Dell Technologies World 2018 – storage.12 – XtremIO X2: An Architectural Deep Dive Notes

Disclaimer: I recently attended Dell Technologies World 2018.  My flights, accommodation and conference pass were paid for by Dell Technologies via the Press, Analysts and Influencers program. There is no requirement for me to blog about any of the content presented and I am not compensated in any way for my time at the event.  Some materials presented were discussed under NDA and don’t form part of my blog posts, but could influence future discussions.

Here are my rough notes from the storage.12 session. This was presented by Rami Katz and Zvi Schneider, and covered XtremIO X2: An Architectural Deep Dive.

XtremIO X2

Efficiency

  • 4x better rack density
  • 1/3 effective price $/GB
  • Multi-dimensional scaling
  • 25% DRR enhancements

Protection

  • NVRAM
  • Expanded iCDM capabilities
  • QoS
  • Metadata-aware replication

Performance

  • 80% lower application latency
  • 2x better copy operations
  • Hardware

Simplicity

  • Simple HTML5 GUI
  • Intelligent reporting and troubleshooting
  • Guided workflows

 

Software-driven architecture driving efficiency and performance

Brute force approach limits enhancements (eg faster chips, more cores). With this approach you can average 20 – 30% performance improvement every year to 18 months. You need to have software innovation.

 

XtremIO Content Addressable Storage (CAS) Architecture

The ability to move data quickly and efficiently using metadata indexing to reduce physical data movement within the array, between XtremIO arrays or between XtremIO and other arrays / the cloud (not in this release).

 

XtremIO In-line, all-the-time data services

  • Thin provisioning – all volumes thin; optimised for data saving
  • Deduplication – inline – block written once, no post-process
  • Compression – inline – compressed blocks only, no post-process
  • XtremIO Virtual Copies – super efficient – in-memory metadata copy
  • D@RE – Always on encryption with no performance impact
  • XtremIO Data Protection – Single “RAID Model” double parity w/ 84% useable

 

XtremIO Virtual Copies (XVC)

  • Performance – read, write and latency as volume
  • Space efficient – no metadata bloat, no space reservation, no moving blocks
  • In-memory – instant creation, immediate deletion, flexible topology, unrestricted refresh
  • Flash optimised – identical data services, always on, always inline

 

Efficiency

  • X2: All the goodness at 1/3 the GBu price of X1
  • X1: High WPD SSD
  • X2: Low WPD (write per day), 4x denser DAE = greater economies of scale

In X1 as you grow capacity, $/GB is the same, in X2 it gets less as capacity increases

 

Better Controller Cost Amortisation

  • X1 – 40TB in 6RU
  • X2 – 138TB in 4RU
  • X2 (Future) – 230TB in 4RU

 

Multi-dimensional Scaling, Hardware Platform and Improved scaling granularity 

  • X2 – up to 72 drives, start with 18, go up in increments of 6 (138TB RAW per Brick). 4 Bricks of 550TB (effective 2.7PB assuming 6:1 data reduction). Scale up in X2 Platform.
  • Resilience – 2 simultaneous SSD failures per XDP group (36 drives)
  • Using an efficient compression algorithm – Intelligent packing algorithm, systems typically experience ~2-3:1 compression ratio

SQL example – even with native compression enabled get some additional array compression

 

Performance

  • 80% lower application latency
  • 2x better copy operations

Install base – over 65% of write Its are small <= 8KB

 

Write Flow – X1

Flow overview:

  • Write arrives
  • Find compute module
  • Calculate hash – read old data if needed
  • Send to data module
  • Harden data
  • Acknowledge host

 

Write Flow – X2 with Write Boost

Flow overview

  • Write arrives
  • Find compute module
  • Harden data
  • Acknowledge host

Huge latency improvement, mainly for small IOs

 

Write Flow – X2 De-Stage

  • Write transactions are aggregated, improving efficiency and bandwidth
  • Benefit from IO folding

 

Hardware and Models

X-Brick: X2S 7.2TB – 28.8TB – Cluster Building Block

  • 48 Haswell cores
  • 768GB RAM
  • 8 Host ports (2 16Gbps FC and 2 10Gbps iSCSI per controller)
  • Up to 72 SSDs

 

X-Brick: X2-R 34.5TB – 138.2TB – Cluster Building Block

  • 48 Haswell cores
  • 2TB RAM
  • 8 Host ports (2 16Gbps FC and 2 10Gbps iSCSI per controller)
  • Up to 72 SSDs

 

New X-Brick Configuration

Configuration X-Brick Minimum (RAW) X-Brick Maximum (RAW) Cluster Size in X-Bricks
X2-S 7.2TB 28.8TB Up to 4
X2-T 34.5TB 69.1TB 1*
X2-R 34.5TB 138.2TB 4 (current)

8 (future)

X2-T – If you think you’ll get to X2-R, it would be more cost effective to go to that straight away

 

Protection

NVRAM – People didn’t like the BBU, so they got that sorted for the X2.

 

NVRAM – Components Failure – Method of Operation

 

 

Xpanded iCDM Capabilities

  • 60% of all XtremIO Volumes are Copies – a large number of XVC copies are writeable
  • 10% of clusters > 500 writable copies
  • 5% of clusters > 1000 writable copies

Other iCDM Statistics

  • Average # of copies has doubled in X2 vs X1
  • 90% of XtremIO iCDM clusters use writable copies
  • Max 6700 writable copies per cluster
  • Average writable copies per volume -5 copies
  • Average copies per volume – 12 copies

2x the number of XVC copies

 

Metadata-aware Replication

A superior method for efficient data replication

  • Unique blocks only
  • Globally unique (not just at the volume / consistency group level)
  • Changes only
  • Compressed

 

Global Dedupe = Global Savings

Replication Use Cases

Unified view for local and remote protection

Easy Operation, Best Protection, Superior Performance

You can read more about XtremIO Native Replication here.

 

Simplicity

Redesigned User Interface

  • Simple and intuitive
  • 1-2-3 provisioning
  • Tagging and search
  • HTML5 (no Java)
    • Nothing to install
    • Popular browser support

 

Flexible provisioning flows – next step suggestions

 

New Reports

  • Weekly peaks
  • Latency heat map
  • Block distribution

“X2 evolves every frontier that X1 redefined”

 

Futures

Better iCDM with QoS

  • Max IOPS or Bandwidth for every volume or consistency group
    • Protect workloads based on importance e.g. critical applications and multi-tenant environments
  • Burst mode gracefully handles applications that temporarily exceed max IOPS

 

Q&A

Is synchronous replication on the roadmap? Give us a little time. It’s not coming this year. You could use VPLEX in the interim.

How about CloudIQ? CloudIQ support is coming in September

What about X1? It’s going end of sale for new systems. You can still expand clusters. Not sure about any buyback programs. You can keep X1 for years though. We give a 7 year flash guarantee.

X2 is sticking with InfiniBand and SAS, why not NVMe? Right now it’s expensive. We have it running in the labs. We’re getting improvements in software. Remember X2 came out 6 months ago. Can’t really talk too much more.

 

Solid session. 3.5 stars.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.