Thoughts to come back to …

[27 August 2008]

Observation: both RDF and Topic Maps seem to aspire to make it easy to find all the ways in which a given thing (topic or resource) may appear in a given body of data.

In both, the basic goal appears to be that if you look for Essex or Washington, you should be able to specify whether you mean the human being or the geographic entity (and probably be able to distinguish the state of Washington from the various cities named Washington), and find it no matter where it appears in the system. In RDF terms, this would mean no matter which triples it appears in, and no matter whether it appears as subject or object of the triples; in topic map terms, it would mean no matter which roles it plays in which associations.

Observation: Codd insists firmly that to be acceptable in his eyes, relational database management systems must not only provide built-in system-level support for domains (which may be regarded as a form of extended data types with slightly more semantics than the basic types in a typical programming language, so you can distinguish people from places even if you use VARCHAR(60) columns to represent both), but also include ways of finding all the values actively in use for a given domain, regardless of relation or column, and of finding all the occurrences of a particular value of a particular domain, without getting it mixed up with any values from different domains which happen to use the same underlying basic datatype in their representation. (For those with The relational model for database management, version 2 (Reading: Addison-Wesley, 1990) on the shelf, I’m looking at sections 3.4.1 The FAO_AV Command and 3.4.2 The FAO_LIST Command).

Question: Are the RDF and TM communities and Codd after essentially the same thing here? Is there some sense in which they fulfil (or are trying to fulfil) this part of Codd’s ideal for data management better than SQL systems do?

What exactly is the relation between this aspect of both RDF and TM on the one hand, and Codd’s notion of domains or extended data types on the other?

I’ve wanted to think about this for years, but have not managed to find anyone to discuss it with who had (a) sufficient knowledge of both semweb or Topic-Map technology and relational theory (or rather Codd’s particular doctrines) and (b) the time and inclination to go into it, at (c) a time when I myself had the time and inclination. But someday, perhaps …

Information and the laws of thermodynamics

[27 August 2008]

Liam Quin has been spending most of his blogging effort recently on his site, but recently he posted a thoughtful piece on his ‘other blog’, In search of XMLence, on the relations among standard off-the-shelf vocabularies, customized vocabularies, what widely deployed systems can understand, and what people want to be able to say. (“The Super Information Archipelago”, 26 August 2008.)

One net result: if you use a vocabulary customized to your application, you get a better fit, at some cost in effort (or money) and immediate interchangeability (or interoperability) with other vocabularies. If you use standard one-size-fits-all vocabularies, you get free interoperability with others, at the cost of a poorer fit for the work you want to do in the first place. Liam imagines the situation of users of custom vocabularies as a kind of archipelago of islands of rich information in a sea of, ah, less rich information, sometimes connected by good bridges.

Something in the tradeoff (perhaps it’s the image of the islands) reminds me of the observation someone made long ago when they were teaching me about the three laws of thermodynamics. This was the kind of conversation where there was a little less emphasis on the laws as formulated in physics books than on the popular paraphrase:

  1. You can’t win.
  2. You can’t break even.
  3. You can’t get out of the game.

If the laws of thermodynamics say that entropy always increases, someone asked, then how is it possible that in some situations entropy seems to decrease, and order to increase? Many of us spend a lot of our time trying to build systems to organize things or information; if the universe is stacked against us, how is that we ever have even the illusion of having succeeded? (And is this a good excuse for not even trying to organize my desk?)

Entropy does increase, was the answer, but only in the universe as a whole — not necessarily at every point in the universe uniformly. You can increase the order, and decrease the entropy, in a restricted portion of the universe — just not in the universe as a whole.

It sheds light from a different angle on the issues of data islands and walled gardens.

Il faut cultiver notre jardin.

Further notes on optimistic concurrency and XML parsing

[22 August 2008]

I just posted some notes on a paper given at Balisage 2008 by Yu Wu et al. of Intel.

A few thoughts occurred to me in writing up those notes which might merit separate consideration.

How effective could pessimization be?

A key part of the optimistic concurrency algorithm presented by Yu Wu et al. is that the process of chunking the document needs to be quick. So they make some guesses, when chunking, that could later be proven wrong; in that case, the chunk needs to be re-parsed.

I suppose the worse-case scenario here is that a sufficiently lucky and malignant adversary could construct a document in which the context at the end of chunk 1 means that chunk 2 needs to be reparsed, and the reparsing of chunk 2 reveals for the first time that chunk 3 now needs to be reparsed, and so on, so that in the end you end up using n time slices to parse n chunks, instead of n divided by the number of threads.

So there’s an interesting question: how long can we keep this up?

