Disclaimer: I recently attended Storage Field Day 13.  My flights, accommodation and other expenses were paid for by Tech Field Day and Pure Storage. 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.

These are my rough notes from a session I attended on “Day 0” of Pure//Accelerate 2017 (aka Storage Field Day Exclusive at Pure//Accelerate 2017). Videos of the session can be found here and you can grab my raw notes from here. I try to avoid dumping a bunch of dot points in Tech Field Day posts, but as this one covered some key announcements, I thought most of the information was useful presented as is.

A Year of FlashBlade

Par Botes spoke to us briefly about the progress made with FlashBlade in the past year. Originally internally codenamed “Wedding Cake” as it was a white box, the performance is “better than you think”, going from 500K IOPS and 15GB/s, to 1.2M IOPS and 15GB/s, to 1.5M IOPS and 16GB/s in the first six months since GA. I first encountered FlashBlade in the flesh at Storage Field Day 10. You can read more about that here.

Scaling Beyond 15 Blades

Rob Lee took some time to talk about scaling beyond 15 blades in a chassis.

• Linear capacity scale – single namespace growing to dozens of PB scale
• Linear IOPS and throughput – single namespace / IP scales IOPS and throughout with capacity
• Preserve simplicity – more capacity adds IOPS & throughout with zero administration

Logical View

• Fabric – scale raw bandwidth without adding management
• Processing – software dynamically schedules processing resources globally
• Data – place data as a single system across all blades

High level Architecture

• Integrated Networking – combined internal and external networks. Load balance connections across all blades.
• Distributed control – partition and distribute control of namespace, data, and metadata across all blades
• Distributed Data – Distribute persistent data across all blades – high-frequency transactions in NVRAM and longer-lived data in N+2 erasure-coded flash

[image courtesy of Pure Storage]

FlashBlade Data Distribution

• Wide-stripe erasure coding with +2 redundancy
• Scaling past a single chassis
• External network load balancer, inter-chassis network switching
• Added external Fabric Module
• intra-chassis network switching
• External Flash Module is 2 rackmounted switches

Scaling Fabric Bandwidth

32port 100Gbs switch (1.6Tb/s north-south). Here’s a photo of Rob talking about this.

• Controller load balancing and capacity load balancing
• East-west traffic
• NVRAM/Flash data access
• Metadata coordination
• 1.6Tbs across chassis, 300Gbs within chassis

Control Placement

Adding a blade is straightforward

• Partitions rebalance to new blade – stops running on old blade and boots on new blade
• No data movement required, only compute (data stays in-place)
• Partition load balancing on a per-blade basis – not chassis constrained

Data Placement

• Data erasure coded across n+2 RAID stripes
• 15 blades – 13-wide stripe (11+2 parity shards)
• RAID Groups are dynamic – selected as needed
• RAID Groups can cross chassis boundaries

As you fill the chassis, it becomes beneficial to constrain the RAID group to a chassis. Note also that there’s enhanced resiliency (n+2 per chassis, without additional overhead) and reduced inter-chassis bandwidth requirements for rebuild operations.

Takeaways

• Software creates parallelism and scale; hardware enables access to data
• Software/hardware integration without tight coupling
• Simplicity/reliability created by software control of the network fabric

Native Objects

Brian Gold presented a section of the session on Why Objects?

Next Generation Apps

• Cloud-native development
• Rich metadata databases

Performance

• Large & streaming: AI training, media serving, analytics
• Small & random: time-series metrics, real-time streams

Efficiency

• No visible partitions
• Unified management

Classic Object Gateways

• Object API gateway -> file system (index to track metadata)
• File system becomes bottle neck when scaling to billions of object
• Purity (FlashArray and FlashBlade) – objects at the core

Object Read Path

Request Arrival

• Extract bucket and object names from request
• Decode bucket and object names
• Get bucket ID from authority
• Get object ID from bucket authority
• Forward read request to object authority

Read data

• Read object data from flash
• Forward back to protocol handler
• Decompress and form response to client

Two takeaways

• Two phases – metadata lookup and data access; distributed everything
• Basically identical to how a file is read via NFS

FlashBlade is S3-compatible for the moment. Purity is really a key-value database

Looking Forward

The next generation of applications require new storage interfaces. There was a demo using TensorFlow.

