Slashdot Log In
Google Open Sources Its Data Interchange Format
Posted by
kdawson
on Tue Jul 08, 2008 03:07 PM
from the it's-fast-that's-why dept.
from the it's-fast-that's-why dept.
A number of readers have noted Google's open sourcing of their internal data interchange format, called Protocol Buffers (here's the code and the doc). Google elevator statement for Protocol Buffers is "a language-neutral, platform-neutral, extensible way of serializing structured data for use in communications protocols, data storage, and more." It's the way data is formatted to move around inside of Google. Betanews spotlights some of Protocol Buffers' contrasts with XML and IDL, with which it is most comparable. Google's blogger claims, "And, yes, it is very fast — at least an order of magnitude faster than XML."
Related Stories
Submission: Google open sources protocol buffers by Anonymous Coward
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
An order of magnitude over XML? (Score:5, Funny)
So is, well, just about anything.
Re:An order of magnitude over XML? (Score:5, Interesting)
Looks like Google just invented the IIOP [wikipedia.org] wire protocol, which is also platform agnostic and an open standard.
I guess the main difference here is that their "compiler" can generate the actual language-domain classes off of the descriptor files, which is a definite advantage over "classic" IDL.
"Google protocol Buffers" is cooler than the OMG terminology, but this kind of thing has been around for 20 years.
Parent
Re:An order of magnitude over XML? (Score:4, Funny)
Oh, I'm a little ashamed that I recognize this message as CORBA flamebait.
Parent
Re:An order of magnitude over XML? (Score:5, Informative)
There have been all kinds of attempts to produce this sort of stuff. RPC, DCE, Corba, DCOM, etc, are programmatic interfaces and handle function calls, synchronization, etc. OPeNDAP is probably the closest to Google's architecture in that it is ONLY data. It's more sophisticated, as it handles much more complex data types than mere structures, but it has its own overheads issues. It isn't designed to scale to terabyte databases, although it DOES scale extremely well and is definitely the preferred method of delivering high-volume structured scientific data - at least when compared to the RPC family of methods, or indeed the XML family. I wouldn't use it for the kind of volume of data Google handles, though, you'd kill the servers.
Parent
Re:An order of magnitude over XML? (Score:5, Insightful)
Technically, you are correct - platform-agnostic data transfer has been possible since Sun's earliest RPC implementations. However, this seems to be considerably lighter-weight (although so is Mount Everest) and because order is specified, it's going to be much simpler to pluck specific data out of a data stream. You don't need to have an order-agnostic structure and then an ordering layer in each language-specific library.
Actually, XDR (used for Sun's RPC) is very lightweight, arguably lighter than PB. (Yes, I forsee a Java implementation called PB&J.) XDR is potentially more compact, since it doesn't encode field identifiers, but it's also big-endian, which made it less attactive as little-endian computer archtectures took over the world. Also, while XDR demands a fixed ordering of fields, field order in PB *isn't* specified; the field identifiers allow you to order the fields anyway that you like.
Overall, I like it. It's obvious that the developers were familar with the flaws of older protocols, and found ways to fix most of them. The only obvious thing I see missing is a canonical way to encode the .proto file as a Protocol Buffer, to make a stream self-describing.
Parent
Re:An order of magnitude over XML? (Score:4, Informative)
The only obvious thing I see missing is a canonical way to encode the .proto file as a Protocol Buffer, to make a stream self-describing.
A-ha! I found it! [google.com] "Thus, the classes in this file allow protocol type definitions to be communicated efficiently between processes."
Why do you need this? Well, you may not. "Most users will not care about descriptors, because they will write code specific to certain protocol types and will simply use the classes generated by the protocol compiler directly. Advanced users who want to operate on arbitrary types (not known at compile time) may want to read descriptors in order to learn about the contents of a message."
Parent
Re:An order of magnitude over XML? (Score:5, Funny)
An order of magnitude over XML? So is, well, just about anything.
Well, let's also not forget that the meaning of the expression "an order of magnitude" depends strongly from the numeric base you're using.
Parent
Re:An order of magnitude over XML? (Score:5, Funny)
Nobody makes jokes in Base 13 [wikipedia.org]!
