cisco hyperflex anywhere€¦ · & solutions veeam lff ai/ml sap hana additional scale lff...
TRANSCRIPT
Jonathan GorlinHyperFlex Product Management
March 20, 2019
Voice of the Engineer
Hyperconvergence from Core to Edge
Cisco HyperFlex Anywhere
@CiscoHXGuru
Agenda
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public 2
• Market Trends & HyperFlex Momentum
• HyperFlex 3.5 Key Innovations
• HyperFlex Anywhere
• HyperFlex Edge 2-Node Installation Demo
• HyperFlex All NVMe, Data Protection, Kubernetes
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
The Future is theDistributed Datacenter
Core Cloud
Edge
3
“Over the last two years, IT
organizations spent 70% on
running the business”.
“By 2020, over 90% of enterprises
will use multiple cloud services
and platforms”.
“By 2022, more than 50% of enterprise-generated data will be created and processed outside the core data center or cloud”.
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
End-to-End Modernization TodayGame Changing Results
Savings vs. 3-tier Infrastructure180% Savings vs. Public Cloud
Over 3 years251%Management time savings375%
Virtual desktop speed42x More Performance for Mission Critical Apps530%Improvement in ERP
Response Time650% Downtime Reduction790%
3500+ Customers Are Saying:
Virtual Desktop Infrastructure
Server Virtualization
Test and Development
ROBO & Edge
Databases & ERP
4
HyperFlex 3.5 Innovations
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
HyperFlex 3.5: Key ThemesNew Workloads
& SolutionsVeeam LFF
AI/MLSAP Hana
AdditionalScale
LFFHyper-V
Stretched Cluster
Container Ecosystem
Dynamic ProvisioningOpenShift
SecurityDISA STIG
Lockdown modeTech Support Mode
Networking4G VIC/FI10G EdgeMulti-VIC
3rd Party NIC
Citrix Cloud Services
Full Stack Upgrades
HyperFlex 3.5
6
HX AccelerationEngine
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
HyperFlex Acceleration Engine
• Purpose Built & Co-Engineered Software & Hardware
• Recommended for Better TCO but not Mandatory
• Not a single point of failure
• Performance Boost
• CPU Utilization & Working Capacity Improvements
• Other service-offloads in-future : EC, Crypto, Hash etc.
• PCIe card offloading CPU cycles
• Higher Compression, Lower $/GB
• Better VM Density, Lower $/VM
Hx 3.5(2a)
7
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
HyperFlex Acceleration Engine Benefits
8
Max Performance Increase
Lower Latency
Offload runs a higher grade compression algorithm
Savings increase of up to 30%
Average savings increase across tested workloads: 5.6%
Higher Compression
Performance numbers on 8-node HX cluster, using vdbench with large working set size running 70% read, 30% write workload with mixed block sizes
100%
117.9%
SW HW AccelIOPS
100%86.2%
SW HW AccelLatency
18
%
14
%
HyperFlex AnywhereData Platform Release 4.0
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public 10
HyperFlex Anywhere
Remote / Branch CloudCore DC
HyperFlex Edge2-4 Node Ultra-Light
Cisco HyperFlex64 Node Scalability
Platform for MulticloudHybrid Solutions
• 2-4 Edge Node Cluster
• 1 or 10 Gbps Branch Networking Integration
• Kubernetes CSI plugin • Intersight Invisible Cloud Witness for HX Edge
• Intersight Cloud-based remote multisite deployment and upgrades
• Centralized Audit Logging
• Crypto Enhancements
• All NVMe Nodeswith Intel Optane
• DISA STIG Rest API
HyperFlex 4.0 Innovations
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
HyperFlex Edge Full Lifecycle ManagementPowered by Cisco Intersight
Invisible CloudWitness
On-Going ManagementIncluding
Full Stack Upgrades
Multi-Cluster Install
1 Node
2000 + Nodes
.
.
.
Connected TAC Experience
Upgrade
11
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
HyperFlex Edge 4.0 Architecture
1 GbE or
10 GbE
2 Node
Remote Office / Branch Office
• Support for single ToR, dual ToR, or stacked switches• Optimized 10 GbE direct connect with 1GbE switching (2-node only)
3 Node
4 Node
HyperFlex Data Center
DC Clusters
1 GbE or
10 GbE
1 GbE or
10 GbE
PoS
IoT
WAN
✓ Full life cycle management✓ Scale ✓ Network Flexibility ✓ Invisible Cloud Witness
12
HyperFlex Edge ConfigurationsHX220c M5
Hybrid Edge
HXAF220c M5
All Flash Edge
Configurable CPU Single or Dual Socket (10-56 cores)
Single or Dual Socket(10-56 cores)
Configurable Memory 128GB – 3TB 128GB – 3TB
2-Node Usable Capacity Range 3.3TB - 13.2TB 2.6TB - 28.2TB
3-Node Usable Capacity Range 4.9TB - 19.8TB 4.0TB - 42.4TB
4-Node Usable Capacity Range 6.6TB - 26.5TB 5.3TB - 56.5TB
Note: Approximate calculations above are before dedupe & compression. Effective capacity will be higher.