• Converting raw pixels (ultimate in unstructured data) to structured data
• Now imagine if you’ve got 10s of thousands of cameras
• Object detection -> message queue -> object indexing, streaming queries, time-series analysis

Conclusion

Par wrapped up by talking about:

• “The big bang of intelligence”
• Modern Compute – parallel architecture driving performance
• New Algorithms – modern approaches for superhuman accuracy
• Big Data – Data is the new oil
• “Massively parallel is the new normal”
• 4th Industrial Revolution (2010 – now) – AI, Big Data, Cloud, IoT, Computing, digital to intelligence

I was a bit confused by FlashBlade when I first heard about it, and suggested that the 12 months post Storage Field Day 10 would be critical to the success of the product. Pure have managed to blow me away with the progress they’ve made with the product since GA, the breadth of customers and use cases they’ve lined up, and the overall level of forward thinking that’s gone into the product. You can use it to do some really cool stuff. The biggest problem I’ve had with the “data is the new oil” paradigm is that, unlike real oil, a lot of companies don’t actually know what to do with their data. FlashBlade is not going to magically fix this for you, but it’s going to give you some pretty compelling infrastructure that solves some of the problem of how to do stuff effectively with massive amounts of data.

Object storage is the new hot, and has been for a little while. Putting together a product like FlashBlade has certainly gotten Pure into a bunch of accounts where they weren’t traditionally successful with FlashArray. It’s also given their more traditional customers a different option for tackling big data problems. Pure strike me as being fiendishly focused on delivering something special with FlashBlade, and certainly don’t appear to be slowing down when adding new features to the platform. There’s been some really cool features added, including support for 17TB blades (almost by accident) and increasing scalability to 75 blades. I’m looking forward to seeing what’s next for FlashBlade. You can read the blog post about the FlashBlade 2.0 announcement here.

Disclaimer: I recently attended Storage Field Day 13.  My flights, accommodation and other expenses were paid for by Tech Field Day and Pure Storage. 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.

These are my rough notes from a session I attended on “Day 0” of Pure//Accelerate 2017 (aka Storage Field Day Exclusive at Pure//Accelerate 2017). Videos of the session can be found here and you can grab my raw notes from here. I try to avoid dumping a bunch of dot points in Tech Field Day posts, but as this one covered some key announcements, I thought most of the information was useful presented as is.

ActiveCluster

Tabriz Holtz and Larry Touchette took us through an overview of ActiveCluster.

What do customers really want?

• Disaster protection
• Consistent performance
• Transparent failover
• Ease of management
• Subscription to innovation

They don’t want

• Life to be difficult because they’re running synchronous replication
• To pay for it

Multi-site Active/Active

• Zero RPO
• Zero RTO
• Zero $
• Zero Additional Hardware

Basic Architecture

• Symmetric Active/Active
• Pure’s pod management model – container where you can store your volumes
• Passive Pure1 Cloud Mediator – prevent split brain

Pods

• Simple management model
• only 1 new command introduced
• serves as a container and a consistency group
• keeps metadata with its data

4 Steps to Setup

The whole point is that it’s super simple to setup, so much so that you can do it in four steps from the CLI.

1. Connect the arrays

purearray connect --type sync-replication

2. Create a stretched pod

purepod create pod1
purepod add --array arrayB pod1

3. Create a volume

purevol create --size 1T pod1::vol1

4. Connect your hosts

purehost create --preferred-array arrayA host
pure host connect --vol pod1::vol1 host

Symmetric Active/Active

I/Os perform symmetrically

• 1 round trip for writes
• reads serviced locally

Host ALUA preferences:

• Active/Optimised
• Active/Non-optimised

There’s a 5ms RTT limit and it uses TCP/IP between arrays (Ethernet). Independent dedupe runs on both sides.

Passive Mediator

• No split brain … ever!
• Intelligence is in the arrays
• Mediator imply records failover
• No third site needed
• Arrays alert if they can’t access mediator

There is also the option to deploy a VM that can be used on-premises. While the cloud mediator runs multiple instances behind a load balancer, the on-premises mediator would have to be protected with HA or similar.

So what if I lose comms to the outside world? (both to the outside world and the partner array). Volumes will be taken offline. The mediator is a per pod setting, so you could conceivably use both in your environment.

Transparent Recovery

1. Snapshots sent asynchronously until arrays are nearly in sync

2. IOs forwarded synchronously along with final snap are merged into target

3. Arrays are fully in sync with no pause in IO for final sync

The goal is to allow different arrays to replicate with each other. At GA these will be qualified. Purity versions (1 minor version different e.g. 5.1 and 5.2). Every customer gets this feature (yay, Evergreen). Assuming you have the appropriate supporting infrastructure. And there’s support for multiple connection types.

DirectFlash Shelf

Pete Kirkpatrick (Chief Hardware Architect) spoke about the recently announced FlashArray//X.

