One day on the Xen-users list

Just one of my little rants about where we could be and how the Xen community(read: we) constantly drops the bucket.
I’m mostly putting this up for entertainment, if we can’t change things we can at least have some fun with them.


Xen limitations : 256 disks devices limit per hypervisor

Someone hardcoded a limit of 256 virtual disk devices into Xen, this even affects the most current Xen version. That means, you can’t have more than 256 disks for ALL VMs on your host. Of course not an issue if you’re just doing dev work, right?

It is unknown what will happen by changing that limit and it seems noone is currently considering working to fix this. Reminds me of the times of the non-PAE i386 Xen kernels, or, even better, recompiling due to xen heap size issues.


PCI-passthru Infiniband card panics the domU at startup

People having issues with Infiniband card PCI passthrough to domU. Fullblown Xen PV domU infiniband support was completed in 2006, but ignored in upstream. So 5 years later, people still have issues with cludgy workarounds.


VM disk I/O limit patch

(The last reply suggests everyone just update to the latest linux 2.6.39 or to the 3.0 dev kernel. This clearly shows a large understanding about running production servers. In think in short it reads “just fuck the users”)

Another highlight was the idea of having all bw limiting right to the (linux/kvm) upstream so they’ll be more easily usable for KVM so that KVM users can more easily switch to Xen, instead of adding them to Xen.


overcommit xen 4.0

A user asks if he can start a second VM if the first one is consuming too much memory. He is told that Xen has no overcommit facilities.

This is wrong 3! times – he could balloon the first VM, he could run tmem, he could run tmem2
On the other hand, the OP definitely lacked any measurable amount of clue.



This was the single non-sucky thread with a pointer to a current howto on using Remus. I also brought up the fact that the Xen Wiki has ~1 yr old “status info”. The poster that linked to it felt sorry for linking to outdated content. I never meant to blame *him* I just wonder why noone on the remus project is able to check the 1(!) single(!) wiki page on Remus a few times a year. Maybe it’s like the FreeBSD/Xen wiki where the maintainers don’t maintain it and ignore requests for permission to edit it?


3 thoughts on “One day on the Xen-users list

  1. Debian 6.0 dom0/xen4 even don’t work with Infiniband HCA. Something to do with xen memory mapping. Possibly it is already fixed as I’v seen some patches (very easy fixes).

    I have this one machine with 10 hardrives and 4 software raid arrays and can not get more than 10mb/sec read/write on domU nor dom0. Be it SAS 15K RAID1 or SATA RAID10 over 4 drives – no more than 10MB/sec. Strange. With less drives working as one would expect.

    Slowly getting tired of all this 😦

    • Hi, I have only one debian host, has 1 raid array and uses the stock kernel / raid1.
      performance is good there.
      But one of my workmates saw something similar to your issue when he reduced dom0_mem in his debian dom0.

      I can’t explain it since there should be no notable buffering on phy: / lvm devices but it happened. Can you give 1-2 extra GB to dom0 and see what happens?

      And don’t get frustrated, take it with some humor, i.e. imagine what a crazy data path the IO must take to have that slow performance. I think a 486 could not have easily handled it!

  2. Good idea, I will try this when go maintenance again. I have 512mb then increased it to 768mb. Not a big diference there.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s