I'm glad to see, that Xen is coming to the PPC platform. The Xen/PPC wiki at http://wiki.xensource.com/xenwiki/Xen/PPC shows, that there is nice progress in porting. Looking this page and at the mailing list archives, some questions come to my mind:
- it is ported now to one version of OF, and it seems to be a tough job. Genesi uses it's own implementation of OpenFirmware. How difficult is to add support for it, when OSW (a dual 970MP machine) becomes available?
- on an x86 machine, the number of virtual machines is just limited by available RAM and independent from the number of processors. Here I have the feeling, that number of virtual machines is tied to the number of processor cores. Is it right, or I just misunderstand something.
CzP

The Open Firmware part shouldn't be too difficult. Xen doesn't do a whole lot with the device tree itself; mostly it just passes it up to the dom0 kernel.

However, it should go without saying that the more divergences Genesi has made from other firmware implementations, the harder it is to port any software to it. I've run into that firsthand on GRUB, and you can see other workarounds in the Linux kernel.
I'm not sure why you think there's some different model for PowerPC Xen; there isn't.

Hi.

We've been following XEN on PowerPC for some time, but due to lack of time we did not have the opportunity to contribute to the effort.

In regard to your questions:
The OpenFirmware should pose no serious problems and workarounds can be made when the other problems are solved
The hard parts of XEN are the virtual memory management, the live migration implementation and the grant tables mechanism.

The number of virtual machines is as far as I know not limited on PowerPC, and this is the first time I heard this. Where did you find that?

We have been interested in XEN on PowerPC for some time, and would like to setup a lab in the University for this and other things based on PowerPC from the embedded to the servers.

Best regards,
Johan

_________________
Johan Dams, Genesi USA Inc.
Director, Software Engineering

Yep, I have a blog... PurpleAlienPlanet

From a development perspective, Xen would be extremely useful. Being able to run full 64bit userland besides a 32bit userland and a mixed userland at the same time on the same machine would be extremely helpful. Chroots are not always useful, especially when you have to run daemons etc. Xen is the way to go and I'm looking forward to seeing it implemented on the OSW.

This is a good discussion. We would like to extend this collaboration.

Here are two sections of a Business Plan we developed for Power.org around the Open Server Workstation and Xen. You are welcome to read them and place your comments here. Of course, this does not provide a complete snapshot of Genesi's activity, but there is plenty of good information included. We hope you find it interesting and useful!

Download Part #1

Download Part #2

Please provide feedback to us here. Thanks.

Raquel and Bill

_________________
http://bbrv.blogspot.com

My original question came after reading a document just posted by BBRV at http://www.genesippc.com/files/power.org/DP-WG_1.pdf

Near to the bottom of page 8 it says: I also read Xen on PPC wiki at http://wiki.xensource.com/xenwiki/Xen/PPC which says indirectly, that Xen PPC runs only on hypervisor enabled processors (there is a low prioryt task for non hypervisor processors). And there is yet another document, a RedPaper about IBM JS21, which also has 970MP. It's available at http://www.redbooks.ibm.com/redpapers/pdfs/redp4130.pdf

For me it's a bit confusing now, how Xen and hypervisor are related, and if the limits described in the RedPaper apply to Xen, etc. Could you explain this?

Thanks,
CzP

I guess that's a little misleading. However, I don't have any control over what other people say about Xen. :)
That is correct. The main Xen/PPC port is for hypervisor-enabled processors, in particular the 970 (and variants). Other processors that have the requisite features include the Cell Broadband Engine, the announced PWRficient chips, and of course POWER5. None of these are current targets.
Yes, the JS21 uses 970MP. I'm not sure how that's confusing, but to elaborate, JS21 allows for partitioning (unlike JS20), which is a feature of IBM's firmware that ships on those systems.

I hope this has cleared things up. (Also, in case it wasn't clear, I'm one of the two main Xen/PPC developers.)[/quote]

OK, thanks.
I know, as I check the mailing list archives regularly I use Xen on x86 on a daily basis and I'm waiting for the PPC version impatiently. I'm not a coder, but a software tester, so I guess, I still have to wait some time. As far as I can remember, I red somewhere, that Xen on PPC is exptected to be out in Q3. Do you think, it's possible?