Consult the HX sizer for the latest sizing & design guidance.
• Initial Cluster Size: 2, 3 or 4
• Networking options:1GbE or 10GbE (bring your own switch)
• Replication Factor 2 (RF2)
• Unlimited cluster scaling, powered by Intersight
Most Flexible
Edge Appliance
in the Industry
Hyperflex DC 1Paris
HX-ROBO NYC
HX-ROBO Denver
HX-ROBO CHS
HX-ROBO Beijing
Sign In with Cisco ID
Converged UCS DC 1 SFO
Ship to site, rack, cable, power-up, IP-address 1
3 Install
2 Login remotely & claim new device
4 Manage
Base includes HX install
HX London
Add new HX-DC or Edge
site
• Cloud based remote deployments
• Monitor HX clusters at scale and across hypervisors
• HCL feature validates firmware & drivers
• Enjoy monitoring, telemetry, analytics, policy, orchestration, proactive TAC, HX Cluster management
Cisco HyperFlex Edge 4.0 2-Node Cluster
Cisco Intersight Invisible Cloud Witness Benefits
• No additional license cost, it is included in the Cisco HyperFlex Edge
license subscription.
• No need for a third site or for existing computing, storage, and network
infrastructure
• Cloud-like operations; with the Invisible Cloud Witness, there is
nothing to manage• No user interface to monitor
• No user-based software updates
• No setup, configuration, or backup required
• No scale limitations
• Built-in high availability and fault tolerance
• Security• Real-time updates to the Invisible Cloud Witness service with the latest
security patches
• All communications encrypted using Transport Layer Security (TLS) 1.2
• Use of standard HTTPS port 443; no firewall configuration required
• Built on an efficient, silent protocol stack• No periodic heartbeats sent across the WAN
• No cluster metadata or user data transferred to the Cisco Intersight platform
• Toleration for high latency and lossy WAN connections
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
2-Node ROBO
Site #1
Hypervisor
VM VM
VM VM
…2-Node ROBO
Site #2000
Hypervisor
VM VM
VM VM
Others 2-Node ROBO
WAN
Cisco HyperFlex Edge 4.0 2-Node ROBO
2-Node ROBO
Site #1
Hypervisor
VM VM
VM VM
…2-Node ROBO
Site #2000
Hypervisor
VM VM
VM VM
Central Data Center
for Witness VMs
Hypervisor
Witness
VMSite 1
vCenter
Witness
VM
Witness
VM Site 2000
Witness
VM
Witness
VM
Witness
VM
WAN
Invisible Cloud
Witness
Witness
VM
Witness
VM
Witness
VM
Witness
VM
Witness
VM
- or -
No additional cost!
Cisco HyperFlex Edge 4.0 2-Node ROBO
Reachability tests
are performed
periodically
between nodes.
When specific failure scenarios
occur, Cisco Intersight holds an
election. The node winning the
election continues to operate and
avoids split-brain scenarios.
HX controller requirements • Each HX controller virtual machine needs
to have public Domain Name System
(DNS) resolution capabilities and the
capability to initiate an outbound
TCP/HTTPS Internet connection on port
443 to the Cisco Intersight platform.
• Typically, these requirements are fulfilled
by allowing outbound Network Address
Translation (NAT) operations from the
subnet assigned to the Cisco HyperFlex
management traffic. If direct outbound
connectivity is not available, an HTTPS
proxy can be configured to enable use of
this service.
Designed with no single point of failure in mind, when individual components fail Cisco HyperFlex continues to operate without service disruption.
Cisco HyperFlex and Common Failure Cases
Common points of failure include:
✓ Cisco HyperFlex nodes (power,
hardware, and software failures)
✓ LAN links and switches
✓ WAN links and routers
Important to note:
✓ The capability of Cisco HyperFlex
Edge two-node clusters to tolerate
drive failures matches that of
traditional Cisco HyperFlex clusters
configured with replication factor 2
(RF2).
✓ For detailed failure tolerances and
recovery procedures, see the Cisco
HyperFlex Administration Guide.
Cisco HyperFlex Connect dashboard when the cluster is fully healthy with fault tolerance
Click on the icon to get additional resiliency health information. Shown in next slide.
