Page 2 of 2
Re: Grossly bloated deduplicated target
Posted: Tue Oct 09, 2012 6:23 pm
by dtrounce
Yes, I enabled the experimental deletion support. This is a 2-host test environment for Hyper-V VMs.
Re: Grossly bloated deduplicated target
Posted: Wed Oct 31, 2012 3:38 pm
by Anatoly (staff)
Have you rebooted the system or restarted the starwind service starting form the moment when the device was created?
Have you doublecheck if all the layers are aligned?
Re: Grossly bloated deduplicated target
Posted: Thu Nov 01, 2012 7:36 am
by dtrounce
Yes, several normal reboots on each server. I've no idea what you mean about "layers being aligned".
Certainly, it was never the case that both targets used exactly the same amount of disk space, which surprised me. It was roughly similar. Possibly each target was writing small changes to the other, which then got sent back ad infinitum?
I have removed the targets now to just use non-deduplicated targets. Given the likely performance impact of deduplication, that Starwind licensing TB is done pre-deduplication, and the Starwind licensing cost/TB is much greater than the cost of underlying disk/TB, it seems the real value of deduplication may not be that high. You save a bit of cheap underlying disk, but are likely to get worse performance.
Re: Grossly bloated deduplicated target
Posted: Fri Nov 30, 2012 3:30 pm
by anton (staff)
It's gonna be changed with a new release. Together with a deletion support we'll represent some special enhacements doing actual ACCELERATION of underlying storage.
Making long strory short: deduplication will make underlying storage working ~10 times faster in some particular workloads (primary storage for VMs for example).
dtrounce wrote:Yes, several normal reboots on each server. I've no idea what you mean about "layers being aligned".
Certainly, it was never the case that both targets used exactly the same amount of disk space, which surprised me. It was roughly similar. Possibly each target was writing small changes to the other, which then got sent back ad infinitum?
I have removed the targets now to just use non-deduplicated targets. Given the likely performance impact of deduplication, that Starwind licensing TB is done pre-deduplication, and the Starwind licensing cost/TB is much greater than the cost of underlying disk/TB, it seems the real value of deduplication may not be that high. You save a bit of cheap underlying disk, but are likely to get worse performance.