Top Technology Trends

Advanced Systems Group

Top Technology Trends for
Virtualization


Today’s Discussion

1 NAND Flash
What you need to know…

2 Infrastructure Continuance
Why DR is so important (really)…

3 Architecting for the Cloud
We won’t always be in Oklahoma City…


Central Region
• Denver, CO – Corporate Headquarters
• Salt Lake City, UT
• Boise, ID
• Portland, OR
• Houston & Dallas, TX
• Seattle, WA
• Louisiana
• Oklahoma City, OK

Western Region
• San Diego, CA – Regional Headquarters
• Phoenix, AZ
• Los Angeles, CA
• Orange County, CA


Best Practices

SONY


Thought Leadership


ASG Recommendations


Commercial Break


Top Technology Trends…

Architecting for
the Cloud

Infrastructure
Continuance

Flash & Storage


I/O, I/O, It’s Off to Work I Go…
Performance Curve
Density IOPs / Density
Gigabytes

IOPs

1985 2010 1985 2010
Capacity vs. Performance Access Density Trend

Spinning Hard Disk Drive (HDD) Trends


Most Disruptive in 50 Years!


First Off, Don’t Believe
Benchmarks…


clear winner for archive
Whoa! good for big sequential

Cap $/GB Reads $/Read Writes $/Write
(8KB/sec) (8KB/sec)

SATA HDD 2,000G $0.12 130 $1.83 130 $1.83

SAS 300G $1.30 17,500 $0.02 17,500 $0.02

SLC 64G $11.00 24,500 $0.03 4,200 $0.16

DRAM 32G $22.00 325,00 $0.002 325,000 $0.002
0
good for random access winner for small R/W

Storage Media Comparison


Killing Latency

1. SSDs Slotted into HDD Slots 4. All-Flash Arrays

2. Flash in Array Controller 5. SSDs Slotted into Server
Cache
6. Flash Used Strictly as
3. Newly Architected Arrays Cache
using Flash/DRAM


Killing Latency


2. Flash in Array Controller 5. SSDs Slotted into Server
Cache
6. Flash Used Strictly as
3. Newly Architected Arrays Cache
using Flash/DRAM
7. Storage Appliances (VSA)


Hey, But I Don’t Have a
Performance Problem?


SSDs Slotted into HDD Slots
• Overlaying flash on existing architectures is expensive and
severely limiting…
Data placement software to determine what is flash-worthy
Net new data is not analyzed, goes to slowest tier first
SSDs as a tier mandates h/w RAID (designed for HDDs)
Inability to respond rapidly to changing performance
demands, read, too many cache misses (think VMware)
Where is the metadata? What is the granularity of data
transferred?
Must use SLC due to all the data movement


The “Disk is Dead” Approach
“Bring flash to the masses…”
All data is on flash so no warming, no copy, no cache
miss, no analysis…
CPUs are cheap, but capacity isn’t (200TB max)
RAID-less all-SSD scale-out storage…
Too much performance (for ~$5-10/GB, and
dropping)?


How to Kill Latency
HDT, FAST VP, Data Progression Nimbus, WhipTail, Violin, Pure
Storage, ExtremIO, SolidFire, TMS
2. Flash in Controller Cache
FlashCache, FAST Cache 5. SSDs Slotted into Server Slots
3. Newly Architected Arrays FusionIO, OCZ, Micron, Project
using Flash/DRAM Mercury, VFCache, Virident, TMS,
SanDisk
NexGen, Tintri, Nimble,
Starboard, Sun Storage 6. Flash Used Strictly as Cache
Appliance Avere, Violin, GridIron


Secret Decoder Ring
• What do you got?
And what are you trying to do?
• Automated tiering?
No refresh planned
Static workloads, hot stays hot, cold stays cold
Consider flash appliances?
• Cache hybrid?
Ideal for SMB
VDI


Secret Decoder Ring
• Flash only?
Storage refresh is planned
Data not cache friendly or can’t be segmented easily
Performance is a high priority (versus cost)
• Flash in servers?
VDI
Data fits easily on form factor
No need for sharing, clustering or vMotion (not true anymore)
Integrated with storage controller


Roads & Roadmaps…


Brave New World…


NFS is the New Black…

FC
NFS NFS NFS

FC VMFS

SAN


Especially With 10Gig Ethernet
(& vSphere 5)

10Gbe Module SFP+

Non-Blocking 1000
Mbps with four 1/10GbE
uplinks (SFP+)

10Gbe PCIe cards
Modular 10Gb Ethernet Switching 10 Gig Multimode
Physical Plant
Cabling


Why is DR Necessary?

Virus Nature
7% 3%
Software Errors
14%
Hardware
44%
Human Error
32%

Source: Ontrack Data International


Infrastructure Continuance
Disaster Business Interruption

An event or chain of events that Any event whether planned or
result in the inability for you to unplanned that disrupts the
provide critical business normal course of your
functions for a period of time operations, which if not
and/or causes the company to addressed within a
move from using SOP to predetermined amount of time,
employing its disaster recovery may turn into a disaster.
plan, potentially at another
facility.


