4 Beryllium

Ask the Expert: Multicore Optimization (MCx) in the New VNX

YOU MAY ALSO BE INTERESTED ON THESE ATE EVENTS...

Ask the Expert: Block Deduplication and FAST Suite Tiering in the New VNX

Ask the Expert: How to Optimize Your VNX

https://community.emc.com/thread/179748

This Ask the Expert event will address your questions about the multicore optimization (MCx) in the next-generation VNX.

MCx distributes VNX data services across all cores, ensuring that cache management and back end RAID management processes take full advantage of multicore CPUs and allowing cache and back-end processing software to scale in a linear fashion. When coupled with FLASH 1st, the VNX with MCx delivers the performance of flash with the efficiency of tiering, providing:

  • Full advantage of the low latency of flash
  • Progressively lower $/IOPS at scale
  • The lowest $/GB for inactive data

Your hosts:

profile-image-display.jspa?imageID=8368&size=350Dan Cummins is EMC's Director of Engineering for MCx.
profile-image-display.jspa?imageID=8385&size=350Phil Trasatti is EMC's Director of Performance Engineering for the Unified Storage Division. He has been a senior engineering manager, consulting software engineer, and architect for products in the networking, storage and video markets, and is experienced in all aspects of product delivery, software development, product deployment and field support.
profile-image-display.jspa?imageID=8282&size=350Andrew Maffessanti was hired by EMC straight out of university in 2006 and has been working on the CLARiiON and now the VNX Product line, specializing in array performance. He has a BA in Information Science & Technology from Mercer University.


This discussion will be moderated through September 16. Follow the thread in your inbox (see option on right) to receive updates.

Reply
48 Replies
Highlighted
4 Beryllium

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

This discussion is now open. We look forward to a fun, interactive and informative discussion.

Regards,

Mark

Reply
Highlighted
7 Thorium

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

dear experts, with so much power at your disposal now ..why dedupe was not implemented as an inline process but rather a post process job.

Reply
Highlighted
1 Copper

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Dedupe is new feature for us and we are excited to bring it to the market.  MCx is a breakthrough as we have literally swapped the engine out of the VNX to enable us to better utilize all the processing power available to us.  Both VNX and VNXe have this new base capability and our components are benefiting from it.  The implementation choice for block dedupe is around time to market and we will continue to improve our offering.

Reply
Highlighted
2 Iron

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

First of all: MCx is great. Second: MCx was inevitable. The old architecture simply was a bottleneck.

My most important question is how to size.

I.e. a VNX5600 sweet spot limits to 50x EFD when their IOPS are fully utilized. What will the limit be if 70% read and all IO's are mirrored by MV/S? And what if they are mirrored with MV/A? And What with RecoverPoint/SE with VNX-splitter?

Are there rules of thumb which indicate a % of 'IOPS-headroom' to reserve for replication? And what the impact is with different flavors of replication?

And what about the impact of RAID5/RAID6/RAID10 on SP utilization with MCx?

Also interesting is the Active-Active dual-lun ownership feature. Is there a whitepaper how this is being done and how cache coherency is maintained?

Reply
Highlighted
1 Copper

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Let me get back to you on the sizing questions.  I will address your concerns around the A/A implementation and cache coherency.

There is a whitepaper that is posted on powerlink VNX MCx Multicore Everything and it will be posted on ECN later today.

WRT to impact of RAID algorithms on SP Utilization.  The previous generation of software performed all the RAID functions on a single core and asymmetrically offloaded parity calculations and block checksums.  With MCx each core independently executes the RAID functions, parity calculations, and block checksums.  All cores execute in parallel. 

Core components of MCx have been designed for symmetric active/active access.  This enables the storage processor to make updates without having to redirect a request.  The MCx stack implements shared locking to coordinate exclusive updates between the SPs.  This is done at the Cache and RAID level (MCC and MCR).  This is described in the white paper.

Reply
Highlighted
2 Bronze

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Q1. What are the technical items, that demonstrate how the VNX with MCx delivers the performance of flash with the efficiency of tiering when coupled with FLASH ???

Reply
Highlighted
1 Copper

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Great question! One impressive example is our new VNX-5600 running the Exchange 2013 Jetstress benchmark test suite on Thin LUNs. We measured a 100% improvement in the IOPs / drive between the older VNX-5500 and the new VNX-5600 platforms.

Add a small FAST VP flash tier and the performance per drive jumps by another 50% over a non-flash configuration under the Jetstress workload. This clearly delivers more value and performance while lowering the TCO for VNX customers.

Thanks,

Phil

Reply
Highlighted
7 Thorium

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Philip Trasatti wrote:

Great question! One impressive example is our new VNX-5600 running the Exchange 2013 Jetstress benchmark test suite on Thin LUNs. We measured a 100% improvement in the IOPs / drive between the older VNX-5500 and the new VNX-5600 platforms.

drives count/specs being equal with the only exception this running on 5600 ?

Reply
Highlighted
1 Copper

Re: Ask the Expert: Multicore Optimization (MCx) in the New VNX

Correct, only change was the VNX-5600 was swapped in.

Thanks,

Phil

Reply