Question about vmfs in active/active?

D

danswartz

2[H]4U
Joined
Feb 25, 2011
Messages
3,715
I've got two hosts with local disks replicated synchronously using drbd (hypervisor is proxmox). It's working fine. I want to try this with vsphere5. I would be running the drbd storage appliances (ubuntu/zfs) and exporting the storage to vsphere via iSCSI. The idea is that each host would have it's local appliance as the primary iCSI target and the other host's as backup. My understanding is that vmfs should be able to work this way, since it is cluster capable, no? I have the vnic in the ubuntu guest as vmxnet3 and have confirmed that it's negotiating '10gb' speed. I created a datastore on the local target (I'm not yet using MPIO, since the other host is still running proxmox), and installed a debian guest on it. That guest is now running dbench and is clocking about 150MB/sec. Does this sound reasonable? Thanks...
 
The appliances make it appear to be the same datastore on failover, so yes, that's fine - a lot of places have various ways of doing that.

Throughput is meaningless for a benchmark - IOPS are what matter.
 
Thanks for the sanity check. To clarify about BW: this was more of a sanity check for me, since I've discovered setups that absolutely sucked (e.g. getting 20MB/sec sustained I/O...)
 
red hat clustering services ... or whatever ubuntu calls them may be worth looking into. then you can float two IPs between your ubuntu hosts and in theory the connection never drops.
 
That's a thought too. Dunno how much they would charge though. Probably not cheap :(
 
all the pieces that make up red hat clustering are freely available. heartbeat, pacemaker, etc
 
Ah, okay. I'd just need to find a HOWTO to 'roll my own'? It uses GFS, I guess...
 
You must log in or register to reply here.
Back
Top