Parent
Re:Between a rock and hard place (Score:4, Insightful)
Funny, I'm tired of seeing YAML in places where XML would work fine.
Like serializing my Ruby objects, for example. When I don't care about performance, XML is best, because almost everything else will read and write it, including my text editor, and I know the syntax. When I *do* care about performance, I'm not going to use YAML either.
I don't see the niche YAML fits, frankly.
Parent
Likely story! (Score:5, Funny)
That is just because they aren't using enough XML!
More XML? EXI, Efficient Xml Interchange! (Score:4, Informative)
Parent
Re:Likely story! (Score:4, Informative)
Are you serious? XML is great for certain applications, but the one thing it *isn't* is fast. It's very believable that something like this could be an order of magnitude faster.
Parent
Re:Likely story! (Score:5, Informative)
Parent
Re:Likely story! (Score:5, Funny)
Parent
Re:Likely story! (Score:5, Insightful)
Being 10x faster than XML to work with is entirely believable: If you're serializing directly to binary structures, those structures can be directly manipulated without any parsing at all... and if you need to do some byte-swapping and alignment adjustments to get them into and out of native form for your current processor, those are still operations which can be performed in a matter of a few CPU instructions, rather than through a few hundred KB of libraries.
I drink the XML kool-aid plenty -- but there are things it's good for, and things it's not. Serializing and parsing truly massive amounts of data is part of the latter set.
Parent
Re:This is a good thing (Score:5, Insightful)
We wanted to give an idea of the speed without trying to boast too much or look like we were directly challenging anyone. Of course every news outlet has chosen to highlight the speed comment -- including the numbers which were intended to be ballpark figures -- more than was intended, but I guess that isn't surprising.
I agree that the tiny "person" example is not a good benchmark case. It was intended as a usage example, not a speed example, but I stuck the speed numbers in there just meaning to give people a vague idea of the difference. The "20-100 times faster" comment is based on testing a variety of formats -- both unrealistic ones and real-life formats used in our search pipeline -- against programmatically generated XML equivalents (which may or may not themselves be realistic, though they contain the same data with the same structure). libxml2 was used for parsing XML. I don't really know how libxml2's speed compares to other XML parsers, but I didn't have a lot of time to investigate. The 20x faster number comes from the largest data set (~100k-ish) while the 100x number comes from a very small message. The most realistic case was about 50x. Sorry that I cannot provide exact details of the benchmark setup since many of the test cases were proprietary internal formats.
In any case, I'm hoping that some independent source conducts some tests because I think anything we produced would probably have unintentional biases in it. Of course, I'll update the numbers in the docs if they turn out to be wildly off-base.
Parent
I bet ... (Score:5, Funny)
No PERL API ??!!?? (Score:4, Insightful)
C++
Python
Java
what about PERL ? :]
Re:No PERL API ??!!?? (Score:4, Insightful)
Go out and write one, sonny!
That's the beauty of open source.
Parent
Re:No PERL API ??!!?? (Score:5, Informative)
According to Brad Fitzpatrick's(of LiveJounral fame) blog [livejournal.com], He's working on Perl support.
Parent
Re:No PERL API ??!!?? (Score:5, Insightful)
Perl is to programming languages what English is to natural languages: easy to fool around with, hard to learn well, but when you do, the expressive power is incredible. And when you mess it up, nobody understands what you're trying to say.
Parent
As a former user of CORBA (Score:5, Interesting)
It looks like Google has taken some of the good elements of CORBA and IIOP into its own interchange format.
While CORBA certainly is bloated in a lot of ways, the IIOP wire protocol it uses is vastly faster and more efficient than any XML out there.. and yes it is just as "open" (publicly documented and Freely available for use in any open source application) as any XML schema out there. J2EE uses IIOP as well and its is technically possible to interoperate (although the problem with CORBA is that different implementations never really interoperated as they were supposed to).
As a side note, I'd rather write IDL code than an XML schema any day of the week too, but that's another rant.
compare to thrift ( from facebook) (Score:5, Informative)
both really from the same design sheet, but thrift has been opensource'd for over a year, and has many more language bindings. its been in use in several opensource projects (thrudb comes to mind), and has much more extant articles/documentation.
http://developers.facebook.com/thrift/
Fast (Score:5, Interesting)
"And, yes, it is very fast â" at least an order of magnitude faster than XML."
