Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014

1
© 2014 Scalar Decisions Inc. Not for distribution outside of intended audience

}  Introduction about Scalar Decisions
}  VDI Challenges
}  Storage for VDI
}  Vendor Solutions
}  Q & A

2

Introduction

3

Founded in 2004
$105M in FY13
Revenues

Nationwide Presence

120 Employees
Nationwide

25% Growth YoY

Toronto | Vancouver
Ottawa | Calgary | London

Greater than 1:1
technical:sales ratio

Background in architecting mission-critical
data centre infrastructure
4

}  The country’s most
skilled IT infrastructure
specialists, focused on
security, performance
and control tools
}  Delivering infrastructure
services which support
core applications

5

6

}  Why am I not getting the desired number of virtual desktops on a
vSphere host?
}  My ThinApp applications run really, really slow…
}  Why does it take 90 minutes to provision 5 linked clones?
}  I can’t print…
}  My physical desktop runs better than my virtual desktop…
}  My connections are load balancing right…
}  I can’t connect in from my home computer…
}  Hey, would you mind validating my design?
7

8

}  Improperly Configured/Sized Environments
–  No Physical/Virtual Desktop Assessment
–  Little Understanding of User Community & Usage Patterns
–  Vendor based “Guesstimation” of VM to Host Density
–  Oversubscription
–  Virtual Machine Configuration (vCPU & vRAM)
–  Lack of proper Plan & Design
}  Dependent Infrastructure Improperly Sized
–  Linked/Full Clone Storage (IOPS & Disk Speed)
–  File (CIFS/SMB) Storage Considerations (User Data & Persona)
–  Networking Bandwidth & Quality of Service

9

}  Customer Environmental Considerations
–  Contending with other network services & traffic
–  Using Shared Infrastructure
﹘ 
﹘ 
﹘ 
﹘ 

Database Server
Storage
Network
Server Infrastructure

–  Application & Peripheral Integration
–  Special Business & Technical Requirements (i.e. BCDR)
–  SLAs
10

Design Validation Benefits

11

}  Understand how their solution will perform and/or scale
–  How many desktops can I get onto a single host?
–  Will my storage or dependent infrastructure scale?
–  How do the components of the solution scale?
–  What will the user experience be when my infrastructure in under duress?
}  “Estimation vs Guesstimation”
–  Provides data to accurately estimate hardware/software needs
}  Identify problems BEFORE solutions are placed into pilot or production
deployments
–  Identify and fix your issues before your users experience problems.
–  Deployment with a minimal amount of issues is considered successful

12

13

14

}  Badly designed storage can cripple a VDI system
}  There are literally dozens of different vendors to choose from
}  Storage technologies are changing rapidly. Where do you begin?

15

}  Desktop workloads do not have the
same storage requirements as
server workloads
}  Desktops are extremely read heavy
during boot and logon but change
to write heavy during steady state
}  You need lots of random IOPS for
desktops (read AND write)
}  VDI desktops display bursty IOPS
throughout the day
}  Accommodate the burst demands

16

}  Implications of volatility
–  Larger swings in workload
–  Greater need for burst capacity; risk of inadequate
capacity
–  More risk to user experience
–  More performance driven user hostility
}  Conclusions
–  You need to lower volatility or provide adequate burst
capacity
17

}  Flash and SSD is very much the rage with storage vendors today
}  Terms like SLC, MLC, EMLC, PCI-based flash, flash acceleration are constantly
being tossed around
}  The real question is can the storage solution deliver a given amount of storage at
a given amount of IOPS, with a given amount of power, in given number of rack
units, at a specific price point?
}  What a vendor uses to accomplish that really doesn’t matter – as long as it
supports your requirements
}  Some vendors use pure flash, some a mix between flash/spinning disk
}  Some vendors accelerate spinning disk with RAM, others use flash as cache

18

}  VDI used to be based on a non-persistent desktop model in
order to achieve storage optimization
}  Users demand a persistent VDI model and storage solutions
can now accommodate that model with storage efficiencies
}  Although there are some great performance acceleration
technologies today – many only focus on non-persistent VDI
}  Make sure your storage solution for VDI supports persistent
desktop – in an optimized way across capacity &
performance
19

Vendor solutions

20

Hybrid Arrays

21

}  Combine the best of what both the SSD and HDD worlds have to offer and create
an array that leverages the benefits of solid state storage for performance while
continuing to be able to rely on the massive capacity benefits inherent in HDDbased solutions
}  Balance between the storage metrics of $/GB and $/IOPS
}  A hybrid array very nicely balances these two metrics and can satisfy a wide swath
of business needs (i.e. VDI)
}  Solid state storage is often used as a mega cache which provides significant
acceleration of all reads and writes from and to the array
}  Newcomers generally have the Greenfield advantage here as they can build
systems with flash in mind and can sometimes take more advantage of flash's
unique characteristics
22

}  New hybrid players rely on commodity hardware and differentiate themselves in software
and because they don't require as many expensive SSDs
}  Hybrid arrays carry a price tag that is a fraction of what one would find with an all-SSD
product
}  While it won't beat the all-SSD array in a head to head test, when one measures the
performance based on true needs, hybrid storage is often the perfect solution.
}  VDI - A hybrid array in this scenario is a perfect fit. Because commonly accessed files are
cached on super-fast SSDs, boot and login storms are a thing of the past.
}  Deduplication/compression savings in VDI is incredibly high, so you may find yourself using
much less disk space than you need
}  Because there is ample capacity in a hybrid device, an organization can consider allowing
users to create persistent, customizable desktops

