SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy,...
Transcript of SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy,...
![Page 1: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/1.jpg)
Sudhanshu (Suds) Jain & Adrian Marinescu
SER2413BE
#VMworld #SER2413BE
NVMe: What Is It? An Interface? A Protocol? A New Drive Technology? An Industry Revolution?
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 2: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/2.jpg)
Agenda
#SER2413BU CONFIDENTIAL 2
1 Why do we need faster storage?
2 Where is the holdup?
3 What is NVMe?
4 NVMe in vSphere
5 Future EvolutionVMworld 2017 Content: Not fo
r publication or distri
bution
![Page 3: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/3.jpg)
Rise of the Third Platform
3
Transforming the World of Applications
Highly parallel processing on very large set of data
Need to deliver business value in
real-time
Low latency for mission critical transactions
Deliver risk and opportunity for future
Business Intelligence
Real-time analytics
Transactional Processing
Machine
Learning
Predictive Analytics
Artificial
Intelligence
Hyper-scale Datacenter
$5.3 trillion by 2020: IDC
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 4: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/4.jpg)
Key Infrastructure Challenges
#SER2413BU CONFIDENTIAL 4
EXPLOSIVE GROWTH OF
DATA
NEAR REALTIME
HIGHLY DISTRIBUTED
Towards Data-centric Future!
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 5: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/5.jpg)
Explosive Growth of Data
#SER2413BU CONFIDENTIAL 5
Stressing your storage and network benchmarks
Source: Intel IDF 2016
Cloudera
ZookeeperVMworld 2017 Content: N
ot for publicatio
n or distribution
![Page 6: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/6.jpg)
Business Value of Data Diminish Fast
#SER2413BU CONFIDENTIAL 6
Time = $$$
Source: http://jtonedm.com/2012/11/21/decision-latecy-revisited/
What we do when data is changing Act fast to gain business value
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 7: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/7.jpg)
Highly Distributed
7
Build, deploy and manage interconnected, collaborative workloads
Big Data Cloud-Native App Multi-tier App
Deploy multiple workloads with strong
demand for inter-VM traffic
Optimize data delivery to applications
Adaptive storage tier critical for improved
application performance
Distributed Storage
Virtual
SAN
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 8: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/8.jpg)
Composable Infrastructure
#SER2413BU CONFIDENTIAL 8
Disaggregation of Infrastructure
Source: http://myeventagenda.com/sessions/0B9F4191-1C29-408A-8B61-65D7520025A8/14/5#sessionID=1484
Performance Curev
All infrastructure delivered as a service
Hyper-scalable to keep up with business demands
Resources automatically tuned to application workloads
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 9: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/9.jpg)
Disruption In Making
9
Future is here now
Source: http://www.theregister.co.uk/2016/09/05/wikibon_server_san_takeover/
Technology Trends
CPU densities
continues to
increase
Hi-Density Flash and
NVDIMM reducing cost
of data access
High speed interconnect
can deliver any to any
fast storage
#SER2413BU CONFIDENTIAL
Application demands for data and key technology trends are disrupting the entire SAN & DAS ecosystem!
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 10: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/10.jpg)
Agenda
#SER2413BU CONFIDENTIAL 10
1 Why do we need faster storage?
2 Where is the holdup?
3 What is NVMe?
4 NVMe in vSphere
5 Future EvolutionVMworld 2017 Content: Not fo
r publication or distri
bution
![Page 11: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/11.jpg)
Anatomy of Data AccessWhere Is the Holdup?
11
Source: http://www.snia.org/sites/default/education/tutorials/2012/fall/solid/AnilVasudeva_NVMe_NextGen_SSD%20Interface-r1-nc1.pdf
* - Typical/Worst case considerationFor long period of time, innovation in silicon (processor
and memory) industry far outpaced the disk I/O speedup
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 12: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/12.jpg)
Anatomy of Data Access
#SER2413BU CONFIDENTIAL 12
Let’s Look Closer on the Various Delays
Source: https://www.snia.org/sites/default/files/SDC15_presentations/gen_sessions/JimHandy_The_Long-Term_Future_PrintVersion.pdf
Various Delays?
