Hi I wanted to get some expert advice on this. I have an ordinary hardware SAN from Promise Tech that presents the LUNs via iSCSI. I would like to use that if possible with my VMware environment where I run several Solaris / OpenSolaris virtual machines. My question is regarding the virtual disks. 1. Should I create individual iSCSI LUNs and present those to the VMware ESXi host as iSCSI storage, and then create virtual disks from there on each Solaris VM? - or - 2. Should I (assuming this is possible), let the Solaris VM mount the iSCSI LUNs directly (that is, NOT show them as VMware storage but let the VM connect to the iSCSI across the network.) ? Part of the issue is I have no idea if having a hardware RAID 5 or 6 disk set will create a problem if I then create a bunch of virtual disks and then use ZFS to create RAIDZ for the VM to use. Seems like that might be asking for trouble. This environment is completely available to mess with (no data at risk), so I''m willing to try any option you guys would recommend. Thanks! -- Mark -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20110214/29960762/attachment.html>
On Tue, Feb 15, 2011 at 5:47 AM, Mark Creamer <whitetr6 at gmail.com> wrote:> Hi I wanted to get some expert advice on this. I have an ordinary hardware > SAN from Promise Tech that presents the LUNs via iSCSI. I would like to use > that if possible with my VMware environment where I run several Solaris / > OpenSolaris virtual machines. My question is regarding the virtual disks. > > 1. Should I create individual iSCSI LUNs and present those to the VMware > ESXi host as iSCSI storage, and then create virtual disks from there on each > Solaris VM? > > ?- or - > > 2. Should I (assuming this is possible), let the Solaris VM mount the iSCSI > LUNs directly (that is, NOT show them as VMware storage but let the VM > connect to the iSCSI across the network.) ? > > Part of the issue is I have no idea if having a hardware RAID 5 or 6 disk > set will create a problem if I then create a bunch of virtual disks and then > use ZFS to create RAIDZ for the VM to use. Seems like that might be asking > for trouble.The ideal solution would be to present all disks directly as JBOD to solaris without any raid/virtualization (either from the storage of vmware). If you use (1), you''d pretty much given up data integrity check to the lower layer (SAN + ESXi). In this case you''d probably better off simply using stripe on zfs side (there''s not much advantage of using raidz if the block device would reside on the same physical disk in the SAN anyway). If you use (2), you should have the option of exporting each raw disk on the SAN as a LUN to solaris, and you can create mirror/raidz from it. However this setup is more complicated (e.g. need to setup the SAN in a specific way, which it may or may not be capable of), plus there''s a performance overhead from vmware virtual network. Personally I''d chose (1), and use zfs simply for it''s snapshot/clone/compression capability, not for its data integrity check. -- Fajar
> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss- > bounces at opensolaris.org] On Behalf Of Mark Creamer > > 1. Should I create individual iSCSI LUNs and present those to the VMware > ESXi host as iSCSI storage, and then create virtual disks from there oneach> Solaris VM? > > ?- or - > > 2. Should I (assuming this is possible), let the Solaris VM mount theiSCSI> LUNs directly (that is, NOT show them as VMware storage but let the VM > connect to the iSCSI across the network.) ?If you do #1 you''ll have a layer of vmware in between your guest machine and the storage. This will add a little overhead and possibly reduce performance slightly. If you do #2 you won''t have access to snapshot features in vmware. Personally I would recommend using #2 and rely on ZFS snapshots instead of vmware snapshots. But maybe you have a good reason for using vmware snapshots... I don''t want to make assumptions.> Part of the issue is I have no idea if having a hardware RAID 5 or 6 diskset will> create a problem if I then create a bunch of virtual disks and then useZFS to> create RAIDZ for the VM to use. Seems like that might be asking fortrouble. Where is there any hardware raid5 or raid6 in this system? Whenever possible, you want to allow ZFS to manage the raid... configure the hardware to just pass-thru single disk jbod to the guest... Because when ZFS detects disk errors, if ZFS has the redundancy, it can correct them. But if there are disk problems on the hardware raid, the hardware raid will never know about it and it will never be correctable except by luck.
Hi I am no expert, but I have used several virtualisation environments, and I am always in favour of passing iSCSI straight through to the VM. It creates a much more portable system, often able to be booted on a different virtualisation environment, or even on a dedicated server, if you choose at a later date (sometimes takes a little work, but it is easier than the alternatives). For ZFS, I would suggest this is even more useful. One could, theoretically, export a pool from one VM, then easily import it on another, or on a random machine. If you are looking for a solution for this, I would suggest looking at gPXE (http://etherboot.org/wiki/start). It allows booting from iSCSI fairly easily, and they have a guide for booting opensolaris. Just my 2p :) Regards Karl> -----Original Message----- > From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss- > bounces at opensolaris.org] On Behalf Of Edward Ned Harvey > Sent: 14 February 2011 23:26 > To: ''Mark Creamer''; zfs-discuss at opensolaris.org > Subject: Re: [zfs-discuss] ZFS and Virtual Disks > > > From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss- > > bounces at opensolaris.org] On Behalf Of Mark Creamer > > > > 1. Should I create individual iSCSI LUNs and present those to the VMware > > ESXi host as iSCSI storage, and then create virtual disks from there on > each > > Solaris VM? > > > > ?- or - > > > > 2. Should I (assuming this is possible), let the Solaris VM mount the > iSCSI > > LUNs directly (that is, NOT show them as VMware storage but let the VM > > connect to the iSCSI across the network.) ? > > If you do #1 you''ll have a layer of vmware in between your guest machine > and > the storage. This will add a little overhead and possibly reduce > performance slightly. > > If you do #2 you won''t have access to snapshot features in vmware. > Personally I would recommend using #2 and rely on ZFS snapshots instead of > vmware snapshots. But maybe you have a good reason for using vmware > snapshots... I don''t want to make assumptions. > > > > Part of the issue is I have no idea if having a hardware RAID 5 or 6 > disk > set will > > create a problem if I then create a bunch of virtual disks and then use > ZFS to > > create RAIDZ for the VM to use. Seems like that might be asking for > trouble. > > Where is there any hardware raid5 or raid6 in this system? Whenever > possible, you want to allow ZFS to manage the raid... configure the > hardware to just pass-thru single disk jbod to the guest... Because when > ZFS detects disk errors, if ZFS has the redundancy, it can correct them. > But if there are disk problems on the hardware raid, the hardware raid > will > never know about it and it will never be correctable except by luck. > > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss