Virtualization Blog

Discussions and observations on virtualization.

Beyond Creedence - XenServer 2015 Planning

In a few weeks James Bulpin and I will be at the Xen Project Developers Summit in Chicago, and some of our discussions will be about the future of XenServer, and more importantly to the community "What comes after Creedence?". With the Creedence alpha program we're seeing a level of community engagement which has honestly exceeded my expectations. I attribute this to the significant improvements in the platform, but also the level of transparency we've had with respect to early access to pre-release builds.

While it was pretty obvious what we needed to do to make Creedence viable, your input is important to the future success of XenServer. With that in mind, we'd like to hear what platform improvements you'd find most valuable. When I speak of platform improvements, I'm thinking of things like storage, networking, core virtualization, performance, scalability and operating system support. I'm not thinking of things which can be classified as data center or virtualization management, so things like network management, disaster recovery, or virtual machine provisioning are out of scope. Based on the blog comments for the various alpha announcements, we already know that CentOS 7 dom0, NFS4 and Ceph are on your wish lists, but what else?

Internally we use a "How would you spend $100?" model to prioritize changes, and if you were interested in providing feedback following that model, it would be ideal. If you've never used this model before, it's pretty simple. Write down the things you'd want to see (optionally with a "why" beside them), and then given a budget of $100. Spend the $100 by allocating it to your desired functionality, and anything with a zero is removed. This has the benefit of focusing on the high value changes without worrying about complexity. If you'd like to provide input, please do so in the comments section below, and let's see what the future of XenServer in 2015 looks like from your perspective.     

2 weeks to Xen Project Developer Summit - What to ...
In-memory read caching for XenServer

Related Posts

 

Comments 103

David Reade on Wednesday, 30 July 2014 17:35

Could Changed Block Tracking (CBT) be considered as a feature please to speed up incremental backups? We use Unitrends to perform backups to an NFS store every night incrementally which can take forever regardless of the amount of data backed up.

0
Could Changed Block Tracking (CBT) be considered as a feature please to speed up incremental backups? We use Unitrends to perform backups to an NFS store every night incrementally which can take forever regardless of the amount of data backed up.
Guest - Rachel Berry on Sunday, 03 August 2014 17:03

Yes, I think that sounds very useful and one I long desired

0
Yes, I think that sounds very useful and one I long desired
Guest - Rachel Berry on Sunday, 03 August 2014 17:03

Yes, I think that sounds very useful and one I long desired

0
Yes, I think that sounds very useful and one I long desired
Keith Walker on Tuesday, 31 March 2015 18:32

CBT, the entire $100.

0
CBT, the entire $100.
Andrew on Wednesday, 20 May 2015 13:12

$80 for CBT, $20 for online disk expansion.

0
$80 for CBT, $20 for online disk expansion.
Jarian on Wednesday, 30 July 2014 19:05

I like that NFS4 and Ceph are mentioned above.

1
I like that NFS4 and Ceph are mentioned above.
JK Benedict on Wednesday, 30 July 2014 19:11

I love this post!

If I had $100 dollars, I'd spend $50.00 of it to roll in older PV tools for Legacy OS/App support as so I don't get stuck on a particular XenServer release.

I'd spend $25 more of my time to make coalescing more dynamic: checking server stats to throttle cleanup (seconds to run/Mibs to coalesce).

Lastly, I would use the remaining $25 to add shadow passwords: allowing non-root users to access XenCenter while logging their activity. Much like RBAC, this would help diagnose who/what/when in a multi-administrator environment.

And I'd give my time away freely to help in all these aspects so the $100 comes back to the community and us!

Thanks for this opportunity, Tim and we appreciate all you do!

-jkbs
@xenfomation

1
I love this post! If I had $100 dollars, I'd spend $50.00 of it to roll in older PV tools for Legacy OS/App support as so I don't get stuck on a particular XenServer release. I'd spend $25 more of my time to make coalescing more dynamic: checking server stats to throttle cleanup (seconds to run/Mibs to coalesce). Lastly, I would use the remaining $25 to add shadow passwords: allowing non-root users to access XenCenter while logging their activity. Much like RBAC, this would help diagnose who/what/when in a multi-administrator environment. And I'd give my time away freely to help in all these aspects so the $100 comes back to the community and us! Thanks for this opportunity, Tim and we appreciate all you do! -jkbs @xenfomation
Guest - Albert on Wednesday, 30 July 2014 19:38

If i had $100, I'd spend all for support Oracle Solaris under a new Xenserver

0
If i had $100, I'd spend all for support Oracle Solaris under a new Xenserver
Guest - Rachel Berry on Sunday, 03 August 2014 17:02

http://discussions.citrix.com/topic/312251-solaris-10-installation-to-xenserver-602/

See my comment (Rachel), XenServer cannot support Solaris alone, you need to persuade Oracle, we would be happy to do so if the did.