How does various
Technology perform?
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 13: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/13.jpg)
Agenda
#SER2413BU CONFIDENTIAL 13
1 Why do we need faster storage?
2 Where is the holdup?
3 What is NVMe?
4 NVMe in vSphere
5 Future EvolutionVMworld 2017 Content: Not fo
r publication or distri
bution
![Page 14: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/14.jpg)
Non Volatile Memory Express: A New Protocol
• NVM Express defines a host controller interface and command set for use with non-volatile memory (NVM)
• Key differentiators
– Streamlined Commands; lighter weight I/O Stack
– Direct access to CPU; lower latency
– Massive parallelism; higher throughput
– Faster media
– Scalable performance
• Key Innovations
– http://www.nvmexpress.org
– http://www.snia.org
– https://www.jedec.org/
– http://www.onfi.org
#SER2413BU CONFIDENTIAL 14
NVMe is architected for NVM
Source: http://www.datacenterjournal.com/answer-nvme-questions/
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 15: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/15.jpg)
NVMe: A New Drive Technology
#SER2413BU CONFIDENTIAL 15
NVMe need new controller and faster media to deliver on its promise!
Source: http://www.snia.org/sites/default/files/files2/files2/contrib_articles2014/S4-S7%20SNIA%20Europe%20v3%20Single%20Pages.pdf
The head has to move to the right place and wait for the right block to
come around
Ah, no need to wait for the blocks to come to the right place – picking
blocks when you need them, one at a time
Getting multiple blocks of information all at once! It is fast
HDD SSD NVMe
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 16: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/16.jpg)
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4Q1 Q2 Q3 Q4
NVM Express, Inc. Roadmap
2014 2015 2017 Future direction
NVM
Express
2016
Released NVMe Planned NVMe Specification releases
• Sanitize
• Streams
• Virtualization
NVMe 1.2.1 May’16 NVMe (next)*
• IO Determinism
• Persistent Cntlr Mem Buffer
• Multipathing
NVMe-MI
NVMe
over
Fabrics
NVMe-MI 1.0 Nov’15 NVMe-MI 1.1*
• SES
• NVMe-MI In-band
• Native Enclosure Mgmt
• Transport and protocol
• RDMA binding
NVMe-oF 1.0 May’16 NVMe-oF (next)*
• Enhanced Discovery
• Authentication
• TCP Transport
• Namespace Management
• Controller Memory Buffer
• Host Memory Buffer
• Live Firmware Update
NVMe 1.2 – Nov ‘14
• Out-of-band management
• Device discovery
• Health & temp monitoring
• Firmware Update
* Subject to change
NVMe 1.3 May’17
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 17: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/17.jpg)
#SER2413BU CONFIDENTIAL
NVMe: Revolutionizing the SAN Architecture
17
NVMe Over Fabric
Scalability
IOPS
LatencyBandwidth NVMe
NVMf
The value of data is based on how fast it can
be accessed and processed
Faster storage access enables cost reduction
through consolidation
Independently scale storage and compute
infrastructures
Data resilience – Mitigate the “server stranded
storage” (with DAS storage) problem
Optimized data delivery to applications
Efficiency of shared storage with no
compromise
Unlocking the Reach of Data
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 18: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/18.jpg)
Storage Industry is Pivoting Towards NVMe
18
NVMe Adoption and Considerations
Currently using
Planning to use within 12 months
Not using and no plans
Drivers to NVMe Adoption
https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2017/20170809_S202A_Burgener.pdf
0% 10% 20% 30% 40% 50% 60%
Better scalability to meet ourperformance requirements over
time
Better cost/performance ($/IOPS)
Higher throughput or bandwidth
Increased storage density (TB/U)
Lower latency
48%
PCIe or NVMe Flash
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 19: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/19.jpg)
Market Trends – High Growth Rate for NVMe
#SER2413BU CONFIDENTIAL 19
It is not a coincidence!
