Ian Pratt, of the University of Cambridge in England and the leader of the Xen project, said there were a number of reasons for the delay in including Xen in the kernel. Primarily, Xen 3.0 had suffered from a bit of feature creep. Physical Address Extension (PAE) 32b support and Virtualization Technology, for example, were added very late in the cycle. "We were aiming for an end-of-summer release, but this now looks on target for December," Pratt said."We hit this point a month or so back, and theres actually been a lot of activity since then," Pratt said. "Weve done a first cut reorganization of our patch into the form that was agreed on at the last Xen summit, forward ported it to the head of Linus [Torvalds] tree, and put out a call for help to Red Hat, SuSE, IBM, HP and all the other stakeholders to help us beat it into shape. Its great to see them stepping up and promising to commit some of their best guys to help." The technology is certainly ready for inclusion in the kernel, he said. Rearranging the patch into a form that fits in better with the existing code base needs to happen first, but this is fairly mechanical. "We maintained our patch in a form that made our life easier, and helped us track stable Linux versions while getting the stability of our own software right. Its now time to make the change," Pratt said. Pratts confident that a patch will be ready to be submitted for inclusion in the kernel within two months, as none of the reorganization and cleanup work that needs to happen is very hard, "but it is essential we get the aesthetics right. But whether Andrew/Linus accept it is a different matter," Pratt said. He welcomes Red Hats support, saying they have good engineers that are well-known and respected in the Linux community, which is bound to make the process run smoother. "SuSE, IBM, HP are all helping the XenSource team too, just maybe not so publicly," he said. Asked what recent contributions the vendors have made to the technology, Pratt said that IBMs MAC stuff is in, as is support for Intel VT-x. Advanced Micro Devices Inc.s Pacifica support is working well too, but will not make Xen Version 3.0.0. "Further down the line were doing some cool stuff with I/O vendors that will result in zero cost I/O virtualization. HP had contributed some useful tools for performance profiling and instrumentation, while SuSE had had helped with PAE support and Intel with x86_64," he said. A lot of companies, most notably IBM, are also helping with testing. "Weve had a lot of support from individuals in the Xen community too," Pratt said. "Xen 3.0 is a big team effort. It is just taking a little longer than wed hoped." John Loiacono, executive vice president for software at Sun Microsystems Inc., welcomed the move to drive the virtualization technology around Linux forward. Any aggressive move by Red Hat to get the technology into the Linux kernel will be fully supported by Sun, which is embracing the Xen virtualization technology across its products and plat-forms. It has some of its brightest engineers working on this and is collaborating with others in the open-source community, Loiacono said. Even Sam Greenblatt, a senior vice president at Computer Associates International Inc., told eWEEK that he is pleased with the progress made with Xen. CA will support anything going into the kernel that supports virtualization. "Its come a long way. We just want to be careful to make sure it goes in the right way," he said. That marks a significant turnaround from earlier this year, when Greenblatt told eWEEK, "We think [Xen] is great innovation, but its concept of virtualization is still not to the point that we want to see in there [the kernel]." Click here to read how some in the community are worried that the rapid feature changes could make the Linux kernel too large. Red Hats Stevens said his goal is to make virtualization as ubiquitous as possible, thereby allowing customers to decide whether they need it or not. "Our strategy is around how to make it ubiquitous, what are all the issues that make it ubiquitous and part of the platform," he said. "But when you get there, a range of great benefits comes with it, like the agility of being able to migrate workloads, suspend workloads, drive up utilization on a system because now you can isolate workloads from each other whereas before an entire box had to be dedicated to a specific application," he said. But this will require an entirely new management infrastructure around it as those that exist today revolve around managing physical boxes. "While people are extending the existing management platforms to virtual boxes, what they are not doing is changing the management paradigm, and that needs to change to one where applications, systems and resources are just meeting for a point in time. This needs to be more of a brokering system than the management of physical systems," Stevens said. While XenSource founder Pratt said an "entirely new management paradigm and infrastructure" is not needed to make good use of virtualization, it would enable this and could "save the big shops a ton of money by doing so." "XenSource will be one of the companies offering management solutions around Xen, along with a bunch of others. Hopefully XenSources will be best as weve been working on this topic in the university for a long time, so we have some deep control and automation facilities rather than just a flashy GUI," he said. Check out eWEEK.coms for the latest open-source news, reviews and analysis.
It didnt make much sense to start preparing patches for sending upstream until the Xen 3 guest API was close to being frozen, because there is a significant resource cost in maintaining multiple trees, he said.