Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

4220

July 30th, 2014 04:00

Virtual Provisioning(Best Performance) in Vmax

We have a VM infrastructure where we are supposed to provide luns of large sizes ranging from 1.2TB - 2TB.

Now my question is:

  • How many hypers should i choose to form my Meta Device? Which option would provide me greater performance(eg:  30 x 69GB or 16 x 128GB or 8 x 240GB or any other option )?
  • Also i would like to understand how this configuration can be depicted, keeping performance in mind ?

I am trying to understand whenever some one asks me to provide a lun size of 1.2 TB or 1.5 TB or 2 TB, how would i select my options (since symmetrix provides me with a range of options to form meta devices) for the best performance.
Also if some one can help me out with any white papers which explains best practice for virtual provisioning.

1.3K Posts

August 1st, 2014 07:00

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2072070

Before adding paths or moving VMs, I would change this parameter if it is not already.

If you still have issues, I think going to 6 or 8 paths would be a better option than splitting the servers.  To me it would be easier to manage.

1.3K Posts

July 30th, 2014 08:00

In Symmetrix, hypers or splits are terms usually used to refer to segments on a physical disk drive.  With VP, your meta members are not associated directly with the backend configuration.

In general, you get the maximum FA performance from a given LUN with about 32 meta members.  16 in most cases is enough.

https://support.emc.com/docu31003_FAST_VP_for_Symmetrix_VMAX_Theory_and_Best_Practices_for_Planning_and_Performance.pdf?language=en_US

My best practices for VMAX from EMC World

Event Launch

July 30th, 2014 09:00

is it better to create a stripped meta if performance is our concern ? i.e if we have no plans to expand the dev .

if performance is not my priority , can I simply create stripped metas for everything ?

1.3K Posts

July 30th, 2014 09:00

Yes!  If there is no need for expansion, always do striped metas. 

See my best practices presentation at about 48 minutes into it for the meta discussion.

1.3K Posts

July 30th, 2014 10:00

The link I posted should have opened the session, but in a pop-up.

Tech sessions, sort by most popular or most liked, and it should be the first one.

July 30th, 2014 10:00

What tab am I looking for ?

Tech Sessions

Federation

Build-A-Hybrid-Cloud

Area 52

Area 53

Demos & 360

I can get into the site but not sure what video to watch .

July 30th, 2014 12:00

Thanks , got it . Watching it now .

2 Intern

 • 

1.2K Posts

July 30th, 2014 18:00

32 Posts

July 31st, 2014 01:00

Thanks a lot Quincy. That was an excellent presentation.     

However i still need to confirm something, if you can help me out with that.

1> How exactly are the "cache slots" booked for thin devices?
2> How to choose how many FAs to allocate for a particular host. What are the things that we should keep in mind?

2 Intern

 • 

1.2K Posts

July 31st, 2014 02:00

The general best practice of FA configuration is "go wide before go deep". We always hope to balance FA workloads, 4 FA ports are better than 2 for sure, with better performance and better fault tolerance. For example, if you have 2 busy FA ports. it could be bottleneck, so better to configure 4 or more.

Besides, PowerPath is designed for balance and failover.

2 Intern

 • 

1.2K Posts

July 31st, 2014 02:00

2> How to choose how many FAs to allocate for a particular host. What are the things that we should keep in mind?

Symmetrix has CPU's dedicated to Front-End ports. Using two ports on the same FA CPU does not give twice as much performance, as the same CPU has to serve both ports. There are several recommendations for FA configuration:

  • Use all the 0 (zero) ports on a director first and then the 1 (one) ports.
  • Do not zone the host to the same port of the same director board.
  • A given host should never be mapped to both ports of one FA CPU.
  • They should also never be only connected to one director board for redundancy.
  • Spread connections across engines / directors first, then on the same director.
  • You get increased availability if you spread the connections "wide" first.
  • Configure multiple paths to FA ports on separate processor slices and directors.
  • Each host should be configured to a minimum of two FA's (SAFs), 4 is recommended.
  • Connect at least two HBAs across redundant fabrics for high availability.

Hope this helps

32 Posts

July 31st, 2014 02:00

Does IOPs also come into play here?

Like if 1 of my host is doing very less IOPs then is it fine to just allocate them to FAs across engines, for eg : 7e0 and 10e0 ? and if its doing heavy IOPS we scale them out to different directors like 7e0 8e0 9e0 10e0 for better performance? Sort of balancing the load on all of the directors?

32 Posts

July 31st, 2014 03:00

I am more concerned with IOps.
In which "precise cases" should we configure with more than 4? Any Particular HOST conditions?

July 31st, 2014 06:00

I watched the presentation last night , was a good one and a wide range of topics covered, very well done  . I've recommended it to my colleagues as well .

Cheers

1.3K Posts

July 31st, 2014 07:00

Cache slots are not pre-allocated for devices.  Any one device could use 100% of cache for read data if it was the only active device.  For writes, there is a maximum WP limit for the system of somewhere between 60-80% depending on some parameters in your system.  A symcfg list -v will show the number of cache slots and the maximum that can be WP at any time.   For a single Symmetrix device, the WP limit is 5% of the system maximum WP limit. 

For the number of FAs needed, 2 should be the minimum for fault tolerance.  For performance you may need more.  As you add FA CPUs to your host, you will get near linear scaling in IOPs.  Many customers have standardized on 4 FA CPUs for most hosts.

No Events found!

Top