HPE Storage Users Group
https://3parug.com/

3PAR OS 3.3.1 MU1 is released
https://3parug.com/viewtopic.php?f=18&t=2462
Page 12 of 19

Author:  mitchellm3 [ Fri Nov 03, 2017 7:29 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

HPE confirmed they won't do any upgrades to MU1 P17. They are on hold until P18 is out. I believe this will all be rolled into an EMU1. It is supposed to be out by the middle of November. They have no problems upgrading to 3.3.1 EGA with P16...just no MU1.

On another note. Our they showed concern that I wanted to upgrade my 7440's to 3.3.1. I told them I needed greater than 16T volumes. They were concerned that I wanted to use the new dedupe on all my volumes which they say the 7k isn't the best at. Even though the new dedupe is better...it just causes real high CPU on the nodes.

Author:  Patrick [ Sat Nov 04, 2017 12:50 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

mitchellm3 wrote:
On another note. Our they showed concern that I wanted to upgrade my 7440's to 3.3.1. I told them I needed greater than 16T volumes. They were concerned that I wanted to use the new dedupe on all my volumes which they say the 7k isn't the best at. Even though the new dedupe is better...it just causes real high CPU on the nodes.


Any idea how high they are worrying about? I've been sitting at 15-20% utilization on a couple of 7000 series arrays.

Author:  Chris212 [ Mon Nov 06, 2017 2:58 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

mitchellm3 wrote:
They were concerned that I wanted to use the new dedupe on all my volumes which they say the 7k isn't the best at. Even though the new dedupe is better...it just causes real high CPU on the nodes.


Dedupe is an ASICS process it should not have any impact on CPU. Compression is a CPU process which on our 8200 causes >70% CPU :?

Author:  MammaGutt [ Mon Nov 06, 2017 3:25 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

Chris212 wrote:
Dedupe is an ASICS process it should not have any impact on CPU.


If that is the case, why does the All-flash models (xx50) have beefier CPUs in addition to memory? This is both true for the 8k/20k and the 7k which doesn't even support compression. My understanding is that the dedupe and dedupe checking etc happens in the ASICs but a lot of other processes related to dedupe doesn't (GC to mention one).

Author:  bajorgensen [ Mon Nov 06, 2017 11:18 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

"Dedupe is an ASICS process it should not have any impact on CPU."

Nonsense, only checksumming is done in ASIC and 3PAR does an active read/compare (in ASIC?)
so not only is dedupe high cpu, but generates additional IO as well.
Dont believe marketing, that is the first thing you should learn in IT.

I have also experienced very high CPU usage on the low end controllers,
disable dedupe and it is down to more manageable levels.

There is also an issue on v1 and v2 with dedupe TPVV where the hash table runs very large.
This is mitigated in v3 on 3PAR OS 3.3.1 but I have not tested yet.
Required migration into new CPG though...

Author:  JohnMH [ Tue Nov 07, 2017 5:03 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

The reason more cores and more memory are beneficial is that dedupe generates a lot of metadata and the fastest way to access that metadata is if it resides in memory. The lower the available memory the more paging to/from SSD has to occur and the more backend I/O is generated. Similarly the ASIC based hashing and dedupe is only the first part of the process, in order to free space in a dedupe store you need garbage collection and if you want that to occur regularly then you need idle compute to patrol and sweep that, hence more cores are better when you want sustained performance. There's a good reason why the all flash arrays have more grunt and that's not only related to all out performance.

3PAR isn't unique in this it affects all dedupe enabled systems, although in many the processes are hidden to a certain extent since you have no comparative measures e.g. you only have a single store so no ability to change, reclaim or move space and there's typically no way to turn off the process to compare performance with and without. The fact that some systems defer the process under heavy load should tell you it's not a free process despite the marketing.

Author:  oby [ Mon Nov 27, 2017 10:31 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

P18 fixes the tunesys issues.

Author:  mitchellm3 [ Wed Nov 29, 2017 11:05 pm ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

EMU1 is approved and should be posted soon.

Author:  storsnapper [ Thu Nov 30, 2017 10:48 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

mitchellm3 wrote:
EMU1 is approved and should be posted soon.


I was told by HPE engineer today that EMU1 went GA on monday. They are applying it to my new array and we are back in queue for upgrading the other. Of course that's assuming it doesn't get pulled between now and when its scheduled.

Author:  Patrick [ Thu Nov 30, 2017 11:34 am ]
Post subject:  Re: 3PAR OS 3.3.1 MU1 is released

mitchellm3 wrote:
EMU1 is approved and should be posted soon.

Thanks for the update

Page 12 of 19 All times are UTC - 5 hours
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/