Steffen Weiberle wrote, On 05/06/2009 16:19:
> On 06/05/09 11:06, Octave Orgeron wrote:
>> Hi,
>>
>> Well the *given* names from the control domain will not carry over to 
>> the device names in the guest OS, as devices are named after the 
>> device name (in this case vnet and c#d#s#) and enumerated based on the 
>> instance order seen in path_to_inst which is inherited from the OBP. 
> 
> Yes, I noticed that, and made a point to not name them vnet* in the 
> control domain to highlight the change in name.
> 
> 
>> Typically, what I do is this:
>>
>> <guest domain name>-vnet#
>> <guest domain name>-vdsk#
>>
>> This way I keep the vnet and vdisk names in the control domain 
>> readable and organized. As for the guest OS's, if the numeration is 
>> important, add them in the order that will give you the 0,1,etc. 
>> ordering that you'd like.
> 
> So you are confirming what the customer complained about and what I saw. 
> More-so, the order they are added is important. So if they want to 
> maintain vdisks 0, 2, and 4 since that their current application 
> configuration, they will have to keep something around to occupy 1 and 
> 3, and (now I am speculating) if they make a change in the middle, they 
> will have to remove all and add them again in the proper order (this may 
> be their real issue). I will try this out some more--it may be easier 
> with vdisks than vnets, as I could have dummy files with a single
>  directory and file to help show changes.
> 
> The customer is asking for an RFE to keep the ordering in place.
> 

What you want is 6801884 Removing a vdisk can change remaining disk paths in 
guest domain
(This is fixed in the soon-to-be-released LDoms 1.2). This allows you to 
explicitly specify
the instance number and prevents it being changed underneath you)

- Liam




> Thanks
> Steffen
> 
>> It unfortunately does take some additional effort, but this is the 
>> same with physical servers where you have to place cards in the lowest 
>> to highest PCI/PCI-E slots per controller. Brings up fun memories of 
>> Serengeti servers:)
>>
>>  
>> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
>>  
>>
>> Octave J. Orgeron
>> Solaris Virtualization Architect and Consultant
>> Web: http://unixconsole.blogspot.com
>> E-Mail: unixconsole at yahoo.com
>> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
>>  
>>
>>
>>
>>
>> ----- Original Message ----
>> From: Steffen Weiberle <Steffen.Weiberle at Sun.COM>
>> To: ldoms-discuss at opensolaris.org
>> Sent: Friday, June 5, 2009 9:45:56 AM
>> Subject: [ldoms-discuss] ?: how do you handle vdisk/vnet device 
>> numbering changes?
>>
>> Since Solaris 2[.0] and SunOS 5.0 came out, one of the hardware 
>> characteristics of Solaris was the remembering of device numbering. If 
>> you put HBAs or NICs into slots a, b, and c, they were numbers 0, 1, 
>> and 2, or however the bus probing found them. Then, if you removed the 
>> device in slot b and added a different one in slot d, the last one was 
>> device 3, and device 1 would be missing.
>>
>> This has its challenges as across systems in trying to get them to 
>> look the same, while keeping a single system the same as hardware is 
>> added and removed.
>>
>> With LDoms, this does not happen. Lets say I add to a guest 
>> mynet1 at vsw-primary, mynet2 at vsw-primary, and mynet3 at vswprimary, they 
>> turn out to be vnet0, vnet1, and vnet2. If I drop mynet2 at vsw-primary, 
>> and add  mynet4 at vsw-primary, the latter will end up as vnet1.
>>
>> This may not as critical for networks, however, I have heard the same 
>> happens for vdisks.
>>
>> How do you handle this situation? A customer is struggling with this 
>> and is keeping all intermediate vdisks in place even if they are 
>> 'empty', just to keep the vdisk ordering consistent over time.
>>
>> Is there a CR or RFE to make the configuration more static?
>>
>> Thanks
>> Steffen
>> _______________________________________________
>> ldoms-discuss mailing list
>> ldoms-discuss at opensolaris.org
>> http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
>>
>>
>>
>>       
> 
> _______________________________________________
> ldoms-discuss mailing list
> ldoms-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss


Reply via email to