work so well. The thing is, 0MQ implements specific interaction patterns, such as request/response, publish/subscribe, etc. Meanwhile, Cap'n Proto RPC is based on a different, more fundamental object-oriented model that doesn't fit into any of these patterns. A Cap'n Proto connection does not have a defined requester or responder -- both sides may hold any number of references to objects living on the other side, to which they can make requests at any time. So it fundamentally doesn't fit into the req/rep model, much less things like pub/sub. On the other hand, you can potentially build a pub/sub system on top of Cap'n Proto's model (as well as, trivially, a req/rep system).
At least, this is my understanding based on what I've managed to read so far of 0MQ's docs. I intend to investigate further, because it would be great to reuse existing work where it makes sense, but at the moment it isn't looking like a good fit. If I've missed something, definitely do let me know.
The killer feature that I like for 0MQ is that you can support message passing asynchronously, even when the other side is not currently up. For instance in a request/response pattern, one side might go away, get restarted, reinitialize, and then they carry on as if there wasn't a period in the middle where there was no connection. This kind of robust handling of network interruptions is very convenient for many use cases.
However what you describe isn't necessarily going to fit into that. The #1 thing that your description makes me wonder about is whether RPCs are going to be synchronous or asynchronous. So, for instance, if you hand me a data structure with a list objects that are references to data that I want to have, and I decide that I need 10 of them, do I have to pay for the overhead of 10 round trips, or can I say, "I need these 10" and get them all at once?
> support message passing asynchronously, even when the other side is not currently up.
That's probably something that could be implemented in Cap'n Proto as some sort of a persistent transport layer. But since the connections are stateful, it does require that when one end goes down, it comes back up with its state still intact. I have a lot of ideas for how to make this possible it big systems but it's a long way off.
Of course, in the simple case where you do have a defined client and server and the server is only exporting one stateless global service object that the client is using -- which is roughly what 0mq req/rep sockets are for -- then it should be no problem to support this.
> whether RPCs are going to be synchronous or asynchronous
The interface will be asynchronous based on E-style promises (similar to futures). In fact, say you call an RPC which returns a remote object reference, and you immediately want to call another method on that reference. With Cap'n Proto's approach, you will be able to do this whole interaction in one round trip instead of two. This is called "Promise Pipelining". There's a bit (just a bit) more detail here:
Looks like I might end up working Cap'n Proto after all. I've no experience with 0MQ, but it seems if Cap'n Proto's RPC is designed flexibly enough it would at least be able to support exclusive pairs as described on 0MQ's Wikipedia page [1]. I'll try to keep it in mind in whatever proposals I have w.r.t. the RPC design. Thanks for the suggestion, Ben!
The claims on this site are pretty impressive, but I have close to zero knowledge of the history here, so can someone comment on how many grains of salt this should be taken with? Otherwise, this looks pretty cool. Something that beats protobufs in overall speed could be really helpful depending on the application.
So, as the author of Cap'n Proto I'm biased -- though I'm also the author of Protobufs v2, so I'm not completely biased. :)
"Infinitely faster" is of course meant more to illustrate how Cap'n Proto works than to be taken as a literal speed measure. Although, if you actually wanted to compare Cap'n Proto to Protobufs, it's unclear what other number you can really come up with. The normal way to compare Protobuf speed vs. anything else is to measure the encode or decode step, but Cap'n Proto has no such step. You can measure an end-to-end system using one vs. the other, but then on the Cap'n Proto side you are basically measuring the speed of everything _except_ the Cap'n Proto code.
The git repo includes some contrived benchmarks along those lines which you can try out. I don't post the numbers because I'm not sure they are meaningful (even though they appear very favorable for Cap'n Proto). I'm really hoping to see a few unbiased third parties benchmark Cap'n Proto vs. Protobufs in real-world systems at some point.
Of course, the larger point here is that Cap'n Proto allows you to do things that Protobuf simply doesn't support, like mmap()ing in a large file and reading one field out of it in constant time, whereas with Protobuf you have to parse the whole thing making it O(size of file) time.
Thanks a lot for the reply, one of my favorite things about HN is getting questions answered by the authors of the tool in question. After reading more, your decision to not post those benchmarks is a smart one. I get where you're coming from with regards to it being hard to make a performance comparison to protobufs, it makes sense now. If/when I need to reach for some serialization I'll certainly try out Cap'n Proto.
There's a big focus on how it's better than protobufs because it's written by the original creator of protobufs and is his attempt at making a better version based on what he's learned. By the time it hits 1.0 I suspect the only reason to use protobufs rather than it will be for backwards compatibility with existing systems.
Just to give a bit of counterpoint, here are some trade-offs that Capn Proto makes compared with protobufs. (Full disclosure: I work at Google and know Kenton from his time here; I have my own protobuf library that I've worked on for several years). I'm sure Kenton will correct me if I get anything wrong. :)
Capn Proto's key design characteristic is to use the same encoding on-the-wire as in-memory. Protobufs have a wire format that looks something like:
[field number 3][value for field 3]
[field number 7][value for field 7]
etc.
The fieldnum/value pairs can come in any order, and may define as many or as few of the declared fields as are present. This serialization format doesn't work for in-memory usage because for general programming you need O(1) access to each value, so protobufs have a "parse" step that unpacks this into a C++ class where each field has its own member.
Protobufs are heavily optimized so this parsing is fast, but it's still a very noticeable cost in high-volume systems. So Capn Proto defines its wire format such that it also has O(1) access to arbitrary fields. This makes it suitable as an in-memory format also.