Source: snia.org & IDC
NVMe Drive Revenues growing at 47%
CAGR; expected to hit $7.3B by 2020
Overall revenue from NVMe and NVMe
related product and services is many fold
Key factors driving adaption
• Higher capacity drive and lower cost
• Availability of higher endurance flash media
• Migration from SLC to MLC
Overall revenue of flash media has already exceeded HDD; NVMe is the main growth engine
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
2016 2017 2018 2019 2020
Component Revenues: NVMe, HDD, SSD
NVMe HDD SSD
$7.3B
$11.8B
$16.2B
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 20: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/20.jpg)
Agenda
#SER2413BU CONFIDENTIAL 20
1 Why do we need faster storage?
2 Where is the holdup?
3 What is NVMe?
4 NVMe in vSphere
5 Future EvolutionVMworld 2017 Content: Not fo
r publication or distri
bution
![Page 21: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/21.jpg)
vSphere Platform
21
Telco & HPC
Cloud
VSAN
NSX
Ultra-low latency, packet
intensive workloads,
scale-out storage
Multi-tenancy, zero-trust,
Specialized hardware and
customization
Low latency, high-IOPS &
predictable performance
High packet rate, fast packet
manipulation, hardware offloads,
All-IP networks
FOUNDATIONAL TO SDDC
vSphere Platform
Transforming World of Enterprise
Scale, HPC-like and Highly
Distributed
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 22: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/22.jpg)
vSphere Driver Architecture Evolution
22
Ran
do
m, IP
MI
RDMA
RoC
E, iW
AR
P
SW
FCoE
FC
oE
Driv
er
Nativ
e S
AS
/FC
Driv
er
Nativ
e N
IC D
rive
r
Nativ
e G
rap
hic
s D
rive
r
Additional Partner Native Drivers
vmklinux
SA
S/F
C D
rive
r
NIC
Driv
er
SATA
Library
SA
TA
Driv
er
SW FCoE
FC
oE
Driv
er
Feature focus area
VMK APIVMK APIVMK API
Legacy support
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 23: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/23.jpg)
Software-Defined Infrastructure
#SER2413BU CONFIDENTIAL 23
NVMe/Flash is Key Focus Area
• vSphere Flash Use-Cases: (KB 2145210)
• Host swap cache
• Regular Datastore
• vSphere Flash Read Cache (aka Virtual Flash)
• VSphere ESXi Boot Disk
• VSphere ESXi Coredump device
• VSphere ESXi Logging device
• Virtual SAN (VSAN)
• Open Source Effort to build for faster innovation and the broad
ecosystem
• Open ecosystem for certification and support
Flash and NVMe
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 24: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/24.jpg)
vSphere Storage I/O Stack
#SER2413BU CONFIDENTIAL 24
Built for Best of the Breed Fabric Solutions!
Virtual VolumesVirtual SAN VMFS
Monitor
SSD PCIe NVMeDIMM/NVDIMM PCI SSD
LSI PVSCSI vNVMe
HBANICCPU
HDD
LSI Driver PCSCI DriverNative NVMe
Driver
Guest
I/O Driver NVMe
RDMANIC SAS/SCSI
iSCSIFCoE
NFS
Physical Hardware
S
C
S
I
S
T
A
C
K
N
V
M
e
S
T
A
C
K
FC/SAS
NVMeOverFabric
OS Stack
vRDMA
Para Virtualized
RDMA
ESXi
vmxnet3
VDS
NIOCScheduler
Memory AllocatorSIOC
Vmxnet3 Driver
T
C
P
/
I
P
S
T
A
C
K
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 25: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/25.jpg)
NVM Express Evolution and vSphere
25
Journey towards All NVMe Stack
• Queueing Interface• NVM Command Set• Admin Command Set• End- to- end Protection (DIF /DIX )• Security• Physical Region Pages ( PRPs )
2009 2011 20122010 2013 2014
NVMe 1. 0 ReleasedMarch 1 , 2011
NVMeTechnical Work Begins
•Implementation and Reporting Refinements
NVMe 1. 2 ReleasedNovember 3 , 2014
•Name Space Management
•Controller Memory Buffer
• Host Memory Buffer
• Ground Work for NVMe Management
2015
• General Scatter Gather Lists (SGLs)
• Multi-Path I/O & Namespace Sharing• Reservations• Autonomous Power Transitions
