At Wed, 3 Aug 2011 09:56:46 +0000,
Fernando Frediani (Qube) wrote:
> 
> Thanks Kazutaka, make sense.
> 
> So as long I keep my nodes either with the similar logical partition size or 
> same individual disk sizes should be fine.
> 
> Does running more individual sheepdog daemons use much extra memory or isn't 
> something to concern about really ?

Sheepdog doesn't use much memory, so I think running many sheep
daemons on one server is not a problem.  If it causes a problem,
please report on this mailing list. :)

Thanks,

Kazutaka

> 
> Fernando
> 
> -----Original Message-----
> From: MORITA Kazutaka [mailto:[email protected]] 
> Sent: 03 August 2011 09:59
> To: Fernando Frediani (Qube)
> Cc: '[email protected]'
> Subject: Re: [Sheepdog] Sheepdog cluster expansion
> 
> At Mon, 1 Aug 2011 14:46:20 +0000,
> Fernando Frediani (Qube) wrote:
> > 
> > Hi Kazutaka,
> > 
> > So if eventually I add a new node which doesn't have the same size as the 
> > others (say 5TB) the maximum data to be stored will be 8.3TB per node ? If 
> > so than I risk my 5TB node run out of space ? Or 5TB will be the limit of 
> > data that I will be able to store on each of my 10TB nodes ?
> > I mean in short what happens or what are my limits with a 3 nodes 
> > configuration 10TB+10TB+5TB.
> 
> The 5 TB disk will be the limit.  The data will be distributed uniformly, so 
> in the above case, the maximum data size you can store would be about 15 TB 
> (each disk has about 5 TB).
> 
> > 
> > I guess I workaround for this is run a sheepdog process for each disk, but 
> > that would be fairly more complex and eventually use more Ram memory ? 
> > Thinking about a server with 24 disks.
> > What you suggest in a scenario like this ?
> 
> Running a sheepdog daemon for each disk also had a problem; to handle node 
> failure, replicated data shouldn't be stored to disks on the same node.
> 
> I've just sent a patch to support location-aware data placement.  This patch 
> can ensure that replicated data is stored on different machines.
> With this patch, I think it would work fine to run multiple sheep daemons on 
> each server.
> 
> 
> Thanks,
> 
> Kazutaka
> 
> > 
> > For the last question I think it is somehow related to the first one. If 
> > eventually I add a node with enough disks for the OS+KVM and some small 
> > storage(say 2 disks, so 500GB) just to run the run the sheepdog process I 
> > would run in a worst scenario as described below ?
> > 
> > Thanks
> > 
> > Fernando
> > 
> > -----Original Message-----
> > From: MORITA Kazutaka [mailto:[email protected]]
> > Sent: 01 August 2011 13:44
> > To: Fernando Frediani (Qube)
> > Cc: '[email protected]'
> > Subject: Re: [Sheepdog] Sheepdog cluster expansion
> > 
> > At Mon, 1 Aug 2011 08:44:15 +0000,
> > Fernando Frediani (Qube) wrote:
> > > 
> > > Hi,
> > > 
> > > If I start my cluster with the following scenario: 2 Servers with 10TB 
> > > each (one single logic drive presented by the Raid controller) and 
> > > obviously copies=2 how should I grow it ?
> > > 
> > > Should I always add a 2 x Servers of 10TB every time I need to increase 
> > > my storage or can I just do one by one and it will always balance the 
> > > data between any number of servers available ?
> > 
> > You can add one by one.  The data will be balanced automatically without 
> > any configuration.
> > 
> > > 
> > > If I can add one by one does these servers need always to be the exactly 
> > > same size of others (10TB) or there is some flexibility on having a 
> > > mixture of that and again the things will balance fine ?
> > 
> > Sheepdog assumes that all the disks have the same free space, so the data 
> > will be balanced uniformly even if there is a much larger disk.
> > It is not impossible to handle different sized disks, but it is one of the 
> > future works.
> > 
> > > 
> > > In another slightly different scenario if I have already enough storage 
> > > for my KVM Virtual Machines and all I need is just CPU and Memory is it 
> > > OK jus to add a node with that but without storage and make it use the 
> > > sheepdog for its VMs as well ?
> > 
> > Basically, Sheepdog needs a local disk to use for a storage system.
> > Though your VMs can connect to the other machine from out side of Sheepdog, 
> > in that case the connected server (gateway) could be a single point of 
> > failure.  The connection failover is also one of TODO items.
> > 
> > Thanks,
> > 
> > Kazutaka
> > --
> > sheepdog mailing list
> > [email protected]
> > http://lists.wpkg.org/mailman/listinfo/sheepdog
> -- 
> sheepdog mailing list
> [email protected]
> http://lists.wpkg.org/mailman/listinfo/sheepdog
-- 
sheepdog mailing list
[email protected]
http://lists.wpkg.org/mailman/listinfo/sheepdog

Reply via email to