RTO is Multi-Dimensional
Current Proposed Solution

RPO RTO RPO RTO
File Recovery
Best-Case Snap Interval 5 min. Snap Interval 5 min.
Typical-Case On-Site Incre.** 1 hour Snap Interval 5 min.
Worst-Case Off-Site Full** 1 day Off-Site Full** 1 day

Mailbox Recovery
Best-Case Snap Interval 5 min. Snap Interval 5 min.
Typical-Case On-Site Incre.** 1 hour Snap Interval 5 min.
Worst-Case Off-Site Full** 1 day Off-Site Full** 1 day

Data Corruption
Best-Case On-Site Incre.** 1 hour Snap Interval 5 min.
Worse-Case Off-Site Full 1 day Snap Interval 5 min.
& Incre.**

Array Failure
Best-Case Off-Site Full 2 days Snap Interval 5 min.
& Incre.**

Data Center Failure
Best-Case Off-Site Full 5 days Off-Site Full 2 days


Infrastructure Continuance
The New Backup
Backup
Production
Target(s)

IP LAN IP LAN / WAN
Virtual Machine

R: R: R: R:
E:
D: E: E: E: E:
D: D: D: D:

Continuous data and application protection

Operating system, system settings,
partition information for BMR
Point in Time
Recoveries


Architecting for the Cloud…
Server?
Performance?
Application?
SLA?
Data?
Connectivity?
Service?
Update?
Security?
Privacy?
Scaling?
Capacity?
SAVINGS??????


Analytics to the Rescue


Create Standards
• Server Vendor X has been preselected as the platform of choice
• Eight 1GbE ports will be used per server
• NAS offering has been preselected as the solution of choice
• All Tier 2 NAS volumes are de-duplicated
• Physical switches will not be configured for QoS
• Existing top of rack environment should be used for the virtual
infrastructure


Best Practices
(or avoiding Worst Mistakes…)
• It is best practice to have a 500GB LUN with a maximum of 15 VMs
• It is best practice to have an 8 host cluster
• It is best practice to have a maximum of 4 VMs per core
• It is best practice to have a dedicated 1GbE link for vMotion
• It is best practice to have “high / medium / low” resource pools
• It is best practice to have an isolated management
• It is best practice to configure vMotion and Management on a VSS
and the other traffic on a VDS


Best Practices?
• It is best practice to have a 500GB LUN with a maximum of 15 VMs
• It is best practice to have an 8 host cluster
• It is best practice to have a maximum of 4 VMs per core
• It is best practice to have a dedicated 1GbE link for vMotion
• It is best practice to have “high / medium / low” resource pools
• It is best practice to have an isolated management
• It is best practice to configure vMotion and Management on a VSS
and the other traffic on a VDS


Seeing vMotion Isn’t Exciting…
(But seeing bad HA and DRS can be very interesting)

• A surprising number of environments have configured
HA/DRS settings incorrectly.
• Some do so because of hardware constraints.
• Others have not designed architecture with HA/DRS in
mind.
• Even others have introduced problems as they scale
upwards.


So What Do You Do?
• As a virtual environment ages, hardware is refreshed and new
hardware is added. Refreshes sometimes create “islands” of
vMotion capability.
• How can we always vMotion between computers?
You can always refresh all hardware at the same time (you
can talk to me after the show ).
You can cold migrate, with the machine powered down.
You can use vMotion Enhanced Compatibility (EVC) to
manage your vMotion-ability.


Caveat Emptor…
• Storage vMotion has some special requirements. Virtual
machines with snapshots cannot be svMotioned.
• Virtual machine disks must be persistent mode or RDMs.
• The host must have sufficient resources to support two
instances of the VM running concurrently for a brief time.
• The host must have a vMotion license, and be correctly
configured for vMotion.
• The host must have access to both the source and target
datastores.


Planning on Overcommit?
(a.k.a. Creating Unnecessarily Big VMs)
• Back during the “hypervisor wars” one of VMware’s big sales
points was memory overcommit. “ESX can overcommit memory!
Hyper-V can’t!”
• So, many of us used it.
• Overcommit creates extra work for the hypervisor Ballooning,
host memory swapping, page table sharing
• That work is unnecessary when memory is correctly assigned.
• Assign the right amount of memory (and as few processors as
possible) to your VMs. Creating “big VMs” also impacts DRS’ load


What About Security?
• Losing a tape with VM images on it is easier to recover and
run than a backup tape from a physical server.
• You can make a snapshot, start up a VM from the snapshot,
changing the password file and run it without problems.
• Traditional network security tools can’t necessarily see the
traffic that passes between VMs communicating with each
other inside a single host, making it harder to monitor
inappropriate traffic flows.
• Change management procedures should also be reviewed to
establish how and when changes are documented.


VMs Easier to Hack Than
Physical Machines
• Memory images are
exposed.
Passwords, crypto
keys, email, AD, …
• Sensitive data can be
left everywhere the VM
travels…
• VM Templates need to
be protected.


Is it Beer Time?

Advanced Systems Group ®
Technology for the Speed of Change

Thank You

Top Technology Trends

More Related Content

What's hot (20)

Similar to Top Technology Trends (20)

More from InnoTech (20)

Top Technology Trends

Editor's Notes