Just wait for the XML zealots to come crashing and not believing that XML is not the fastest, best, solution to all the world's problems (including cancer) and of course people at Google are amateurs and id10ts and WHY DO YOU HATE XML kind of stuff.
Or, as Joel Spolski once said: http://www.joelonsoftware.com/articles/fog0000000296.html [joelonsoftware.com]
No, there is nothing wrong with XML per se, except for the fans...
Ok, I'll bite... (Score:5, Interesting)
Obviously, those at Google felt XML didn't work well for them. They have the resources to invent a protocol and libraries to support it. And, they are big enough to be their own ecosystem, which means as long as everyone at Google is using their formats, interop is no biggie. Good for them, I don't begrudge that decision.
I'm actually a game developer, not a web developer, so I'll speak to XML's use as a file format in general. Here's a few points regarding our use of XML:
* We only use it as a source format for our tools. XML is far too inefficient and verbose to use in the final game - all our XML data is packed into our own proprietary binary data format. .NET platform (Windows is our development platform, of course). It's astoundingly easy to serialize data structures to XML using .NET libraries - just a few lines of code.
* We also only use it as a meta-data format, not a primary container type. For instance, we store gameplay scripts, audio script, and cinematic meta-data in XML format. We're not foolish enough to store images, sounds, or maps in a highly-verbose, text-based format. XML's value to us is in how well it can glue large pieces of our game together.
* All our latest tools are written in C# and using the
* Because it's a text-based format and human readable, if a file breaks in any way, we can just do a diff in source control to see what changed, and why it's breaking.
I'll make a concession that I've heard of some pretty awful uses of XML. But those who dismiss XML as a valuable tool in the toolchest are equally as foolish as those who believe it's the end-all and be-all of programming (I'm not saying that's true of you, just pointing out foolishness on both sides). Like any tool, it's most valuable when used in it's optimal role, not when shoehorned into projects as a solution to everything.
Parent
Smart move (Score:5, Insightful)
And as a bonus, they help undermine opponents who use competing technologies by helping train the workforce away from their practices. Overall I think it's very intelligent and well done strategic move.
The killer feature is simplicity (Score:5, Insightful)
I'd use it because 95% of the time all I need is something simple like this, and the other 5% of the time I should go back and rethink my design anyway.
That said, there is still a case for XML, especially the self documenting and human readable nature of the document, but there are a lot of cases where it is used today where it only adds unnecessary complexity and actually makes your code more difficult to maintain instead of simpler.
XML is a crappy format (Score:5, Insightful)
I always told people that -- it's optimized for:
1. Easy parsing by parsers written by people who slept through their compiler classes.
2. Verification in situations when it's impossible to devise a meaningful reaction to a failure (other than either "everything failed, turn off the computers and go home" and "assume the data to be valid anyway because ALL of it will have the same formatting error because the same program generates it")
3. Dealing with data that arrives in neatly packaged "documents" and "requests", as opposed to being constantly produced and consumed.
4. Either communicating between programs that have the same knowledge of message semantics, or preparation of pretty human-readable documents.
None of the above even remotely applies to anything practical except UI/display formats -- this is why XHTML and ODF (and because of that at some extent XSL) are usable, SOAP is a load of crap, and for the rest of purposes XML is used as a glorified CSL with angle brackets. XML is widespread because monumentally stupid standard is still better than no standard.
So here is your example of how superior can be ANY format that is not based on this stupid idea.
Re:XML is a crappy format (Score:4, Insightful)
Y'know, I usually give low-UID Slashdotters a modicum of respect, but this diatribe is off-the-charts nonsense.
And your evidence of this assertion is...what exactly? Not to mention the minor detail that XML and compilers are orthogonal: you can use XML (or many other data interchange formats) with non-compiled languages, and most compilers know nothing about XML (or many other data interchange formats).
And your evidence of this assertion is...what exactly? XML-consuming programs that are aware of the data structure can have as detailed a "reaction to a failure" as a JSON-consuming program, or a YAML-consuming program, or a Protocol Buffer-consuming program. XML-consuming programs that are not aware of the data structure can, if the XML supplies it, validate against a DTD or schema, things which are not possible in some other data interchange formats (e.g., JSON, YAML).