• 100% NVMe Enterprise AFA
• “This is just a flash array”
• The data is the array, the hardware and software comes and goes over time

They were working on the FlashArray//M chassis about 4 years ago, and had gone to some length to future proof the design. “DirectFlash Modules” are now replacing the SAS SSDs. SSDs emulate HDDs – this isn’t ideal.

Flash Transition Layer needs Garbage Collection

• Severely limits sustained throughout
• Destroys latency distribution
• Causes excess wear
• Needs over provisioned capacity

Legacy protocols and interfaces

• Assumed high latency
• Inherently serialised

DirectFlash

Purity has always been designed for Flash, so get disk legacy out of the way. The goal of DirectFlash was to

• Start with NVMe: efficient, low latency, high throughput, high parallelism,
• Design an API providing knowledge of the Flash geometry
• Place data intelligently, and schedule operations with high precision
• No FTL is required, so GC is eliminated
• High sustained throughput and low, deterministic latency
• No over provisioning
• Extended flash endurance

Here’s a happy snap of one of the 18.3TB (I think) modules.

You can now fit 1PB of useable Flash in 3RU. DirectFlash Shelf is this week’s news. They’re using NVMe/F. It’s over RoCE (RDMA over Converged Ethernet). You can start with 1 shelf at this stage, but Pure are looking to extend that capability.

VVols Support

Cody Hosterman took us through VMware Virtual Volumes (VVols) support with Purity//FA 5.0. I enjoy watching Cody present and I wasn’t disappointed by this session. So, VVols eh? Why?

• Virtual disk granularity on array – use array-based technology on a virtual disk basis
• Automatic volume creation and configuration
• Storage Policy Based Provisioning

Virtual Volumes – The Full Picture

[image via Pure Storage]

VVols

Every VM has individual volumes on the array:

• Config VVol—4 GB—holds the configuration information of the VM. Created automatically when a VM is provisioned
• Swap VVol—is for the VM swap file. Sized according to the VM memory. Created automatically when the VM is powered-on and deleted when powered-off
• Data VVol—for every virtual disk added to the virtual machine there is a new data VVol. Sized by the requested size of the virtual disk

VVol Snapshots

• VMware snapshot and array-snapshot is created automatically – no performance penalty

The Data Plane

Protocol Endpoints

• A mount-point for VVols
• Presented in a traditional fashion via iSCSI or FC as LUN
• VVols are sub-LUNs to a PE. IO goes to the PE on the array, the array distributes.
• FlashArray automatically “binds” VVols to the appropriate PE

The Management Plane

• How does vCenter manage the FlashArray? Through a VASA provider

FlashArray VASA Provider

• VASA version 3 (includes replication)
• Redundant service on both controllers
• Automatically configured during Purity upgrade
• Active-active configuration
• Entirely stateless – no configuration is tied to /stored on the hardware of the controllers (no special VASA database on the array, it is part of the array configuration)

Policy Based Management

Create VMs and individual virtual disks (VVols) independently

Use customised capabilities advertised by VASA provider to configure volumes

• Replication
• Snapshot policy
• QoS
• Etc.

Are VVols Special Volumes? On the FlashArray, not really. Just normal volumes with special metadata tags. So if you want to present a VVol to a physical server for instance? You can just connect it as a standard LUN.

Conclusion

I’m enthusiastic as all get out about ActiveCluster. There have been rumblings in the market about this type of capability for some time, so it’s great to see Pure deliver. I need to dig a bit deeper into it, but it feels a lot like it has the cross-site capability of Dell EMC’s VPLEX without a lot of the palaver traditionally associated with that product (which to be fair, does more than just cross-site volumes). The great thing is that it’s available to existing customers without additional expense (at least on the Pure side) or messing about. This has always been a big selling point for Pure, and it’s great to see it continue here.

I think the DirectFlash shelf is certainly a step in the right direction and the appetite is there for this kind of solution. I’ll be interested to see how many shelves they end up adding, as the scale and speed possibilities here are potentially pretty tremendous. It will also be interesting to see the uptake of the solution over the next 12 months.

I liked a lot of what I saw with Cody’s presentation on VVols support. It certainly appeared fairly straightforward. I remain underwhelmed by VVols in general though. I know there needed to be a change in the way we presented storage to VMs but it feels like we’ve somehow missed the boat with this solution. In another year it might just be that everything sits on VVols by default but I feel like that’s been the feeling for the past five years and it hasn’t yet transformed to the extent we expected. I am more than happy to be proven wrong on this point though, and my surliness regarding VVols shouldn’t be taken as criticism of what Pure have managed to deliver here. Also, it’s worth checking out Cody’s post on Virtual Volumes support here – he covers it way better than I do.