Resiliency health information showing that the cluster is healthy and can tolerate failures
Case 1: Loss of WAN connectivity to Cisco Intersight platform
WAN
• During the WAN outage, both Cisco
HyperFlex nodes and all workloads
continue to operate uninterrupted.
• Invisible Cloud Witness is required
only when a local failure of a LAN link,
switch, or Cisco HyperFlex node
occurs.
• Additional LAN or node failures cannot
be tolerated. It is a best practice to
design WAN connectivity with reliability
in mind.
• When WAN connectivity is restored,
the device connector automatically
reconnects to help ensure that the
cluster can tolerate a LAN or node
failure.
Note:
Monitor the Invisible Cloud Witness by
using stcli commands or launching Cisco
HyperFlex Connect.
View from Cisco HyperFlex Connect dashboard when connectivity to Cisco Intersight Invisible Cloud Witness is lost
Clicking the information
icon under resiliency health
brings up a window with
more detailed information
Resiliency health information showing that there is no connection to the Cisco Intersight platform
WAN
Case 2: Cisco HyperFlex node failure as a result of a hardware or software failure
• Local reachability test fails, then the
surviving node immediately reaches out
to the Cisco Intersight platform to
obtain permission to continue
operating.
• If the election state is clean for this
cluster, the Cisco Intersight platform will
return a success response, and the
cluster will continue to operate on the
single surviving node.
• The virtual machines that were running
on the failed node will be automatically
restarted on the surviving node by
VMware HA. The virtual machines that
were on the surviving node will
continue to operate without interruption.
• When the failed Cisco HyperFlex node
is recovered, the reachability test
between nodes will succeed, and the
surviving node will relinquish control.
• The restored node is resynchronized
with the surviving node.
• Failures can be tolerated again.
Cisco HyperFlex Connect dashboard during node failure
WAN
Scenario A: Single link loss between nodes
• If a single direct-connect cable (1 Gigabit Ethernet
topology) or a single network uplink (10 Gigabit
Ethernet topology) is severed, all services will
immediately fail over to the secondary link, and no
additional action is required. The running cluster is not
affected.
• After the link is restored, the services that failed over
will fail back to their original failover priority, and
network fault tolerance will be restored.
Scenario B: All links between Cisco HyperFlex nodes
lost
• If both direct-connect cables (1 Gigabit Ethernet
topology) are lost, or if switching infrastructure or both
uplinks from any Cisco HyperFlex node to the switch (10
Gigabit Ethernet topology) are lost, an election process
occurs.
• When the local reachability test fails, both nodes will
immediately reach out to the Cisco Intersight platform.
• The winning node will continue to operate, and the
rejected node will enter a suspended state and any
Cisco HyperFlex datastores will go offline.
• VMware HA can be configured to automatically restart
the virtual machine on the online node.
• After connectivity is restored, the reachability test
between nodes will succeed and the operating node will
clear the election state from the Cisco Intersight platform
so that future failures can be tolerated.
• Next, the previously offline node is resynchronized, and
the cluster will become fully healthy. From this healthy
state, any new failure scenario can now be tolerated.
Case 3: Loss of connectivity between nodes
Cisco Intersight Invisible Cloud Witness Whitepaper
Simple Worldwide Infrastructure Upgrades Powered by Cisco Intersight
With Cisco HX 4.0, one simple wizard enables optimized full stack upgrades. Choose one or many
Edge clusters to upgrade in parallel!*Intersight
Non-Disruptive online upgrades!
HX-ROBO NYC
HX-ROBO Denver
HX-ROBO CHS
HX-ROBO Beijing
HX-ROBOSFO
Upgrade
Embedded Firmware Management
With Cisco HX 4.0, data path firmware is now embedded!