During Idle
NVMe 1. 1 ReleasedOctober 11 , 2012
2016
•Defines extension to NVMe, for non PCI
NVMe Over Fabric 1. 0 ReleasedJune 5 , 2016
•Primary focus on RDMA
•Compatible to FC-NVMe (INCITS 540)
• Host Memory Buffer
• Ground Work for NVMe Management
vSphere 5.5
• Introduce first async NVMe driver 1.0e
• Launch IOVP cert program for NVMe
vSphere 6.0
• Introduce first inbox NVMe driver
• Bring broader ecosystem support
vSphere 6.5
• vNVMe• Optimized
performance for NVMe driver
Future Direction
• Multiple name spaces, Queues
• NVMe Over Fabric• End-to-end NVMe
Stack
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 26: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/26.jpg)
NVMe Driver Ecosystem
• Available as part of base ESXi image from vSphere 6.0 onwards
– Faster innovation with async release of VMware NVMe driver
• VMware led vSphere NVMe Open Source Driver project to encourage ecosystem to innovate
– https://github.com/vmware/nvme
• Broad NVMe Ecosystem on VMware NVMe Driver https://www.vmware.com/resources/compatibility/search.php?deviceCategory=io
– Close to 300 third party NVMe devices certified on VMware NVMe driver
• Also available for download (async) VMware ESXi 5.5 nvme 1.2.0.27-4vmw NVMe Driver for PCI Express based Solid-State Drives
#SER2413BU CONFIDENTIAL26
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 27: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/27.jpg)
NVMe Focus @VMware
27
Build Your Infrastructure with Confidence!
• Reduced serialization
• Locality improvements
• vNVMe Adaption layer
• Multiple completion worlds support in
NVMe
Core Stack
• Boot (UEFI)
• Firmware Update
• End-to-end protection
• Deallocate/TRIM/Unmap
• 4K
• SMART, Planned hot-remove
Driver
• NVMe 1.0e spec
• Hot-plug support
• VM orchestration
Virtual Devices
• Optimized stack - Highly parallel
execution for single path local NVMe
devices
• Reach target of 90%+ performance of
device spec
• Performance enhancements
• Extended CLI/UI
• Name space management
• Async event error handling
• Enhance diagnostic logs
• Performance improvements
• Async mode support
• unmap support
vSphere 6.5 Tech Preview
• Next Generation Storage Stack
with ultra-high IOPS
• End-to-end NVMe Stack
• NVMe Multi-pathing
• NVMe Over Fabric
• Multiple fabric option
• SR-IOV
• Sanitize
• I/O Determinism
• Rev the specification
• Parallel execution @backend
• 4K Support
• Scatter-gather support
• Interrupt coalescing
Future Direction
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 28: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/28.jpg)
Key Challenges for NVMe Based Architecture
#SER2413BU CONFIDENTIAL 28
NVMe Require Fresh Approach for Software Stack Designs
Parallel Execution
Elevator / C-Scan forced stack to be serialized
File systems are also not optimized for concurrent
NUMA architectures pose additional challenges for locality
Interrupts distribution and queue selection
IO scheduling
Performance
More processing power needed to saturate NVMe device
IOPS increase by almost order of magnitude
IO overhead reduced by 2-3x
Queuing == increases latencyVMworld 2017 Content: N
ot for publicatio
n or distribution
![Page 29: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/29.jpg)
Application Application Application
File System
IO
Scheduler
Block
storage
Driver
Implie
d s
erializ
ation
SSD
Source: https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2013/20130812_PreConfD_Marks.pdf
Processing
resources
100 KIOPS
Typical Storage Stack
29#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 30: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/30.jpg)
NVMe Focus @VMware
30
Build Your Infrastructure with Confidence!