0
http://discussions.citrix.com/topic/312251-solaris-10-installation-to-xenserver-602/ See my comment (Rachel), XenServer cannot support Solaris alone, you need to persuade Oracle, we would be happy to do so if the did.
Tobias Kreidl on Friday, 08 August 2014 15:14

How about supporting an open Solaris variation instead, like OpenIndiana or Illumos?

0
How about supporting an open Solaris variation instead, like OpenIndiana or Illumos?
Guest - ONOV on Wednesday, 30 July 2014 23:49

Except it would then cost you $1000 per core per year to license running your Solaris VM according to Oracle's licensing models. :-(

0
Except it would then cost you $1000 per core per year to license running your Solaris VM according to Oracle's licensing models. :-(
James on Wednesday, 19 August 2015 04:14

Consider positioning yourself with Joyent SmartOS. It is the spiritual and commercial successor to OpenSolaris after OpenIndana without the Oracle (Formerly Sun) or SGI IRIX license (Contract) model of per socket per annum for service packs or per-line fixes. If docker and droplets are any indication of what can happen, it should be considered despite the issues which appear on the surface of this argument.

The whole idea is NFSv3/4, non FUSE ZFS, multi-pathing and RBAC/ACL's exposed through all network-level services, which would be more robust than a Hodge-podge of components, despite as flexible as Linux is. Since "Xen Project" has released every component freely, and Xen is supported since OpenSolaris B85.

The only remaining issues that I can see would focus on missing (if any) device drivers including unprivileged or Linux specific, De-coupling certain key integration which may exist with XenServer as of this writing with Linux and moreover CentOS itself.

I would like to see integration for the following, though it may be a pipe dream as I cam into eclectic and vintage things:

Support for PowerPC(64), MIPS(64), PA-RISC, VAX, Alpha AXP, m68K guest emulation through Xen but if not repackaged SIMH, PersonalAlpha, EmuVM, SIMICS, QEMU, Bochs, GXemul automation and provisioning support so it would be possible to host and offer commercial and personal "migration" services for hobbyists, computer scientists, etc. to run and maintain systems such as 2.11BSD, OpenVMS, (Free/Net/Open) BSD, (Open)Solaris, and other "eclectic" systems of our yesteryear without the monotonous labor that is required to set such things up piecemeal.

The performance arguments are no longer warranted as even the free Alpha AXP emulators (As well as VAXen) can run 2x more computations than the real thing on realistic x86_64 server hardware such as 3.4GHz Xeon 6-Core's with hyper-threading (E1650/v2) in single or multiple socket configurations.

Even gaming applications have proven it is as of 2015 feasible to emulate a Sony Playstation 1/2 (MIPSSel) and games such as Gran Turismo on a i7 4790k with more FPS than was ever possible then, let alone on supported industry scaled tech such as Oracle EXAData (1000 Cores Per Rack over FC/iSCSI HBA) or a custom Linux cluster running on top of instances of XenServer for example.

While there's OpenIndiana, the issue of integration and standard of quality would have to be addressed as well as the small but still existent encumbrances of specific device driver code that were never released under CDDL, gratis and/or libre license.

I would say $25 to provide a level of assurance for compatibility while moving forward with developments in the upstream Linux kernel itself, $25 for at least getting experimental support going for host HVMs such as (Open)Solaris or FreeBSD which have the legal and functional advantages of mature and proper ZFS support, while XenServer on CentOS only exposes ext3/reiserfs (MurderFS) by default, and finally $50 for domU and/or automation/provisioning support for non ARM/x86/x86_64 processors.

If only a handful of operating systems could be supported it would be a shock to the competition in big data businesses still paying a premium for licensing specific products that provide this level of portability. The irony of these legacy systems is that incidentally can boot from ISO in other industry offered emulator packages despite predating CD-ROM's in their entirety as by way of tape drive emulation for instance with an ISO as a back-end. (Since it's sequential all the same as to such)

Thanks for reading.