Disclaimer: I recently attended Storage Field Day 13.  My flights, accommodation and other expenses were paid for by Tech Field Day and Pure Storage. 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 General Session notes from Day 1 of Pure//Accelerate 2017. I’ll be digging into some of the announcements in the very near future so this is just a rough teaser. I call it death by dot point.

David Hatfield

David Hatfield in a Dubs cap. Sorry about the traffic, but it can be hard “[w]hen you setup an event for 3000 people in kind of a crack area?.” He then talks about Game 5 of the NBA Finals. Thanks to people for coming a long way (some travelled around 36 hours). 227800 square feet of space. Used for manufacture of iron originally. This will be the last event ever in this building. Will be knocked down soon. There’s also a 110 feet wide screen, 420 million pixels, FlashBlade helps to render.

“New and Possible”

• 25 new software capabilities, new hardware, new cloud capabilities, new partners (friendships?)
• “The new possible” – Help you break free from legacy way of doing things.
• New – look around and identify what real innovation looks like.

Partner and sponsor shoutout

• It takes an ecosystem of partners to help too, and sponsors as well.
• Shoutout to Veeam – will be directly integrating Veeam and Pure Storage in next release of Availability Suite (v10?)
• Cisco – FlashStack (over 1400 customers together). 7 CVDs in place at the moment. 2 new offers – NVMe over Fabric to the host, Cisco Capital – full offering for FlashStack

Scott Dietzen

Scott Dietzen (CEO) takes the stage, and reclaims his Dubs cap. Am I in the right place? I knew I was when I got inside. Blue is the colour of so many of our competitors, but it’s a Warriors cap, so it’s okay.

This is our second //Accelerate conference. Thanks to 3300 customers (and partners).

“The world’s most valuable resource is data”

• Companies are competing to amass huge datasets (are they doing useful things with it though?)
• AI rates only behind cloud and mobile in terms of impact people think it will have
• “Massively parallel AI demands massively parallel storage”

“By the year 2020 the amount of data created will be 50+ zettabytes”

• Capacity of the Internet will only be 2.5 zettabytes
• It’s going to stored close to where it’s generated

A new model for DCs

• Multi-cloud
• Core
• Edge

Pure uses tonnes of SaaS to run its business, but it also has its own DCs. Believes Edge is going to be larger than multi-cloud and core together because of IoT, etc.

Disruptors? (based on a recent survey)

• The digital gold rush
• The great workload debate
• Cloud: migrations and mistakes
• Data: taking back control

Multi-cloud = photo of Snoop Dogg blazing. Pure are looking to deliver “The data platform for the cloud era”.

Differentiators?

• Big data bandwidth
• Performance for deep learning
• ultra-density
• Uptime
• Subscription to innovation

Tomorrow’s cloud block:

• 1300 cores
• 2.6PB Flash
• 2x density and 5x performance improvement (for leading SaaS company)
• 100x reduction in space for enterprise DC (20 racks -> 4RU)

$1B revenue and cash flow positive, 6th year in selling

All the new software updates are given to customers for free

Liz Centoni

Liz Centoni (Cisco SVP and GM, Computing Systems Product Group) takes the stage to talk with Scott Dietzen.

Every DC was built to do one thing: run applications. But these applications are changing – how they’re built, where they reside, etc. They’re a lot more distributed. A lot more endpoints to manage and secure.

How does FlashStack help?

• Helps customers build their private cloud infrastructure
• This year – adding hybrid cloud (via Cisco CloudCenter)

FlightStats example

Servers existed before DCs or cloud. Customers want any workload, anywhere

• Compute is moving closer to the data.
• Security is top of mind for everyone.
• UCS is a “system” – fabric-centric design, 100% programmable, 60K customers.
• Start small and scale

What about the impact of NVMe?

• SSDs changed the storage bottleneck, but NVMe really puts it back on the network. Cisco is happy about having the opportunity to improve the performance of the whole stack.

Kelli Zielinski

Kelli Zielinski, Domino’s takes the stage. Traditional arrays just weren’t keeping up. Invested in Pure Storage FlashStack. Now “[t]he application gets what it needs”.

Matt Kixmoeller

Matt Kixmoeller takes the stage. It’s the “dawn of a new cloud era – yet the old never really disappears”.