It’s pretty clear that if we know exactly where the pre-scanner will break the chunks, then we can devise an XML document that forces chunk 2 to be reparsed. Can we construct a document in which only the second, correct parse of chunk 2 reveals that chunk 3 now needs to be reparsed (i.e. in which the first parse of chunk 2 makes chunk 3 look OK, and the second one shows that it’s not OK)?

Can we make a document in which every time we reparse a chunk with the correct context, we discover that the next chunk also needs to be reparsed? How much reworking can an omniscient and malevolent XML author cause this algorithm to do? Remember that comments and CDATA sections do not nest; the worst I can figure out off hand is that a comment or CDATA section begins in chunk 1 and doesn’t end until the last chunk.

How many chunks do you want?

The paper says fewer chunks are better than many chunks (to reduce post-processing costs), and that you want at least as many chunks as there are threads (to ensure that all cores can be busy). To simplify the examples I’ve been thinking about, I’ve been imagining that if I have eight threads, I’ll make eight chunks.

But if I’ve read the performance data and charts right, the biggest single reason the Horatian parser is not getting an eight-fold speedup when using eight threads is the need to reparse some chunks, owing to bad guesses about parse context made during the first parse. If we have eight threads and eight chunks, everything is fine for the first pass over the chunks. But if we need to reparse two of the chunks, then it rather looks as if six threads might be sitting idle waiting for the re-parsing to finish.

I wonder: would you get better results if you had shorter chunks, and more of them, to keep more threads busy longer? What you want is enough chunks to ensure that while you are reparsing some chunks, you still have other chunks for the other threads to parse.

As a first approximation, imagine that we have eight threads. Instead of eight chunks, we make fourteen chunks, and give the first eight of them to the eight threads. Let’s say two of them need to be reparsed; the reparsing goes on at the same time that the remaining six threads parse the remaining six chunks. The minimal path through the speculative parsing step remains the time it takes to parse two chunks, but the chunks are somewhat smaller now. The only question is how much additional time the post-processing step will now take, given that it has fourteen and not eight chunks to knit together.

And of course you need to bear in mind that if one chunk in four turns out to need re-parsing, then three or four out of the fourteen chunks are going to need reparsing, not just two. By the time you factor that in, and try to ensure that your last round of parsing doesn’t generate any new re-parse requests, things have gotten more complicated than I can conveniently deal with here (or elsewhere).

Maybe that’s why the Intel paper was so non-committal on the way to choose how many chunks to make in the first place: it can get pretty complicated pretty fast.

Optimization and context independence in schema languages

One of the things that intrigues me about these results is that so much of what people have said needs to be done to schema languages to ensure that validation can be fast has nothing much to do with the speed gains shown by optimistic concurrency.

I thought for a while that this work did benefit from the fact that elements can be validated against XSD types without knowledge of their context (no reference to ancestors or siblings in any assertions, for example), but on reflection I’m not sure this is true: in order to find the right element declaration and type definition to bind an instance element, you need to know (a) the expanded name of the element (which means knowing the in-scope namespaces, which in practice means having looked at all of the ancestors of the element), and (b) the type assigned to the element’s parent (unless this element is itself the validation root). Once you have a type, it’s true that validation is independent of context. But the assignment of a type to an element or attribute does depend, in the normal case, on the context. It’s not clear to me that allowing upward-pointing XPath expressions in assertions or conditional type assignment would make much difference.

To really exploit parallelism in validation, it would seem you want to eliminate the variable binding of expanded names to element declarations and to types.

Back to DTDs plus datatypes, anyone?

Optimistic concurrency and XML parsing and validation (Balisage report 3, in which chronology is abandoned)

[22 August 2008]

My brief hope (it would be misleading to refer to it as a “plan”) to report daily from Balisage 2008 has bitten the dust — it did that shortly after noon on the first day, when my account of Sandro Hawke’s work on XTAN turned out to take more time than I had available — but there is still a lot to say. I’m going to abandon the chronological approach, however, and write about things that come to mind, in a more or less random order.

One of my favorite papers this year was the one submitted by Yu Wu, Qi Zhang, Zhiqiang Yu, and Jianhui Li, of Intel, under the slightly daunting title “A Hybrid Parallel Processing for XML Parsing and Schema Validation”. (I think they are all members of the XML Engineering Team at the Intel China Software Center in Shanghai, but I am not sure I’ve read all the affiliation info correctly; I keep being distracted by the implications of an Intel software center having an XML engineering team.)