0
Consider positioning yourself with Joyent SmartOS. It is the spiritual and commercial successor to OpenSolaris after OpenIndana without the Oracle (Formerly Sun) or SGI IRIX license (Contract) model of per socket per annum for service packs or per-line fixes. If docker and droplets are any indication of what can happen, it should be considered despite the issues which appear on the surface of this argument. The whole idea is NFSv3/4, non FUSE ZFS, multi-pathing and RBAC/ACL's exposed through all network-level services, which would be more robust than a Hodge-podge of components, despite as flexible as Linux is. Since "Xen Project" has released every component freely, and Xen is supported since OpenSolaris B85. The only remaining issues that I can see would focus on missing (if any) device drivers including unprivileged or Linux specific, De-coupling certain key integration which may exist with XenServer as of this writing with Linux and moreover CentOS itself. I would like to see integration for the following, though it may be a pipe dream as I cam into eclectic and vintage things: Support for PowerPC(64), MIPS(64), PA-RISC, VAX, Alpha AXP, m68K guest emulation through Xen but if not repackaged SIMH, PersonalAlpha, EmuVM, SIMICS, QEMU, Bochs, GXemul automation and provisioning support so it would be possible to host and offer commercial and personal "migration" services for hobbyists, computer scientists, etc. to run and maintain systems such as 2.11BSD, OpenVMS, (Free/Net/Open) BSD, (Open)Solaris, and other "eclectic" systems of our yesteryear without the monotonous labor that is required to set such things up piecemeal. The performance arguments are no longer warranted as even the free Alpha AXP emulators (As well as VAXen) can run 2x more computations than the real thing on realistic x86_64 server hardware such as 3.4GHz Xeon 6-Core's with hyper-threading (E1650/v2) in single or multiple socket configurations. Even gaming applications have proven it is as of 2015 feasible to emulate a Sony Playstation 1/2 (MIPSSel) and games such as Gran Turismo on a i7 4790k with more FPS than was ever possible then, let alone on supported industry scaled tech such as Oracle EXAData (1000 Cores Per Rack over FC/iSCSI HBA) or a custom Linux cluster running on top of instances of XenServer for example. While there's OpenIndiana, the issue of integration and standard of quality would have to be addressed as well as the small but still existent encumbrances of specific device driver code that were never released under CDDL, gratis and/or libre license. I would say $25 to provide a level of assurance for compatibility while moving forward with developments in the upstream Linux kernel itself, $25 for at least getting experimental support going for host HVMs such as (Open)Solaris or FreeBSD which have the legal and functional advantages of mature and proper ZFS support, while XenServer on CentOS only exposes ext3/reiserfs (MurderFS) by default, and finally $50 for domU and/or automation/provisioning support for non ARM/x86/x86_64 processors. If only a handful of operating systems could be supported it would be a shock to the competition in big data businesses still paying a premium for licensing specific products that provide this level of portability. The irony of these legacy systems is that incidentally can boot from ISO in other industry offered emulator packages despite predating CD-ROM's in their entirety as by way of tape drive emulation for instance with an ISO as a back-end. (Since it's sequential all the same as to such) Thanks for reading.
David Taylor on Wednesday, 30 July 2014 23:59

I'd have to say our key issues are around P2V conversion and more specifically Linux HVM to PV conversion. I'd probably spend about $35 on that. Linux support for XenCenter is also a major one, the current Linux clients aren't there for production yet, I'd put about $15 there. The other $50 I'd love to see hit a lot of small operational issues like screen buffer support for linux hosts and stability/recoverability following a hypervisor crash.

0
I'd have to say our key issues are around P2V conversion and more specifically Linux HVM to PV conversion. I'd probably spend about $35 on that. Linux support for XenCenter is also a major one, the current Linux clients aren't there for production yet, I'd put about $15 there. The other $50 I'd love to see hit a lot of small operational issues like screen buffer support for linux hosts and stability/recoverability following a hypervisor crash.
Tim Mackey on Thursday, 31 July 2014 13:51

David,

If you can give examples of the pain points in your "operational issues"; I'd like to hear them. I personally have a bunch of pain points I'd like to see addressed, and some of them have been around for years but have always tended to fall below the cut line for one reason or another.

-tim

0
David, If you can give examples of the pain points in your "operational issues"; I'd like to hear them. I personally have a bunch of pain points I'd like to see addressed, and some of them have been around for years but have always tended to fall below the cut line for one reason or another. -tim
David Taylor on Monday, 04 August 2014 22:57

A lot of the issues have been around stability, visibility and migration. So for instance in a two Xenserver cluster we had an issue where we were doing a rolling update and the master was put into maintenance mode but didn't pass the master function to the second server correctly and we ended up with 4 isolated machines that then couldn't find their virtual disk correctly and consequently couldn't start.
In terms of visibility just the errors being so generic, it would be much better to provide more detail in the XenCenter error log. Also graphing can be unpredictable sometimes it works correctly other times graphs just aren't there. So for me XenCenter is a big one especially the lack of a production ready linux client.
And the last big thing is the conversion of linux hosts to PV from HVM. THe lack of a clear path to full virtualisation is really putting a crimp on our ability to move to Xen.

