- From: Noah Mendelsohn/Cambridge/IBM <noah_mendelsohn@us.ibm.com>
- Date: Thu, 20 Feb 2003 09:44:38 -0500
- To: Elliotte Rusty Harold <elharo@metalab.unc.edu>
- Cc: michael@neonym.net, www-tag@w3.org
Elliotte Rusty Harold writes: >> You could have created the right technology >> for your needs, without impacting anyone >> else who didn't need it. If it happened to >> look a lot like XML, no big deal. I'm sympathetic to this argument (remember I said that I'm not particular at this point an advocate of binary XML). However, I think the problem is a bit more subtle than the above might imply. Yes, many of the advantages of XML come from not being all things to all people. XML solves certain problems at moderate performance using text-based approaches. What I think is more subtle is that my employer (and our competitors as far as I can tell) are trying to >unify< their application models. The requirement to build binary serializations of XML is not primarily for applications that are independent of the traditional uses of XML, it's for the same or related applications. Let's say you take in orders over the public network using traditional XML 1.0 documents. The cost of character parsing is well justified by the known advantages of XML: platform independence, self-description, tremendous availability of tools, etc. The problem is that once those orders get inside your organization you have to deal with large numbers of them. You also have to build related applications (inventory perhaps) that have high-volume data processing requirements and that work on some of the same data that's in the orders. The point is that the high-volume applications are processing the same data that's coming in and out using traditional XML. Even if they weren't directly processing the same data, it's quite compelling to have the same data models, same typing system, and where possible the same tools to operate on that data. While it's true that the serializers and parsers are different, many other tools can be shared. Most notably these include XML-aware databases, message queueing systems, modelling systems and to some degree data binding logic. Organizations such as IBM are committed to building high performance, high throughput applications using XML-compatible data models. The only questions I see are whether this can best be achieved by optimizing the processing of traditional character serializations, whether there are times when binary serializations are a better tradeoff, and if so whether it's better on balance to standardize those representations. I do agree that these are not easy tradeoffs and that the answers aren't immediately obvious. So, for better or worse, these are key reasons why I see some push toward binary serializations of the XML model. It's not just to build applications that are similar to XML but with higher performance, it's to get architectural synergy between the higher and lower performance design points. ------------------------------------------------------------------ Noah Mendelsohn Voice: 1-617-693-4036 IBM Corporation Fax: 1-617-693-8676 One Rogers Street Cambridge, MA 02142 ------------------------------------------------------------------ Elliotte Rusty Harold <elharo@metalab.unc.edu> Sent by: www-tag-request@w3.org 02/20/03 08:33 AM To: Michael Mealling <michael@neonym.net> cc: www-tag@w3.org, (bcc: Noah Mendelsohn/Cambridge/IBM) Subject: Re: [binaryXML-30] Binary XML problem statement. Categories: At 7:47 AM -0500 2/20/03, Michael Mealling wrote: >We looked at just about everything we could find. ASN.1 had its typical >problems with encapsulation of unkown data types. Most of the other were >to simplistic since we had to deal with things such as limited use of >namespaces, encodings, schemas, etc. In other words, to handle the >applications we're after we would've ended up creating something almost >exactly like XML 1.0 in a binary form, just not called XML. That >would've been a rather silly thing to do given perfectly reasonable >solutions that already exist. Actually, that's exactly what you should have done. You could have created the right technology for your needs, without impacting anyone else who didn't need it. If it happened to look a lot like XML, no big deal. >I find it much more productive to start with something that solves all >of the required problems except one and fix that rather than throw out a >perfectly good solution and build one from scratch. For now I've decided >to use WBXML since its about as good as you can get for our application. >But if the W3C were to come up with a better or more accepted standard >we would use that (if it solved the problem, of course. gzip doesn't). The problem is that this so-called binary XML may save you time, but it pollutes the space for the rest of us. Maybe it's a net-gain for your project. It's a net loss for the community, though. -- +-----------------------+------------------------+-------------------+ | Elliotte Rusty Harold | elharo@metalab.unc.edu | Writer/Programmer | +-----------------------+------------------------+-------------------+ | Processing XML with Java (Addison-Wesley, 2002) | | http://www.cafeconleche.org/books/xmljava | | http://www.amazon.com/exec/obidos/ISBN%3D0201771861/cafeaulaitA | +----------------------------------+---------------------------------+ | Read Cafe au Lait for Java News: http://www.cafeaulait.org/ | | Read Cafe con Leche for XML News: http://www.cafeconleche.org/ | +----------------------------------+---------------------------------+
Received on Thursday, 20 February 2003 09:48:54 UTC