All data comes in neatly packaged buckets of varying types. We call them "bytes" and "packets" and "structures" and "records" and "frames" and "rows" and the like. The only way I can interpret your claim in a way that makes sense is to translate it as "XML sucks for streaming audio and video", which is undoubtedly true, and I don't think anyone uses it in that arena.
On the contrary, this is one of XML's primary strengths — handling cases where programs lack the "same knowledge of message semantics".
With most data interchange formats, from CSV to JSON to Protocol Buffers, either you know everything about the data structure you're receiving, or you're screwed. In other words, there is no discoverability and no standardized means of being able to only deal with a portion of the data. This is particularly true for binary formats, like Protocol Buffers — either you know exactly what structure you received so you can parse it, or you're SOL, since it's just a bunch of bytes.
With XML namespaces, it is entirely possible for Program X to publish data that Program Y has no intrinsic knowledge of in its entirety, but might know in part. If Program Y knows how to handle documents containing Dublin Core elements, for example, it can work with just those elements and ignore the rest of the document.
You're welcome to have any opinion of XML you like. Heck, I even agree that XML tends to be used in places where it's overkill or too verbose. But if you want to convince others that your opinion is the correct one, you'll need to do a better job than this.
Parent
Re:XML is a crappy format (Score:4, Interesting)
In the world I live in, data is frequently valid, but not always:
Many of those can be caught by the general-purpose validators that you decry, and that limits the number of validation routines programmers have to deal with. And your complaints re: CPU, memory, and cache place a value on them that may or may not be proper in every context. Or, as my former business partner put it, "in six months' time, computers will be faster and cheaper, but programmers will be neither".
You obviously have a very different definition of "streaming" than I do, as I'd argue virtually nothing uses streaming, from the days of FORTRAN and COBOL to the present day.
Precisely. Decomposable formats, like XML, allow programs to have semantics for part, but not all, of a data structure. Non-decomposable formats, like C structs, require semantics for all of a data structure. In situations where you know 100% of all use cases for a data structure, non-decomposable formats are fine. If, however, you want to allow for what Jonathan Zittrain refers to as "generativity" (i.e., unanticipated uses for existing technology as a means of advancing said technology), decomposable formats can be a benefit.
Take, for example, ODT vs. classic binary Word documents, which are pretty much just a serialization of a big-ass binary structure as I understand it. I've written programs that parse and generate ODT, or, more precisely, the portions of ODT that I need. Frankly, I don't care what the rest of it is, so long as my generated documents work properly. And I didn't need to refer to the ODT documentation on OASIS or anything to write them, as the XML was sufficiently human-readable that, accompanied with experimentation, I was able to determine how to generate valid ODT. With Word, even if there were OOXML-sized documentation for it, I'd have to hand-roll my own parser for the whole damn format, just to pick out the pieces I need to work with. Now, if I worked for Microsoft on the Word team, I wouldn't have that problem, because I'd already have the parser. However, I, like most people, don't work for Microsoft, and even if Microsoft's parsers were available, they might not fit my environment (e.g., won't run on Linux).
Don't get me wrong, XML definitely gets overused. That's a problem with the uses of XML, not XML itself.
Parent
JSON (Score:5, Interesting)
Looks kinda like JSON to me.
Re:JSON (Score:5, Informative)
Structurally Protocol Buffers are similar to JSON, yes. In fact, you could use the classes generated by the Protocol Buffer compiler together with some code that encodes and decodes them in JSON. This is something some Google projects do internally since it's useful for communicating with AJAX apps. Writing a custom encoding that operates on arbitrary protocol buffer classes is actually pretty easy since all protocol message objects have a reflection interface (even in C++).
The advantage of using the protocol buffer format instead of JSON is that it's smaller and faster, but you sacrifice human-readability.
Parent
Re:JSON (Score:4, Informative)
The major difference between this and something like JSON or YAML or even XML is that those formats all include the format information (variable names, nesting, etc) along with the data. This does not.
message Person {
required int32 id = 1;
required string name = 2;
optional string email = 3;
}
What you are looking at above is the Protocol Format (.proto file) for a single message, which is analogous to an XML schema. No data is stored in that file - the numbers you see are unique ids for the different fields, and they are used in the low low-level representation of the data (not all fields have to be included in every instance of a message)
The actual data is serialized using a compact binary format, not ASCII like JSON/YAML/XML which makes it much more efficient both to transfer over a network as well as to parse.