When I paraphrased this paper to a friend recently, her response was “Wow! That’s a lot more accessible than I would have guessed from the title.” So perhaps it’s worth while to try to record here the high points of the work, in a way that’s accessible to people to people with no more than lay knowledge of the relevant technical disciplines. (This goal is made easier, of course, by the fact that I don’t have more than lay knowledge of those disciplines myself.)

For technical details, of course, readers should go to the paper in the online proceedings of the conference; all errors in this summary are mine.

The elevator speech

The quick executive summary: XML parsing, and validation, can be a lot faster if performed by a multi-threaded process using optimistic concurrency.

By “optimistic concurrency”, I mean a strategy that parallelizes aggressively, even if that means doing some work speculatively, based on guesses made when parallelizing the work, guesses that might prove wrong. When the guesses are wrong, the process has to spend extra time later fixing the resulting problems. But if the speedup gained by the concurrency is great enough, it can outweigh the cost of wrong guesses. (This is a bit like the way Ethernet chooses to clean up after packet collisions, instead of attempting to prevent them the way token-ring networks do.)

A fast and simple-minded process divides the XML document into chunks, and multiple parallel threads handle multiple chunks at the same time. The fragmentary results achieved by the chunkwise parsing can be knit back together quickly; sometimes the fixup process shows that one or the other chunk needs to be reparsed.

What does Moore’s Law have to do with XML parsing?

OK, so much for the elevator speech. If you’re still interested, here is a little more detail.

First, some background. Moore’s Law says, roughly, that the number of transistors it’s possible to put on a chip doubles every eighteen months. For many years, this doubling of transistor count was accompanied by increases in clock speed. (I don’t understand the connection, myself, not being an electrical engineer. I just take it on faith.) But higher clock speeds apparently require more power and generate more heat, and eventually this causes problems for the people who actually use the chips. (Very few laptop designers are persuaded that water-cooled systems can be made to have the requisite portability. And liquid nitrogen, which would be the next step? Don’t get them started.)

So nowadays the doubling of transistors appears to be reflected in the rise of multi-core chips; dual-core chips in the current crop of off-the-shelf machines, with every expectation that the number of cores on standard chips will rise. Intel has already shipped chips with four and eight cores, although I haven’t seen any four-core machines on my list when shopping for laptops. (I’m not sure whether cores are expected to double every eighteen months indefinitely, or not; if they do, will we end up with a 1024-core chip vaguely resembling the Connection Machine in our laptops in another fifteen years?)

It used to be that performance rose about as fast as the transistor count, because the clock speed kept going up; even if software didn’t do anything smarter, it kept getting faster because the chip was faster. But to get performance benefits out of multi-core chips, a system is going to want to keep all of the cores busy whenever possible. People have been thinking about parallel computing for a long time, and at the 30,000-foot level, the answer so far seems to boil down to the simple, general principle “Gosh, that’s hard!”

Under those circumstances it seems plausible that a manufacturer of multi-core chips might see it as in the manufacturer’s own interest to show people how to multi-thread common applications, so as to make it easier to get as much performance as possible out of multi-core chips.

Parallelizing parsing

How do you parallelize the task of parsing an XML input stream? (There may be other ways to keep multiple threads busy, but this seems like the obvious choice if we can manage it.)

The answer wasn’t obvious to me. There are references to parallelism in the titles or summaries of a number of papers in Grune and Jacobs’s online bibliography on parsing techniques (part of the supplemental material to their book on Parsing Techniques), but none that leap out at me as being easy to understand.

One way to parallelize XML parsing is to run a pre-scanner over the document to select suitable places to divide it into chunks, and then to parse the chunks in parallel. (There is earlier work on this by Wei Lu et al., which Yu Wu et al. cite. They also cite work on other ways to parallelize XML parsing, but I don’t understand them and won’t try to describe them.)

The problem with (the existing state of) the pre-scanning approach, according to the Intel team, is that the pre-scanning takes a lot of time by itself, and once the parsing process itself is optimized, the overhead imposed by the pre-scanner ends up being prohibitive (or at least deplorable).


So Yu Wu et al. use a simpler (and I guess faster) pre-scanner. They don’t attempt to find wholly optimal chunk boundaries, and they don’t attempt to ensure that the parse context at the chunk boundary is completely understood by the pre-scanner. They don’t go into great detail on the chunking method, but if I have understood it correctly, the primary criteria for division of the XML document into chunks are that (1) the chunks should be of about the same size, (2) there should be at least one chunk per thread, (3) other things being equal, fewer chunks is better (because this reduces the cost of post-processing), and (4) each chunk should start with a left angle bracket.

Until I looked at the paper again just now I thought each chunk had to begin with something that looks like a start-tag, but I can’t find that constraint in the paper. Maybe I hallucinated it.

