[Discuss] Thin Provisioned LVM

[markw at mohawksoft.com (markw at mohawksoft.com)]

>> From: Discuss [mailto:discuss-bounces+blu=nedharvey.com at blu.org] On
>> Behalf Of markw at mohawksoft.com
>>
>> says give ZFS whole disks, which is stupid.
>
> Mark, clearly you know nothing about ZFS.

Think what you wish. Maybe I'm not explaining the problem....

Commercial SAN systems provide disks as "LUNs" over fibre channel or
iSCSI. These LUNs are allocated from a pool of disks in a commercial
storage system. Ideally, a number of servers would use storage from the
SAN. Each of the servers or VMs will be presented with their disks.

Now, EXT2, XFS and many other file systems keep their data allocation
conservative, opting to re-use blocks in-place instead of using new
blocks.

The problem arises when you have something like a 100 VMs, each with a 2TB
LUNs, running off a SAN with only 20TB of actual storage. Without ZFS, the
systems only use space as they need it. 100VMs with 2TB of logical storage
each, can easily come out of 20TB as long as block allocation is
conservative. When you use ZFS the 100VMs will, far more quickly than
actually needed, gobble up 2TB each and force 200TB physical storage even
though most of the VMs have largely free space used by ZFS.

This is representative of a *real* and actual problem seen in the field by
a real customer. ZFS is not compatible with this strategy, and this
strategy is common and not something the VERY LARGE customer is willing to
change.

>
> Also, it's clear you have an axe to grind, which makes anything you say
> about it "take it with a grain of salt."

Believe what you will, I have posted nothing but real issues that myself
and other people have had.

>
> I've personally used a lot of zfs, and a lot of lvm, and there is barely
> any situation that I would ever consider using lvm ever again.

Agreed, ZFS does a lot of things "right," unfortunately it does a lot of
things incorrectly and renders itself as a sub-optimal for a class of
applications, specifically ones which manage their own block cache and
block I/O strategy.

You can make ZFS "faster," but in the configuration I describe, not as
fast as a simpler volume management system.
>