23

}  Patented Cache Accelerated Sequential Layout (CASL) architecture
}  CASL accelerates applications by using flash memory as a read cache in
conjunction with a write-optimized data layout
}  Accelerated performance for greater throughput and more IOPS, and
latencies of less than a millisecond
}  Greater storage efficiency (compression), reducing the storage footprint
needed by 30 to 75 percent
}  It protects data by supporting instant snapshots for easy backup and
restoration
}  Efficient replication for disaster recovery
24

Key Benefits for VDI
}  Deliver a consistent highperformance user experience
}  Enable business continuity for end
users
}  Independently scale performance
and capacity without downtime
}  Reduce costs with efficient
capacity utilization
}  Leverage SmartStack reference
architectures to speed deployment
25

}  3U appliance
}  Tintri Flash First Design: VMstore is a hybrid storage solution
}  VMstore integrates flash as a first-class storage medium rather than as a bolt-on cache or tier
}  Tintri’s patented Flash First Design incorporates algorithms for inline deduplication, compression
and working set analysis to service more than 99 percent of all IO from flash
}  Desktop VM performance and responsiveness that meets or exceeds the performance of fast desktop
PCs and laptops
}  Desktop VM Pool creation and provisioning operations
}  User acceptance is sure to suffer if it takes an excessive amount of time to log in and gain access to
their desktops
}  Desktop pool refresh or “recompose” operations require at the very least at least one reboot
operation for each virtual desktop
26

}  VM-aware architecture (NFS based)
}  Each VMstore deploys quickly as a datastore to vSphere over 10GbE Ethernet.
}  Low latency performance, high storage I/O bandwidth
}  High VM density (large numbers of VMs) drives down $/VM costs
}  Inline deduplication and compression
}  Per VM snapshots and rapid cloning
}  Unique server, network and per VM statistics and correlations (Tintri Bottleneck
visualization)
}  A unique, well designed and modern UI
}  Tight vSphere and vCenter integration
27

Hyperconverged

28

}  Solutions comprised of hardware and software
}  Bundling the software with pre-architected hardware into hyperconverged
appliances allows them to provide customers with better support and be sure that
the total package can provide a predictable level of performance and capacity
}  Support for this predefined hardware set is much easier than simply supporting
whatever customers decide to pull off the shelf and use
}  Avoids the need to choose, test and optimize the environment, which results in a
longer time-to-value
}  best of breed” environment can be considered “undifferentiated heavy lifting”
due to the amount of work that it takes to make sure everything is operating as
expected

29

30

NDFS Cluster
Infinite Scalability
• 
Controller VMs create NDFS cluster
• 
Storage Pool created from all Disks (RF:2)
• 
NFS datastore presented to hypervisor (SMB for Hyper-V)
• 
I/O reads local, I/O writes synchronous
31

Built for Speed
The great majority of Read/Write I/O is serviced locally
•  vMotion or HA event triggers RF review and data locality
•  Only accessed blocks are brought local (RF maintained)
•  Background process (no flooding network)

32

NDFS

Pooled resources

NDFS

10Gbps Ethernet

SINGLE NODE

Compute:16 Cores

Storage:

Flash: 800GB SSD

HDD: 4TB

Controller: Nutanix Storage controller

33

}  An abstracted pool of storage that
looks like a single volume of
storage but acts like a virtual
hybrid-flash storage array
}  Built directly into the hypervisor,
so supports features like HA, DRS,
vMotion, Storage vMotion
}  Pool of storage is based on locally
mounted SSDs and HDDs
}  ESXi hosts without any local
storage can consume VSAN

34

}  Writes to VSAN are handled by SSD
and flushed down to HDD
}  Reads from VSAN are pulled from
SSD cache
}  Does not require a local RAID set,
just a bunch of disks
}  Virtual machines can be striped for
performance (data written to
multiple HDDSs) and protected by
replicas (data written to multiple
ESXi hosts)

35

Flash Hypervisor Software

36

} 

Flash Hypervisor software, like PernixData FVP, aggregates server-side Flash (PCIe or SSD) to accelerate reads and writes
to primary storage

} 

Allows IT administrators to scale storage performance independent of capacity, and without modifying existing
applications or storage

} 

10x faster application performance - accelerating reads and writes in server-side Flash, FVP accelerates storage I/O
where it has the most impact, on the host where the application resides

} 

Substantially more IOPS at a fraction of the cost - When FVP is used in conjunction with a single $600 SSD, you can get
approximately 50,000 IOPS. In comparison, a $100,000 SAN delivers the same 50,000 IOPS

} 

More capacity at a lower price - Since storage performance is coming from the server tier, you can optimize the SAN for
capacity and data services - less expensive, higher capacity drives in your SAN, lowering your cost per GB

} 

Increased VM density and minimized server hardware

} 

No Virtual machine modification

} 

Kernel mode integration and supports all hypervisor features for data resilience

37

} 

As data moves between VM2 and the LUN (shown
with gold arrows), the FVP layer acts as a read
and write-back cache while also mirroring the
write-back data to another host node (shown with
a white arrow).

} 

This is done over the vMotion network by default,
but you can change this to use a different
network, or make a dedicated network, if
desired.

} 

If the virtualization administrator or DRS migrates
VM2 from host ESXi A to B with a vMotion, the
flash will already contain the hot write-back data
and performance should remain relatively high.

} 

In the case of write-through data, FVP will allow
the host to read from a peer host’s flash tier to
avoid going back to the SAN until the local flash
has warmed up.

38

39

THANK YOU.

40

QUESTIONS?

41

Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014

More Related Content

What's hot (20)

Viewers also liked (14)

Similar to Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014 (20)

More from Scalar Decisions (20)

Recently uploaded (20)

Storage Considerations for VDI - Scalar presentation at Toronto VMUG 2014