CzP

By all means let's clear that up! If we have something therein incorrect we want to fix it. We actually were hoping for this sort of collaboration.

Thanks Hollis for stopping by! Thanks to you as well Peter and all the rest of you for participating.

We seek the best possible success for the OSW and Xen sure does seem like a good thing on which to direct our focus.

R&B

_________________
http://bbrv.blogspot.com

It's possible. There are two important questions though:
- What are your functionality/performance expectations? Even if it's "out", performance may suck, SMP guests may not be supported, etc.
- What hardware do you expect to run it on? Right now Maple is the only well-supported option. We haven't seen an OSW yet, so hard to guess about that one.

I think, I can join testing as soon, as there is console support with disk and/or network support. I have a good sense of finding ways to crash and/or freeze applications, find performance problems, DoS possibilites, etc., just by starting to use the given application. I'm a long time SUSE beta tester with an impressive list of discovered bugs
Ooops, you got a point: I can start testing as soon, as OSW is here. And that's another two months.

Another question: Xen is able to run unmodified Windows guests on some recent Intel processors. Do guests need to be modified to run under Xen PPC? As far as I can undertand, hypervisor-enabled processors mean hardware based support for running multiple OS-s on the same machine.

And back to BBRV-s question: I guess, this part of the sentence can be left out completely: "Each of the cores in the 970MP can run one operating system and"

CzP

Great to hear it. We will definitely need help flushing out bugs, and I expect it will be a target-rich environment initially... ;)
The guests will need modification. The hypervisor extensions that IBM developed for PowerPC assumed some (but minimal) guest modification, because this gets you superior performance. Also, IBM controlled both the hardware and the OSs, unlike Intel and AMD.

The non-hypervisor mode work being done requires much more extensive guest modification, because we don't have hardware assists (in particular, a hypervisor privilege level).

The Intel VT and AMD SVM extensions support unmodified OSs (and we can guess which OS might have made that a priority for them). I'm not familiar enough with those architectures to say for certain, but I think it's safe to say that modified guests are expected to get better performance on Xen than unmodified.

As far as i can see Xen will need a lot of testing for each release, which i can join as soon as possible. Im not that familiar with the codebase, but when i become familiar with it i will also contribute code.

BB & RV,

Right now I'm working on virtualizing the development infrastructure at a large client (7,000+ employee health care system with 4 primary locations). Due to this, I have some expertise

If you're going to virtualize and mention your solution as cost-effective, you should mention:

1. RAID array controller support. Many virtual machines = much I/O. Good RAID controller support, specifically for high-performance models, is a must. Right now, Apple can't do what you do with the BIOS extensions to more easily write drivers for their platform. Do these extensions also extend to PCI Express and PCI-X?

Talk this one up .

2. mySQL and Postgres Support. This is also critical here, as people look at IBM and think DB2. If you can get decent performance from those two (and I know you're partners with mySQL), you've got another good tack.

3. Java support. Again, there are many popular application servers out there, and many of them run Java. A partnership with IBM for Websphere and/or JBoss would work really well here.

4. HPC support. This is another big one, as Apple is bowing out of the PPC market. IBM's already got their foothold. People can and will be looking for PPC vendors due to the fact that Altivec makes lots of complex parallel vector calculations faster. Apple does have excellent support for high-performance network and I/O cards for clustering. However, since you're primarily Linux, you can do the same, and potentially support more cards, such as the 10GbE cards.

5. HA support for certain apps. Linux has it. I know your solution does. However, every vendor out there is touting it.

6. NX/Tarantella/Remote Desktop support. You and I have talked about this in the past, and I ran an ODW through the paces doing system admin work with VNC and rdesktop. If you're going to talk virtualization, using this as the basis for a remote desktop server running NX or Tarantella is not out of the question .

Again, RAID/SAN support here is a biggie. Linux already has decent support for it. What good is it to buy a few racks of these when they don't plug into your EMC or Pillar SAN? . Apple's actually been doing quite well with their SATA RAID solution.

Thanks,

MBP

Yes they do. Anything which presents a PCI-defined "Option ROM".

_________________
Matt Sealey