Parent
Re:JSON (Score:5, Interesting)
Modify JSON so unquoted attributes are 'type labels' and define the type of an attribute by giving a label or a default value. For instance:
... now you have pretty much exactly the same message definition as protocol buffers, but in pure JSON. It could also use some convention like "@WORK" for labels/classes so that a normal JSON parser can parse the message definitions. You can write a code generator to make access classes for messages just by walking the json and looking at the types. I don't see that 'required' and 'optional' keywords help much... imo defaults are generally better (even if they are nil). But this could easily be expressed in a json message definition.
It's easy to make a binary JSON format that is fast and also small, so there is little advantage to protocol buffers there. It's also easy and ridiculously fast to compress JSON text using say character-based lzo (Oberhumer).
Maybe somebody can explain, but it doesn't seem like protocol buffers really have much advantages over JSON. It sounds like it is effectively just a binary format for JSON-like data (name-value pairs they say) along with a code generator to access it. The code generator is nice, but this is like a day's work max. Maybe I'm not understanding google's problems, but I'll stick with JSON since it actually is a cross-platform, language neutral data format... and you can always optimize it if actually needed.
Parent
Have they ever heard of BER/DER? (Score:3, Insightful)
How is this either implementationally or conceptually different from BER/DER encoding (commonly used and available all over the place)?
Looks to me like it is exactly the same thing, reimplemented. I am sure bearing a mark of Google is nice and all, but they are definitely reinventing the wheel here.
ASN.1 encoded with BER/DER just needs tools (Score:4, Informative)
ASN.1, from 1985, really is very similar. Here's a message defined in ASN.1 form:
Note that this has almost exactly the same feature set as Google's representation. There are named, typed field which can be optional or repeated. It just looks more like Pascal, while Google's syntax looks more like C.
Parent
Re:Have they ever heard of BER/DER? (Score:4, Funny)
Parent
I have an XML alternative format too. (Score:4, Funny)
I think this "delimited" format has a lot of potential.
Old-School Property lists? (Score:4, Insightful)
The similarity between these things and NeXT's Property Lists (now called "Old-School Property Lists" that Apple/NeXT has standardized on XML) is incredible. Some things are changed, like having a specification instead of just assuming that the recipient will parse it and figure it out, but the likeness is there. I wonder if any of the proto people at google had experience with plists, or if it's just a case of convergent design.
Everything old-school is new-school again, I guess.
Re:Back to the 70's night? (Score:4, Insightful)
The article is actually completely wrong there. The protocol buffer binary format uses tag/value pairs, not fixed positions. Parsers simply ignore any tag they don't recognize and move on to the next.
Parent
Re:WTF am I missing (Score:5, Informative)
Parent
Re:WTF am I missing (Score:5, Insightful)
Seems like you are missing the code they released that allows you to implement this in a number of languages from the 'get-go'.
You've also missed that they've just told the world how the majority of their systems talk, something most people would find interesting given how much Google does and the fact that one of Google's strong points is mangling huge amounts of data in a relatively quickly manner.
PS. Your format stinks and is horribly slow and unscalable when it comes to adding to the library. Genre's are so unbelievably grey defined that you might as well just sort them by the dominate color of the cover. Google would have done better.
Parent
Re:Why another encoding scheme? (Score:4, Insightful)
This is just yet another way in which Google demonstrates that it is suffering from NIH syndrome [wikipedia.org]. Instead of improving existing tools, they have to go off and re-invent all the bad mistakes of past, including non-relational databases [wikipedia.org], clunky [google.com] binary encodings, and a bizarre non-POSIX filesystem [wikipedia.org].
Just imagine how far we ahead we would be today if Google had put the same effort into creating tools the rest of the SQL-writing, open(2)-using world could use.