• Reduced serialization
• Locality improvements
• vNVMe Adaption layer
• Multiple completion worlds support in
NVMe
Core Stack
• Boot (UEFI)
• Firmware Update
• End-to-end protection
• Deallocate/TRIM/Unmap
• 4K
• SMART, Planned hot-remove
Driver
• NVMe 1.0e spec
• Hot-plug support
• VM orchestration
Virtual Devices
• Optimized stack - Highly parallel
execution for single path local NVMe
devices
• Reach target of 90%+ performance of
device spec
• Performance enhancements
• Extended CLI
• Name space management
• Async event error handling
• Enhance diagnostic logs
• Performance improvements
• Async mode support
• unmap support
vSphere 6.5 Tech Preview
• Next Generation Storage Stack
with ultra-high IOPS
• End-to-end NVMe Stack
• NVMe Multi-pathing
• NVMe Over Fabric
• Multiple fabric option
• SR-IOV
• Rev the specification
• Parallel execution @backend
• 4K Support
• Scatter-gather support
• Interrupt coalescing
Future Direction
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 31: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/31.jpg)
ESXi Storage Stack Enhancements for NVMe Performance Boost
#SER2413BU CONFIDENTIAL 31
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
1 2 4 8
Th
rou
gh
pu
t (I
OP
S)
# Workers
• Hardware:
– Intel® Xeon® E5-2687W v3 @3.10GHz (10 cores + HT)
– 64 GB RAM
– NVM Express* 1M IOPS @ 4K Reads
• Software:
– vSphere* 6.0U2 vs. Future prototype
– 1 VM, 8 VCPU, Windows* 2012, 4 VMDK eager-zeroed
– IOMeter:
• 4K seq reads, 64 OIOs per worker, even distribution of workers to VMDK
vSphere today
vSphere prototype
Source: SDC Conference 2016 https://www.snia.org/sites/default/files/SDC/2016/presentations/performance/Chagam_Dutta_Rajaopal_Bridging_the_Gap_Between_NVMe_SSD_Performance_and_Scale_Out_Software.pdf
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 32: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/32.jpg)
vSphere Virtualized I/O Stack
#SER2413BU CONFIDENTIAL 32
vSphere Offers Rich Set of I/O Virtualization
Virtual VolumesVirtual SAN VMFS
Monitor
NVMe SSD PCI SSD
LSI PVSCSI vNVMe
Interconnect like
IB and OPANIC
HDD
NIC Driver PCSCI DriverNative NVMe
Driver
Guest
I/O Driver NVMe
RDMANIC SAS/SCSI
iSCSIFCoE
NFS
Physical Hardware
P
C
I
FC/SAS
NVMeOverFabric
OS Stack
vRDMA
Para Virtualized
RDMA
ESXi
vmxnet3
VDS
NIOCScheduler
Memory AllocatorSIOC
I/O Device Driver
D
i
r
e
c
t
P
a
s
s
-
t
h
r
o
u
g
h
PCIe GPU/GPGPU FPGA
Hardware Accelerations
e.g QAT
Device
Emulation
Para-
Virtualized
Device
S
C
S
I
S
T
A
C
K
N
V
M
e
S
T
A
C
K
T
C
P
/
I
P
S
T
A
C
K
NSX
DatapathvSAN
Datapath
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 33: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/33.jpg)
Virtual NVMe
#SER2413BU CONFIDENTIAL 33
High Performance Guest Block I/O
New in
vSphere
6.5
Feature:
• NVMe 1.0e Device Emulation
• Hot add/remove support
• Integrated with VM orchestration
• Multi-Q support – 16Queues, total 4K depth
Benefits:
• Improved application performance, better
IOPS and latency numbers
• Leverage native NVMe stack in various
Guest OS including Linux and Windows
• Reduced in-guest I/O stack overhead
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 34: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/34.jpg)
Virtual NVMe
#SER2413BU CONFIDENTIAL
Rich Set of Features & Potential!
Implemented
34
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 35: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/35.jpg)
NVMe Focus @VMware
35
Build Your Infrastructure with Confidence!
