Sorry for the late update.

First of all I have to thank the whole VMWare staff and all participants for this awesome time. Almost every part of the event was precisely planned. Especially none of the Breakout-Sessions had any delay or inconvenience.

DAY 3 – General Session, Keynote and Breakout Session

General Session – Innovation Keeps Companies Competitive:

• Cloud strategy – any cloud
• NSX better inclusion in vSphere – pricing etc.
• HTML 5 client 90% features at the moment – next update almost 100%
• Today will be your slowest tech day in your life!
• Quantum computing
• Edge – IoT
• Dell EMC and VMWare independence
  • own ecosystems
  • acceleration, vibrant and divers
• PCI compliance on premise
• VCenter Cloud management and on premise SSO
• VMWare HCX (hybrid cloud experience) high performance cloud interconnect- cloud migration – seamless platforms – security

Keynote – Transforming the datacenter:

• Technologies have to be everywhere and close to the customer
• 500 GB data emergence per scheduled flight
• Smart buildings 250GB up to 1PB per day in the coming years
• Low latency important
• Edge mastery is a competitive advantage
• Fast & secure simultaneous
• Cloud provider speedy implementation but sticky – lock in
• Globally consistent data center with infrastructure as a code

Breakout Sessions:

Performance VCenter Deep Dive:

• Appliance is outperforming VCenter Windows
• VCenter and Database should be in the same net
• Extensions cost Performance
• Operations mainly done on hosts (i.e. ROBO)
• Tags and associations in VCenter DB
• More VCenter – more Heap size in flash
• CPU + Memory for operations < 70% usage at avg!
• VCenter: 640 tasks simultaneously before queuing
• Concurrent tasks limits:
  • Per Host Limit = 16:
    • vMotion costs=2
    • svMotion costs=8
  • Per datastore limit = 128
    • vMotion costs=1
    • svMotion costs=16
  • Network Interface Card limits:
    • 1 Gbit 4 concurrent
    • 10 Gbit 10 concurrent
    • Usage of separate v motion nic or port group recommended

Preventing Public and Private Cloud Lock-In

What is a cloud?

Not just a datacenter

Not just virtualization

Elastic – on demand – resource pooling – measured service – broad network access

• Azure – high level services – not on premise
• AWS – zero redundancy – resiliency – availability zone no guarantee of uptime
• Google Compute – 700 engineers @ security – owns 30% of optic fiber – tracking and recording data by google ML / AI
• VMWare Cloud on AWS:
  • HCX allows mixing different versions of vsphere
  • platform as a service to prevent cloud locking!
• Each solution has its purpose – no silos! infrastructure as a code!
• Cloud choice considerations
  • SLA and availability
  • Data management and governance
  • Ticketing integration
  • CMDB integration
  • Security
• Automation -> Resources for innovation

DAY 4 – General and Breakout Sessions and final conclusion

Breakout Sessions:

vSAN 6.6 a day in a life of an I/O

• No clustered file system its an object store resulting in a low overhead
• Granular availability and performance requirements, objects distributed across cluster based on storage policy

• On flight corruption verification before send to host
• Old blocks get read and scrubbed to avoid data corruption
• Deduplication and Compression
  • only available on all flash systems
• Shot lifespan -> no compression

• Data at rest encryption
  • Incoming buffer and then destaging
  • Data in flight is not encrypted
• Snapshots – no redo logs like VMs – max 31 snapshots

• Intelligent rebuilds – smart, efficient repairs

NVMe – A new drive technology

Why faster storage?

NVMe:

• Streamlined commands
• Direct access to CPU
• Massive parallelism; higher throughput
• Faster media
• Scaleable performance

Maybe replacement of SCSI

 

Allflash – best practice

settings for every all flash storage regardless of vendor

• Mulipathing round robin
  • Default is MRU – which is less than ideal
  • Maximizes performance for array devices by using all available paths
• I/O operations limit
  • How often ESXi switches logical paths for a given device – set to 1
  • 1000 is default
• Queue depth (outstanding I/Os, Concurrent threads, In flight I/Os)
  • Handling things a one
  • High = more I/O
  • If queue depth limit is 32, 32 I/O can be processed at once
  • 33rd has added latency because it has to wait
  • Storage vs. device vs. vscsi vs. virtual disk – queue depth limit
• Changing queue depth requires reboot of host – be aware if new cluster hosts are added -> host profile – settings must be documented
• Paravirtual SCSI adapter – higher performance to a single VM
• Math on IOPS
  • 1 second = 1000ms
  • 1000ms / 0.5 ms per IO = 2000 IOPS
  • 2000 IOPS * 32 outstanding I/O = 64000 IOPS
  • -> I/O in queue change limits
• IOPs limit as policy in vSphere 6.5!

Final conclusion:

A very tech heavy event. Still trying to digest all the information. The complexity of IT is breathtaking and will get more confusing over the next years. Hope I will be able to take part at VMWorld 2018!

 

BR, Nicolas Frey