Parent
Re:Why another encoding scheme? (Score:5, Informative)
You think? Take BigTable. Wikipedia describes it as: '"a sparse, distributed multi-dimensional sorted map", sharing characteristics of both row-oriented and column-oriented databases'. Sounds, to me, like a specialized solution to a very specialized problem, a problem that, I presume, didn't fit with any existing solution. Same goes with GFS. After all, do you really think they didn't evaluate existing solutions before embarking on building an entirely new distributed filesystem? Do you really think they're that stupid?
As for Protocol Buffers, given the existing solutions out there (such as ASN.1 and CORBA) are generally ugly and/or over-engineered, it sounds to me like they're simply addressing a gap in the industry... after all, XML and SOAP aren't the end-all and be-all of generic object-passing protocols.
Parent
Re:Why another encoding scheme? (Score:4, Insightful)
I dont think its NIH syndrome. They no doubt tested other solutions before doing their own thing.
Dont forget this code is in widespread use and works very well. Googles server farm aint exactly small and the load they see is probably second to none.
A couple of percents of better efficiency for Google probably means millions in saved costs. Tossing a couple of months on development on something like this is money well spent.
I guess if all you have is SQL everything is a SQL SELECT no matter what you want to achieve.
Parent
need something different (Score:5, Insightful)
If Google had tried to build their system on relational databases, XDR, and NFS, they would have spent huge amounts of money and spent lots of time trying to shoehorn their software into those constraints. And it's not just Google that did this: Amazon did the same thing, with their SimpleDB, S3, and SQS.
The actual mistakes were relational databases, XML, and distributed POSIX file systems; all of those were systems designed by people with too much time on their hand and no real-world, large scale problems to solve. Finally, those mistakes are getting corrected, at least when it comes to high-end computing. At the low end, I suppose people will continue to tinker around with those toys.
Parent
Re:Why another encoding scheme? (Score:5, Insightful)
You think it's a "mistake of the past" that Google wrote things like GFS and BigTable that run on commodity hardware, scale basically horizontally (eg. you can just throw machines at the problem) and survive machine failures without human intervention?
You don't "improve" on an existing tool like a relational database by adding a "feature" like fault tolerance. You have to redesign from the base up with those assumptions.
Parent
Re:Why another encoding scheme? (Score:5, Interesting)
Just imagine how far we ahead we would be today if Google had put the same effort into creating tools the rest of the SQL-writing, open(2)-using world could use.
We wouldn't be ahead at all. We use different tools than they do because they are dealing with different volumes of traffic, data and demands. Let's take a moment and look at your specific complaints. You say Google suffers from NIH syndrome. Having previously worked at Google, I think you are half right. The difference is that Google both benefits _and_ suffers from NIH syndrome. Sometimes the company spends too much time reinventing the wheel, but sometimes the tools out there aren't (and shouldn't be) useful to Google. Apache shouldn't be changed to support the kind of traffic that Google handles because then it wouldn't nearly as good for all of the rest of the world. General software is great because it solves so many problems. However, general software isn't the right solution for all problems, especially extreme ones. Just about all of Google's needs are extreme ones due to the volume of traffic. You dislike the idea of BigTable. Why not use the right tool for the right job? BigTable is a ridiculously fast database system that works beautifully with petabyte sized databases. SQL isn't the right answer to all solutions. They DO use SQL... but when it is the appropriate solution. They have some really sexy internal tools for dealing with SQL and such and I'm hoping those are coming down the open source pipeline soon. :)
You claim the Protocol Buffers are clunky. I've used them and developed with them extensively. They aren't clunky at all, they are actually quite elegant and easy to use. They streamline development, are incredibly reliable, and are incredibly fast.
You obviously are confused by GFS as well. The system is transparent to the application by using standard i/o stream classes. It is inherently redundant to ensure data security. It is so fast in its response time that Google search is the fastest of any major player. The list goes on and on.
I don't really see how you can be upset at Google for making awesome software and then giving us access to it.
Parent
Re: (Score:3, Informative)
It's not hard because XML has to be the most bloated (and yet still, ironically, nowhere near human-readable) format ever invented. That it has not only not been discarded, but is now being used to store binary blobs by guys like Microsoft and OO.org is testimony to the sheer overwhelming stupidity of a lot of developers.
Re:How about C? (Score:4, Funny)
Well, I can't translate C++ to C until after it is DECLASSIFIED...
*rimshot*
Cheers!
Parent