Purity

• Reduce
• Assure
• Protect
• Secure
• Direct Flash

Pure1

• Manage
• Analyse
• Support
• Meta

FlashBlade for big data, FlashArray//X for low latency/high IOPS apps

It’s time for software to take centre stage – 25 new features (delivered in an Evergreen fashion)

• The new Tier 1
• Big Intelligence
• Self-driving Storage ()

Tier 1

• “no one ever got fired for buying [blue]”.
• You want reliability and innovation (dedupe, compression, simple, automated, open cloud integration, NVMe, NVMe/F)
• Delivered 2 years of 6 9s since GA

Metro Stretch Cluster – 1994 – EMC launched SRDF – 20 years later – still hard and expensive.

Purity//FA 5.0

Purity ActiveCluster.

Steve Hodgson (Software Architect) does a brief demo on setup.

Jason Nadeau

Jason Nadeau takes the stage.

• Compression 2.0 – 25% improvement – self-selecting compression engines
• Simplest VVols implementation in the industry
• Granular VM-level operations and transparency
• Cloud automation and security compliance

Purity//FA Snap and CloudSnap

“Portable snapshot”

• Snap Local
• Snap to FlashArray
• Snap to FlashBlade
• Snap to NFS
• CloudSnap to AWS
• DeltaSnap to API

So now you can:

• Native, two-way cloud connection
• Backup, restore, migrate, and DR
• Fully leverage all PaaS services

Purity //Run

• Run VMs and Containers Directly on Purity
• Ideal for Edge Analytics, Custom protocols
• Flexible, open, secure, HA platform

Windows File Services for Purity

• Best of Breed – FlashArray meets MS File services

DirectFlash Shelf

• Native NVMe/F expansion shelf (photo)

“The new Tier 1 is Evergreen”

Arthur Riel

Arthur Riel (Director, The World Bank) takes the stage. The World Bank is neither Wall Street nor Main Street. Their job is to decrease poverty. When he got there, they were a risk-averse organisation (“Let’s keep doing what we’re doing because it works”).

• But “[i]f I save money and can’t deliver my services what good is that?”.
• “Slow storage covers up a lot of sins up the stack”

Sumit Dhawan

Sumit Dhawan (runs EUC at VMware) talks with Scott Dietzen.

“Tech is going out of tech” you need a more platform-centric approach (?)

Dietzen: Is it hard to work with us when your parent company is Dell? (I’m paraphrasing).

Par Botes and Rob Lee

Par Botes and Rob Lee take the stage to talk about FlashBlade.

• “The big bang of intelligence”
• Medium blade – 17TB (fits in between 8TB and 52TB)
• 75 blade-scale FlashBlade (start as small as 7 blades, scale 1 at a time)

Why Object Storage?

• Cloud-native applications use object storage
• Next-generation developers code with object
• Cloud primary storage is object
• >10x faster time to first byte vs S3
• >100x faster indexing image objects vs existing solution at leading web scale company

Brian Gold joins them on stage – it’s about AI from edge to cloud

Rob Ober

Rob Ober (Tesla Chief Platform Architect, Nvidia) takes the stage with Scott Dietzen.

Deep learning (about 5 years ago) has taken off because of:

• Neural nets (this have been around a while)
• Massive amounts of data (tremendous volumes)
• Computation

“You need good data”

Sandeep Singh

Sandeep Singh on stage to talk about self-driving storage

• Automate and simplify
• Sense and model world around
• Constantly learn and re-train
• Global effect

>7PB telemetry data

> 1 trillion data points per day

Pure1

>500 Sev1 incidents avoided to date

Performance sizing has been the final frontier

• Too many variables
• Complex interaction and inter-dependencies
• Over provisioning = wasted expense
• under-provisioning = downtime

This is a perfect problem for AI and machine learning.

Pure1 Meta

• Global sensor network
• real-time scanning
• data lake
• AI engine

Sergey Zhuravlev, Chief Data Scientist

• >1000 measures
• “workload DNA”
• Meta learns from everyone’s workloads to make better predictions
• Meta Workload Planner

David Hatfield takes the stage to wrap up. Here’s a summary of today’s announcements.

Purity//FA 5.0

• ActiveCluster
• Snap to FlashBlade
• Windows File Services for FlashArray (SMB and NFS)
• Policy QoS
• Snap to NFS
• Compression 2.0
• VVols
• Hybrid Cloud for AWS
• Microsoft ODX
• CloudSnap to AWS
• Purity /Run
• Docker Persistent Volumes

Purity//FB 2.0

• Object / S3
• SMB
• Snapshots
• Scale to 75 Blades
• HTTP
• REST
• LDAP
• IPv6
• NFS NLM

Pure1

• Pure1 Meta
• Workload Planner
• Cloud Mediator
• Global Dashboard
• Workload DNA
• Cloud REST

Good session. 4 stars. Stephen did a nice live blog as well – you can read it here.