HXDP Storage Stack
EXSi Hypervisor OS
UCS Server Firmware
HXDP Storage Stack
EXSi Hypervisor OS
UCS Server Firmware
HXDP Storage Stack
EXSi Hypervisor OS
UCS Server Firmware
Data Path Firmware (HDD/SSD/NVMe/SAS HBA/Hercules)
Server Component Firmware(BIOS/CIMC/VIC/NIC/GPU)
HXDP Storage Stack
Most important components to HXDP:• Delivered through Intersight &
On-Prem Installers• Hot add of components• Upgrade for HX Edge through
Intersight
Less Frequent Updates• UCSM for FI attached• Intersight offline upgrade for Edge
(post FCS)
Benefits:• Drastic reduction in interop issues• Painless data path firmware upgrades with each HXDP release• Faster & more reliable upgrades, due to online firmware application
• Upgrade multiple clusters in parallel, even if starting versions do not match
• Upgrade logic decoupled from platform, following true CI/CD model
• Provisions for granular upgrades
• Robust, built-in retry mechanisms
• All execution local to cluster
• Telemetry for continual improvement to upgrade services
• Includes HXDP and associated online firmware
• Automatic filtering of compatible versions
• Automated evacuation of hosts without requirement for DRS (2 node clusters)
New Upgrade CapabilitiesUpgrade
SW Packages New FeaturesNY
Denver
San Fran
4.0
FirmwareTelemetryMultiple Clusters
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
10G Edge for both on-prem and Intersightinstallers
Field re-image for Intersight Installer
Additional Intersight Installer Enhancements
Wipe & Redeploy
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Internet Connectivity Required at FCS*
Price Sensitive Deals
Cost over self-contained availability
2 Node Edge
Completely self-contained
Better failover design with higher resource utilization
Additional performance boost and future expandability
3 & 4 Node Edge
*Intersight appliance will be qualified after FCS
WAN
BOM
Demo2-Node ROBO Installation via Cisco Intersight
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
Enterprise Ready NVMe Solution
34
• NVMe drives have higher perf than SAS/SATA SSDs
• Need PCI lanes in the platform to leverage drive perf
• Need low latency, high bandwidth N/W
• Performance without compromising RAS
• Ability to handle hotplug, surprise removal, etc
• Higher perf requires very high endurance write cache
• No compromise on manageability
• Software stack that can leverage higher perf
• Software optimization reqd to N/W and I/O datapath
Performance RASReliability, Availability, Serviceability
Optimized
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
RAS in NVMe SystemHotPlug, Surprise removal, Firmware mgmt, LED mgmt, etc
35
CPU
PCIe
Storage controller
SATA SSD
SATA System
SATA SSD
HX Controller VM
VMD DriverExisting Datapath
RAS Features
RAS Features
CPU
NVMeSSD
NVMe System
NVMeSSD
All NVMe Datapath
VMD
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
Introducing All NVMe HyperFlex
36
Hx 4.0
• All NVMe UCS C220 M5 Servers
• Integrated Fabric Networking
• Ongoing I/O optimizations
• NVMe Optane Cache, NVMe Capacity
• Up to 32TB/Node, 40GB Networking
• Fully UCSM Managed for FW, LED etc.
• Co-Engineered with Intel VMD for Hot-Plug &
surprise removal cases.
• Reliability Availability Serviceability (RAS)
Assurance
• Up-to 50% IOPs Increase
• Further improvements expected with future
SW optimizations
• Higher VM Density, Lower Latency
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
Long Distance Native Replication
HX Data Platform
VM
HX Data Platform
HyperFlex 4.0 Data ProtectionDisaster Recovery – Runbook Automation
VM VM VM VMVM
Support for VMWare Site Recovery Manager (SRM) Powershell based DR Runbooks
Support for VMWare SRM 8.1 & 6.5
VMWare certified Storage Replication Adapter(SRA)• SRA will be listed for download on VMWare site• HyperFlex SRA will use HX Native Replication• All VMs on the SRM protected datastore get replicated
Coexist with HX Native DR workflows• Different datastore for HX native DR
Powershell based cmdlets for DR operations• Hosted on Powershellgallery.com (Microsoft site)
(https://www.powershellgallery.com/packages/Cisco.HXPowerCLI/)• Commandlet- Examples: [Get]/[Add]/[Remove]/[Invoke] - [Protect]/[Failover]/[ReverseProtect] - VM• Will work with HX 3.5 and 4.0.
Auto Generation of Powershell based DR Runbooks scripts• VM load-order in an autogenerated Powershell script• Powershell scripts for Test Recovery, Migration or Unplanned-Failover for user-
selected VMs
vCenter SRM
HX-SRA
vCenterSRM
HX-SRA
Runbook/ Recovery Plan
37
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public
Kubernetes Storage Integration Evolution
Kubernetes FlexVolume• “Out-of-tree” exec based API for external volume plugins,
essentially scripts executed by Kubelet• Enabled 3rd party storage vendors to write storage
integrations• Requires access to root filesystem of nodes and master
machines to install plugin and it’s dependencies, not desirable
Kubernetes Container Storage Interface (CSI)• Enables 3rd party storage vendors to develop out-of-
tree pod deployed on Kubernetes cluster• All code and dependencies self-contained in pod• No need for root filesystem access• Interfaces with standard CSI sidecar (helper) containers
to interact with Kubernetes
Containers & Kubernetes Introduced• Containers are stateless• Little thought about the need for
storage
Kubernetes In-Tree Storage Support• Introduced storage support• Tightly integrated into and dependent on
Kubernetes code• Checked into and shipped with Kubernetes• Fixes/updates required new versions of
Kubernetes
Evolution driven by Kubernetes Storage Special-Interest-Group (SIG) Community
SupportedSinceHX 3.0
Comingin
HX 4.0
1 2 3 4
38
Thank You