Subject: Re: [vserver] Need advice on which path to take for IPv6 support
From: Paul Kyzivat <pkyzivat@cisco.com>
Date: Fri, 08 Jan 2010 12:11:40 -0500



Ed W wrote:
> On 07/01/2010 18:04, Paul Kyzivat wrote:
>>
>> (We had previously been using patch-2.6.14.3-vs2.01.diff. When we 
>> migrated to patch-2.6.22.10-vs2.2.0.5.diff some of our guests 
>> encountered incompatibilities that we didn't discover until after the 
>> fact. That is making people gun shy.
> 
> Opinions are ten a penny, so just to offer one...

:-)

> Given that one of the major benefits I find from virtualisation is that 
> you can clone the image very easily and test it elsewhere.  Would this 
> not be a viable strategy with regards to upgrading no matter what you go 
> for?
> 
> I concur with Herbert though that it's always worth striving to stay as 
> current as possible in the Linux world.  For better or for worse all the 
> bug fixing and features tend to follow HEAD and backports tend to 
> struggle a bit (in general)...
> 
> I personally didn't notice much difference upgrading from exactly your 
> kernel to something around the 2.6.29 mark with the newer patch line.  
> However, you should also not take my experience too seriously since I 
> barely stress the vserver stuff and have only very simple needs...  
> Perhaps others will offer you some better notes on compatibility.
> 
> I think this upgrade should be workable for you, but I guess the point 
> is to get some spare hardware and get some testing in ahead of changing 
> out the live machine? (I think someone previously asked about running 
> vserver under vmware/kvm/xen or similar...)

Of course we know we have to test. Part of the querying is to get some 
sense of "how much" we need to test, and particularly how much others 
who use our stuff have to test. I'm personally in favor of getting as 
current as we practically can without being on the bleeding edge. But it 
isn't in the cards to go beyond 2.6.22 right now, so that is a limitation.

Its other people's code running in our vservers that we are most 
concerned with. And we can't test many of those before we release. We 
have to rely on alpha/beta releases, etc. for that. So I'm just trying 
to understand how big a perturbation this upgrade will be.

> Good luck

	Thanks,
	Paul