So while I don’t know the details of the Intel pre-scanner, I imagine a pre-scanner that just looks for an angle bracket in the neighborhood of the desired chunk boundary: if you have an 800-MB document you want to divide into eight chunks (in order to allocate them to eight threads, say), my imaginary pre-scanner just looks in the vicinity of byte offset 100,000,000 for a left angle bracket, scanning forwards and/or backwards until it finds one. If we are reading the document from random-access storage, the pre-scanner doesn’t even have to scan the body of the chunk.

Some readers will be saying to themselves at this point “But wait, not everything that looks like a start-tag or an end-tag is necessarily a start-tag or an end-tag. It might be in a CDATA section. It might be in a comment. What then?”

Hold that thought. You’re quite right, and we’ll come back to it.

Parallel parsing

Now, each thread is given a chunk of XML to parse, but only one thread gets to begin at the beginning of the document, so only one thread actually knows where it is. The other threads are all following the advice of the Latin poet Horace, who recommends beginning in the middle (medias in res). (I’m tempted to see if we could call this parsing technique Horatian parsing, for that reason. That’s probably not going to fly in the community at large, but it’s more compact than “hybrid parallel-processing parser” or any other descriptive phrase I can come up with denoting the algorithm presented by the Intel team, so I’ll use it in the rest of this description.)

The nature of XML, however, is that the element structure is fairly explicit in the document and doesn’t require a lot of context information. When in a chunk you see a start-tag, some content, and a matching end-tag, you know that’s an element, and you can build the appropriate data structure. But you will also run into some things that you can’t handle all by yourself: unmatched start-tags (for elements that end after the current chunk), unmatched end-tags (for elements that started in earlier chunks), and undeclared namespace prefixes (for namespace bindings declared in some ancestor element). The Horatian threads keep track of each of these phenomena, and produce as their output a convenient representation of parts of a parse tree (think DOM, not SAX), and lists of unmatched start-tags, end-tags, and namespace prefixes, together with some miscellaneous context information.

The post-processor can use the lists of unmatched bits to knit the data structures together: the last unmatched start-tag found in chunk 1 matches the first unmatched end-tag found in chunk 2, and so on.

The post-processor can also use the context information to see whether any parsing needs to be redone. (Those of you who were worried about misleading angle brackets in comments and CDATA sections and so on? Remember I told you to hold that thought? OK, put it down now. This is where we deal with it.)

If chunk n ended in the middle of a comment or in a CDATA section, then the parsing of chunk n + 1 will need to be redone. But if we have divided the document into eight chunks, and one chunk, or four, need a second parsing, we are still running at about four times the speed of a sequential parser.


Once the chunks have been knit together (and, crucially, once all namespace prefixes are bound), the same chunking is used to perform schema-validity assessment: one thread validates chunk 1, another validates chunk 2, etc.

I won’t go into the fixup needed to knit together the various partial results; it suffices to observe that in both XSD and Relax NG, the validation needed for an element cannot be determined reliably without knowing its ancestry. The consequence is that validation cannot be performed reliably without doing the post-parsing fixup first. (You could guess which element declaration to use, I suppose, but I think the rate of wrong guesses would be too high.)

If you want to pass appinfo from the governing element declaration to the consumer, you will also in pathological cases need to know the left siblings of the element, in order to know where it occurs in the content model of its parent, so you can know which appinfo to use. I expect any schema validator designed for this kind of optimistic concurrency will therefore decline to expose appinfo from element declarations.


In theory, if parallelization imposes no overhead, then by using two threads you get a two-fold speedup, four threads gets a four-fold speedup, etc.

In practice, the test results presented by the Intel group show, as one might expect, that there is some non-negligeable overhead in this scheme. For four threads, their stand-alone parser shows a slightly better than two-fold speedup; for eight threads, slightly better than four-fold. Some of the overhead, clearly, is in the pre-scanning and the post-processing, but if I’m reading their graphs correctly, for eight threads neither of these processes comes close to the amount of time spent in parsing and validation, so I’m guessing that the main performance hit comes from the need to re-parse some chunks.

Small documents (which the paper describes as those smaller than 64 KB) do not benefit from the parallelization, and larger documents (larger than 1 MB) benefit more than medium-sized ones. They didn’t talk at any length about the nature of the test data, beyond saying that it was real data from real customers.

Optimization can be a real pain sometimes, and listening to people talk about optimization can be a good cure for insomnia. But sometimes, someone has a really simple Big Idea that makes sense even to someone as ignorant as me, and then it can be an awful lot of fun to hear about it and think about the implications.

Optimistic concurrency for XML processing: a Big Idea whose implications can be huge.