Our Dell MD3660i arrays want by default for me to set up DDP (disk pools), rather than disk groups.
I understand how to set up multipathing with disk groups and ESXi 5.1, but I am unsure about disk pools (DDP). There is little documentation on this as DDP arrived in a relatively new firmware release.
Has anyone integrated shared storage on a Dell MD3 array set up with DDP with ESXi 5.1?
DDP is very appealing as it has fast rebuild times, lots of spindles (IOPS) and great coverage on disk failure.
ok, I nave completed setting up a test environment with DDP and it seems to work well with multipathing.
The setup is
So two subnets across two controllers.
The controllers are attached to two dedicated switches, one for each subnet.
Create vSwitch1 with one vmkernel 192.168.130.103 with a 1:1 ratio to one vnic
Create vSwitch2 with one vmkernel 192.168.131.103 with a 1:1 ratio to the other vnic
Plug the nics into the relevant switch for each subnet.
Add the iSCSI SW adapter. DO NOT port bind the vmks as you should not do this with multiple subnets.
Add to the dynamic discover tab of the iSCSI SW adapter -
Create a disk pool (DDP) with a Virtual Disk.
Create a host group with a host with the iqn of the ESX iSCSI SW adapter.
Map the Virtual disk to the host (LUN mapping)
Enable ALUA ( Asymmetric Logical Unit Access) on the CLI of the ESX host (ssh)
esxcli storage nmp satp rule add -s VMW_SATP_ALUA -V DELL -M MD36xxi -c tpgs_on;
esxcli storage nmp satp set --default-psp VMW_PSP_RR --satp VMW_SATP_ALUA;
Rescan the iSCSI SW adapter and there is your LUN!
I tried various multipath/ failover scenarios by unplugging cables, changing preferred controllers etc and all worked perfectly.
As far as I can tell the configuration looks ok.
I only used disk groups so far but I there shouldn't be a difference in multipathing compared to a disk pool.
So if the MD is properly dicovered as VMW_SATP_ALUA and the right PSP has been asigned you should be ok.
Did you also rebooted a controller and a switch during your failover tests?
Thanks for the reply.
I have not tried rebooting the controller and switch yet. I have tried moving the preferred controller.
What I can see so far from the ESXi perspective is 4 active paths, two with IO only to the preferred controller. If I reallocate the preferred controller the Active IO changes to the other two paths in ESXi. This seems correct.
If I create a situation where there is only a path to the non preferred controller by unplugging cables, then I get Active IO through the non preferred controller. This seems correct too as the ALUA stuff I think takes the path of greater latency in order to continue IO as the controllers do not know to change preferred controller.
DDP whilst cool, new, offering fast rebuild times and giving great multiple disk failure protection, is slower for streaming reads and random writes than RAID 6.
Dells implemetation is also relatively untested in the real world and so hard to find others using it for comparison. I consider this a large risk factor.
Perhaps RAID 10 is a better option. It gives much better performance than RAID 5,6,DDP, and minimal rebuild times. The only downside is the very small chance of two disks in the same mirrored pair failing at the same time. This is VERY unlikely.
DDP is certainly a safer option in this respect, but you could argue that RAID 10 can deal with half the disks failing as long as they are not in the same mirrored pair!
I always think that performance is key on a VMware storage network. If I commit all my 42 2TB drives to a DDP pool and performance is lacking I am stuck. If I RAID 10 all 42 drives then I know performance will never be an issue due to RAID type (maybe just segment size or something else annoying like that!).
The path configuration and failover behavior is looking good.
Personally I prefer performance and usually go with RAID10 to meet the I/O requirements. If you still achieve some reasonable capacity with RAID10 (if you can afford it ) I would recommend to stick with such a configuration.