0
A lot of the issues have been around stability, visibility and migration. So for instance in a two Xenserver cluster we had an issue where we were doing a rolling update and the master was put into maintenance mode but didn't pass the master function to the second server correctly and we ended up with 4 isolated machines that then couldn't find their virtual disk correctly and consequently couldn't start. In terms of visibility just the errors being so generic, it would be much better to provide more detail in the XenCenter error log. Also graphing can be unpredictable sometimes it works correctly other times graphs just aren't there. So for me XenCenter is a big one especially the lack of a production ready linux client. And the last big thing is the conversion of linux hosts to PV from HVM. THe lack of a clear path to full virtualisation is really putting a crimp on our ability to move to Xen.
Tobias Kreidl on Thursday, 31 July 2014 00:01

$30: Thin provisioning for block I/O (yes, Ceph can handle that) and NFS4 (YES!), $20: RAM Cache with Overflow to disk for MCS so XenServer can support it with that "other" provisioning service, $40: the ability to copy either running or shut-down VMs including storage to other locations and either leave them down if already running or start them up if not already running or copy from down-to-down state to provide easy, cheap ways of data replication and/or backup, $10: better granularity in XenCenter performance meter. Shucks, I'm already out of money...
-=Tobias

0
$30: Thin provisioning for block I/O (yes, Ceph can handle that) and NFS4 (YES!), $20: RAM Cache with Overflow to disk for MCS so XenServer can support it with that "other" provisioning service, $40: the ability to copy either running or shut-down VMs including storage to other locations and either leave them down if already running or start them up if not already running or copy from down-to-down state to provide easy, cheap ways of data replication and/or backup, $10: better granularity in XenCenter performance meter. Shucks, I'm already out of money... -=Tobias
Guest - Tessio Fechine on Sunday, 15 February 2015 22:52

Don't XenServer already has thin provisioning?
http://cdn.techgenix.com/media/upls/image012_23.jpg

0
Don't XenServer already has thin provisioning? [img]http://cdn.techgenix.com/media/upls/image012_23.jpg[/img]
Terry Wang on Thursday, 31 July 2014 00:39

If I had $100, I'll spend
- $30 to upgrade dom0 user-space to CentOS 7 x86_64
- $40 official support for running new Xen modes (especially for Linux): PVHVM (PVonHVM) and PVH (PV in HVM containers - new in Xen 4.4 and will stabilize). Benchmarking results show Linux PVHVM outperforms pure PV in many ways. PV is no longer the best mode for Linux guests.
- $10 Btrfs as local storage option - probably as tech preview - See how cool it is in CoreOS (Linux + systemd + LXC + Btrfs + Docker).
- $10 update e2fsprogs to support ext4 with Linux 3.10+, ditch ancient ext3 (local storage - thin provisioning)
- $10 Certify more OSes , e.g. PV Ubuntu 14.04 LTS, RHEL 7, CentOS 7, Oracle Linux 7, openSUSE 13.1 etc. (in fact, if PVHVM is officially supported, this is not a problem)

0
If I had $100, I'll spend - $30 to upgrade dom0 user-space to CentOS 7 x86_64 - $40 official support for running new Xen modes (especially for Linux): PVHVM (PVonHVM) and PVH (PV in HVM containers - new in Xen 4.4 and will stabilize). Benchmarking results show Linux PVHVM outperforms pure PV in many ways. PV is no longer the best mode for Linux guests. - $10 Btrfs as local storage option - probably as tech preview - See how cool it is in CoreOS (Linux + systemd + LXC + Btrfs + Docker). - $10 update e2fsprogs to support ext4 with Linux 3.10+, ditch ancient ext3 (local storage - thin provisioning) - $10 Certify more OSes , e.g. PV Ubuntu 14.04 LTS, RHEL 7, CentOS 7, Oracle Linux 7, openSUSE 13.1 etc. (in fact, if PVHVM is officially supported, this is not a problem)
Van Rue on Wednesday, 24 September 2014 18:09

$50 for btrfs support

0
$50 for btrfs support
Guest - Frediano Ziglio on Friday, 08 August 2014 10:22

- Ubuntu 14.04 Centos 7, RHEL 7 Oracle Linux 7 are going to be supported on Creedence.
- There is an installer option to use ext4 instead of ext3, at least for root partition.

0
- Ubuntu 14.04 Centos 7, RHEL 7 Oracle Linux 7 are going to be supported on Creedence. - There is an installer option to use ext4 instead of ext3, at least for root partition.

About XenServer

XenServer is the leading open source virtualization platform, powered by the Xen Project hypervisor and the XAPI toolstack. It is used in the world's largest clouds and enterprises.
 
Commercial support for XenServer is available from Citrix.