While this avoids a parsing step, it also means that your wire format has to preserve the empty spaces for fields that aren't present. So to get the "infinitely faster" advantage, you have to accept this cost. For dense messages, this can actually be smaller than the comparable protobuf because you don't have to encode the field numbers. But for very sparse messages, this can be arbitrarily larger.
As Kenton points out on http://kentonv.github.io/capnproto/encoding.html , lots of zeros compress really well, so even sparse messages can become really small by compressing them. To do this you lose "infinitely faster", but according to Kenton this is still faster than protobufs.
In both cases though, the tight coupling between the (uncompressed) wire format and the in-memory format imposes certain things on your application with regards to memory management and the mutation patterns the struct will allow. For example, it appears that the in-memory format was not sufficiently flexible for Python to wrap it directly, so the Python extension does in fact have a parse step.
Other cases where you could need a parse/serialize step anyway: if you want to put the wire data into a specialized container like a map or set (or your own custom data classes), or if the supported built-in mutation patterns are not flexible enough for you (for example, the Capn Proto "List" type appears to have limitations on how and when a list can grow in size).
It's very cool work, but I don't believe it obsoletes Protocol Buffers. I'm actually interested in making the two interoperate, along with JSON -- these key/value technologies are so similar in concept and usage that I think it's unfortunate they don't interoperate better.
Generally a fair analysis. A few comments/corrections:
> For example, it appears that the in-memory format was not sufficiently flexible for Python to wrap it directly, so the Python extension does in fact have a parse step.
This is not correct. The Python wrapper directly wraps the C++ interface. You might be confused by Jason's claim that "The INFINITY TIMES faster part isn't so true for python", but this was apparently meant as a joke.
It is true, though, that the constraints of arena-style allocation (which Cap'n Proto necessarily must use to be truly zero-copy) mean that working with Cap'n Proto types is not quite as convenient as protobufs, although most users won't notice much of a difference. Lists not being dynamically resizable is the biggest sore point, though most use cases are better off not relying on dynamic resizing (it's slow), and the use cases that really do need it can get around the problem using orphans (build an std::vector<Orphan<T>>, then compile that into a List<T> when you're done).
OTOH, over the years, many people have requested the ability to use arena allocation with Protobufs due to the speed benefits, especially with Protobufs being rather heap-hungry. I always had to tell them "It would require such a massive redesign that it's not feasible."
And yes, there is the trade-off of padding on the wire. You have to decide whether your use case is more limited by bandwidth or CPU. With Cap'n Proto you get to choose between packing (removing the zeros, at the cost of a non-free encode/decode step) and not packing (infinitely-fast encode/decode, larger messages). For intra-datacenter traffic you'd probably send raw, whereas for cross-internet you'd pack. Protobufs essentially always packs without giving you a choice. And because it generates unique packing code for every type you define (rather than use a single, tight implementation that operates on arbitrary input bytes), Protobuf "packing" tends to be slower.
> OTOH, over the years, many people have requested the ability to use arena allocation with Protobufs due to the speed benefits, especially with Protobufs being rather heap-hungry. I always had to tell them "It would require such a massive redesign that it's not feasible."
Yes totally, I agree that arena allocation is great. I think we both agree on this point, though we've taken two different paths in attempting to solve it.
Your approach is to say that arena allocation can be made pretty convenient, and sparse messages can compress really well, so let's design a message format that is amenable to arena allocation and then implement a system that uses this format both on-the-wire and in memory.
My approach is to say that we can solve this (and many other related problems) by decoupling wire formats from in-memory formats, and having the two interoperate through parsers that implement a common visitor-like interface. Then a single parser (which has been optimized to hell) can populate any kind of in-memory format, or stream its output to some other wire format. Of course this will never beat a no-parser design in speed, but the world will never have all its data in one single format.
I think of these two approaches as totally complimentary; to me Capn Proto is simply another key/value serialization format with a particular set of nice properties, and I want it to be easy to convert between that and other formats.
Since your approach is much more focused, you have been able to turn out usable results orders of magnitude faster than I have. I'm spending time implementing all of the various protobuf features and edge cases that have accumulated over the years, while simultaneously refining my visitor interface to be able to accommodate them while remaining performance-competitive with the existing protobuf implementation (and not getting too complex). As much as I believe in what I'm doing, I do envy how you have freed yourself from backward compatibility concerns and turned out useful work so quickly.
It's more like I started from "Let's design a message format that can be passed through shared memory or mmap()ed with literally zero copies", and then arena allocation was a natural requirement. :)
> Since your approach is much more focused, you have been able to turn out usable results orders of magnitude faster than I have.
To be fair, the fact that I'm working on it full-time -- and with no review, approval, or other management constraints of any kind -- helps a lot. :) (Down side is, no income...)
Yep, that is a big trade-off. There's room for someone to write a pure-Python implementation as well, to fill that niche. You could probably get a lot of the way using Python's `struct` module. But it would be slower.
First: thanks for your work on Protocol Buffers. I've used it fairly extensively for RPC communications between C++/Java clients and a Java service. It made things so much easier to get native objects in each language using a well-defined protocol.
One thing that bugged me about Protobuf is that it provided a skeletal mechanism for RPC (e.g. RpcController/RpcChannel) but later deprecated the use of that mechanism in favor of code-generating plugins. Since Cap'n Proto is billed as an "RPC system", do you have plans to include a more fleshed-out version of RPC than was provided in Protocol buffers? Having abstract classes for event-handling and transport mechanisms is a good idea for extensibility but it sure would make it easier for your users if there was at least one default implementation of each.
I imagine that Google has standard implementations of these things internally but balked at trying to support them for multiple languages as an open source project.