>From Sahara (and Hadoop workload in general) use-case the reason we used
BDD was a complete absence of any overhead on compute resources
The results show that the LVM+Local target perform pretty close to BDD in
synthetic tests. It's a good sign for LVM. It actually shows that most of
the storage virtualization overhead is not caused by LVM partitions and
drivers themselves but rather by the iSCSI daemons.
So I would still like to have the ability to attach partitions locally
bypassing the iSCSI to guarantee 2 things:
* Make sure that lio processes do not compete for CPU and RAM with VMs
running on the same host.
* Make sure that CPU intensive VMs (or whatever else is running nearby) are
not blocking the storage.
I understand that BDD is not really following the trends in Cinder and
OpenStack general approach to virtualization. So deprecating it in favor of
LVM base solution make total sense to me. However there may be other
consumers besides Sahara that rely on BDD so it would be grate to hear
their opinion as well.
On Mon, Sep 19, 2016 at 9:01 PM, Ivan Kolodyazhny <e...@e0ne.info> wrote:
> + [sahara] because they are primary consumer of the BDD.
> Thanks for the answer. My comments are inline.
> Ivan Kolodyazhny,
> On Mon, Sep 19, 2016 at 4:41 PM, John Griffith <john.griffi...@gmail.com>
>> On Mon, Sep 19, 2016 at 4:43 AM, Ivan Kolodyazhny <e...@e0ne.info> wrote:
>>> Hi team,
>>> We did some performance tests  for LVM and BDD drivers. All tests
>>> were executed on real hardware with OpenStack Mitaka release.
>>> Unfortunately, we didn't have enough time to execute all tests and compare
>>> results. We used Sahara/Hadoop cluster with TestDFSIO and others tests.
>>> All tests were executed on the same hardware and OpenStack release. Only
>>> difference were in cinder.conf to enable needed backend and/or target
>>> Tests were executed on following configurations:
>>> - LVM +TGT target
>>> - LVM+LocalTarget: PoC based on  spec
>>> - LVM+LIO
>>> - Block Device Driver.
>>> Feel free to ask question if any about our testing infrastructure,
>>> environment, etc.
>>>  https://docs.google.com/spreadsheets/d/1qS_ClylqdbtbrVSvwbbD
>>>  https://review.openstack.org/#/c/247880/
>>> Ivan Kolodyazhny,
>>> OpenStack Development Mailing List (not for usage questions)
>>> Unsubscribe: openstack-dev-requ...@lists.op
>>> Thanks Ivan, so I'd like to propose we (the Cinder team) discuss a few
>> things (again):
>> 1. Deprecate the BDD driver
>> Based on the data here LVM+LIO the delta in performance (with the
>> exception of the Terravalidate run against 3TB) doesn't seem significant
>> enough to warrant maintaining an additional driver that has only a subset
>> of features implemented. It would be good to understand why that
>> particular test has such a significant peformance gap.
> What about Local Target? Does it make sense to implement it instead BDD?
>> 2. Consider getting buy off to move the default implementation to use the
>> LIO driver and consider deprecating the TGT driver
> +1. Let's bring this topic for the next weekly meeting.
>> I realize this probably isn't a sufficient enough data set to make those
>> two decisions but I think it's at least enough to have a more informed
>> discussion this time.
>> OpenStack Development Mailing List (not for usage questions)
>> Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscrib
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
OpenStack Development Mailing List (not for usage questions)