• Reduced serialization
• Locality improvements
• vNVMe Adaption layer
• Multiple completion worlds support in
NVMe
Core Stack
• Boot (UEFI)
• Firmware Update
• End-to-end protection
• Deallocate/TRIM/Unmap
• 4K
• SMART, Planned hot-remove
Driver
• NVMe 1.0e spec
• Hot-plug support
• VM orchestration
Virtual Devices
• Optimized stack - Highly parallel
execution for single path local NVMe
devices
• Reach target of 90%+ performance of
device spec
• Performance enhancements
• Extended CLI/UI
• Name space management
• Async event error handling
• Enhance diagnostic logs
• Performance improvements
• Async mode support
• unmap support
vSphere 6.5 Tech Preview
• Next Generation Storage Stack
with ultra-high IOPS
• End-to-end NVMe Stack
• NVMe Multi-pathing
• NVMe Over Fabric
• Multiple fabric option
• SR-IOV
• Sanitize
• I/O Determinism
• Rev the specification
• Parallel execution @backend
• 4K Support
• Scatter-gather support
• Interrupt coalescing
Future Direction
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 36: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/36.jpg)
NVMe Focus @VMware
36
Summary
• Reduced serialization
• Locality improvements
• vNVMe Adaption layer
• Multiple completion worlds support in
NVMe
Core Stack
• Boot (UEFI)
• Firmware Update
• End-to-end protection
• Deallocate/TRIM/Unmap
• 4K
• SMART, Planned hot-remove
Driver
• NVMe 1.0e spec
• Hot-plug support
• VM orchestration
Virtual Devices
• Optimized stack - Highly parallel
execution for single path local NVMe
devices
• Reach target of 90%+ performance of
device spec
• Performance enhancements
• Extended CLI/UI
• Name space management
• Async event error handling
• Enhance diagnostic logs
• Performance improvements
• Async mode support
• unmap support
vSphere 6.5 Tech Preview
• Next Generation Storage Stack
with ultra-high IOPS
• End-to-end NVMe Stack
• NVMe Multi-pathing
• NVMe Over Fabric
• Multiple fabric option
• SR-IOV
• Sanitize
• I/O Determinism
• Rev the specification
• Parallel execution @backend
• 4K Support
• Scatter-gather support
• Interrupt coalescing
Future Direction
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 37: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/37.jpg)
Agenda
#SER2413BU CONFIDENTIAL 37
1 Why do we need faster storage?
2 Where is the holdup?
3 What is NVMe?
4 NVMe in vSphere
5 Future EvolutionVMworld 2017 Content: Not fo
r publication or distri
bution
![Page 38: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/38.jpg)
Byte addressableBlock storage
Storage Evolutionary Path & Key Drivers
38
Building the Stack for Future
System Applications
File System
SCSI
Storage
Architecture
SCSI DriverNVMe/NVMeoF
Driver
NVMe
Storage
Architecture
NVDIMM
Storage
Flash evolution in cost, endurance and
capacity
NVMe as a replacement of SCSI
Composable Infrastructure
NVMe Over Fabric as primary way to
access storage
SAN Evolution towards Server Attached
Storage
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 39: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/39.jpg)
Next-Generation Hardware EvolutionNVMe as Caching Tier & Performance Tier
Persistence Tier
Today’s Hardware
NVMeSSD NVDIMM
Evolution of Storage Tier
NVMe
NVDIMM
Caching Tier
NVMe
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 40: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/40.jpg)
NVMe Over Fabric: Demo
40
POC
Next Generation SAN is Here!
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 41: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/41.jpg)
Key Takeaways
41
Foundation to deploy Software Defined Infrastructure
vSphere and its NVMe ecosystem is constantly innovating
vSphere is ready to power your NVMe infrastructure
NVMe is the choice of technology for your storage needs
#SER2413BU CONFIDENTIAL
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 42: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/42.jpg)
VMworld 2017 Content: Not fo
r publication or distri
bution
![Page 43: SER2413BE NVMe: What Is It? An Interface? A Protocol? A or ... · scale-out storage Multi-tenancy, zero-trust, Specialized hardware and customization Low latency, high-IOPS & predictable](https://reader030.fdocuments.in/reader030/viewer/2022041215/5e0423fff3134a5ea2355448/html5/thumbnails/43.jpg)
VMworld 2017 Content: Not fo
r publication or distri
bution