Zach Rosenfield Senior Program Manager Microsoft zrosenfield Meet Cost Drivers Overview MSIT Consumption How Does SPO Scale Topics Tips for Running OneDrive for Business Storage Costs ID: 178792
Download Presentation The PPT/PDF document "Give users all the storage they want wit..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1Slide2
Give users all the storage they want without going broke
Zach Rosenfield Senior Program ManagerMicrosoft
@
zrosenfieldSlide3
MeetSlide4
Cost Drivers Overview
MSIT Consumption
How Does SPO Scale?
Topics
Tips for Running
OneDrive
for BusinessSlide5
Storage Costs
File Count And Size
Cost Drivers
File Sync Overhead
Growth Management
OneDrive
for Business costs beyond CAPEX and OPEXSlide6
Microsoft’s own tenant - A year after launch…
92% of growth is from users who use the sync client
Know your organization!
Will grow another 20TB in the next 90 days.
File Counts & Size
60TB
of files on disk
150k
new
files per
day
2 MB
average
file size
150k
u
sers
40k
s
ync dailySlide7
What is Sync Traffic?
OneDrive Sync
+
Office Document Cache
Sync
LIST.ASMX
ITEMS TO SYNC
OneDrive for Business Client
Content DB
CELLSTORAGE.SVC
Content-Type:
multipart/related
< 3.5 MB
Content-Type
:
multipart/relatedSlide8
Compute & Bandwidth Heavy
File Sync and background processing50 cores for 40,000 syncing users. All of Microsoft will require nearly 200 cores
Avoid
sharing with front ends used for user “experience” traffic
Sync Overhead
50
CoresSlide9
backups
No downtime comes at a large cost!
The more machines, the more overhead
OPEX is real!
Security
Monitoring
Overhead & Design
DB Integrity
Physical Machine Failures
Patching
Storage Costs
database
raid
database
raid
database
raid
database
raid
secondary datacenter
T-logs
T-logs
mirror
mirror
backupsSlide10
Database
operations are hardSingle tenant instances can spike unexpectedly
Growth rates require constant “DB Splits”
Constant rebalancing across machines
Read-only window required for users
Database-level growth projections are unreliable – just-in-time maintenance
Onboarding Strategies
Avoid Pre-provisioning. Impossible to balance correctly upfront and empty sites consume resourcesThe search index grows rapidly with the number of sites and files (over 25 indexed items per site by default!)
Manage Growth and Onboarding
Team site growth
OneDrive
growthSlide11
How Does SharePoint Online Work Then?Slide12
SP2013 average creation
time per site: 30s
Assume serialized
site
creation per C
ontent DB
Average 2,880 sites per day per
Content DBPer Content DB, that’s 120/hr and ~1000 per day during work hours
Beware
of averages!
30 users/hour != 300 users at 9amShared resourcesOther SQL actions like migrating filesOther types of site creations for the same Content DB or serverSharePoint Online QueuesNormally under a minute, but can get long during times of high activity
Onboarding QueuesSlide13
Enabling
SPFeatures is the “long pole”
of
creating a new site
Rolling into SharePoint Online now: Fast Site Provisioning
“Site Master” per language, template, and
databaseCopied directly in SQL to new site
Not all features support this yetUpcoming improvements
Current release
improves speeds by 30%
In testing—an additional 75% improvementFast Site Provisioning
s
econds
fast site provisioning: phase 1
2013 provisioning
Latency distributionSlide14
Predictive Capacity
Storage on
disk
as indicator of
growth
Only model one thing (maintain a consistent
ratio of compute)Proactive Growth Modeling
Storage capacity: Project exponential growth at observed rate -- adjust rate monthly.
@Scale – Apply storage as needed
Lead time dictated by hardware onboarding speed
Actual Limit
Observed
Growth
Goal
State
Aggressive
Projection
Standard
ProjectionSlide15
boundless cloud storage platform
metadata
1Tb
Site Collections
& OneDrive!
Total
Storage
∞
Unlimited storage.
Unparalleled Security.
Fort
Knox
Introducing…
You need to know your data is securely protected
You want infinite storage that expands to meet your growing needsSlide16
RBS vs Fort Knox
Content DB
File Read Scenario
Request file
get location
request file
request file
request file
request file
2013 RBS
Fort Knox
RBS does garbage collection & orphan cleanup using SQL Tables & Indexes
RBS does full blob enumeration to find orphans
RBS does not allow for fine-grained permission
RBS uses windows timer jobs (harder to scale/coordinate)
RBS pools are an un-necessary management layer for SPO
Fort Knox only has the permissions it requires
Fort Knox can write and read directly into the blob store
Fort Knox lowers the impact on SQL IO—leaving it for metadata
Fort Knox uses storage account pools without intermediate abstractionSlide17
Blob C
Blob B
Blob A
Version 1
A,B,C Uploaded
Initial
Shred
Blob G
Blob D
Blob E
Blob F
Blob B
Blob A
Version 2
D,E,F,G Uploaded
Blob C
Update
Blob G
Blob D
Blob H
Blob FBlob BBlob AVersion 3H Uploaded
Blob EUpdate
A Blob’s Life
Unsaved Doc
How do Blobs shred in SharePoint?Slide18
Fort Knox: Advanced Encrypted Store
A
B
C
D
E
Key Store
A
B
C
D
Content DB
A
B
C
D
E
cryptoSlide19
Windows Azure Storage
a=2
b=3
erasure
coding
standard
replication
a=2
b=3
a=2
b=3
a+b
=5
Append only distributed file system (immutable store)
Local and geo
r
eplication
Erasure
coding
(from 3 local copies to 1.33*)Can handle as many as 4 local failures – better than traditional replication *Reaches 1.33 through Reed-Solomon 12 fragments + 2 local par + 2 global par
a=2
Azure: 12+4Slide20
Managing SQL Growth
Backup & Restore
Log Shipping
sitemap
update
sitemap
update
catch up
clean up
clean up
No windows for maintenance
Eliminate shrinks and defrags
Seamless rebalance with move
But, split
r
equires RO time
Introducing DB Mitosis
New technology to enable very fast database growth management
“Splits” a database into two parts within seconds
flag
deletions
flag
deletionsSlide21
Healthy SQL
SPO-Aware Monitoring
Availability
High I/O
Full Disk
Blocking
Host & Mirror Auto-Heal
SQL Aware Patching
Patchable
Normal
Runtime
HA
Patching
failovers
Split Aware
Keep Books In Sync
Auto-Manage HostsSlide22
On Prem ImpactSlide23
SQL Storage Layout (Blobs inline)
Location
Array
Logical drive
RAID
Size (GB)
Used For
Chassis
2x
900GB
C
0
500
system
D
0
1,176
system
logs
6x 900 GB
E
0
5,000
content & temp
logs
MSA
24x 900GB
H
1+0
9,500
content
data
O
1+0
505
temp
data
Hard Drive: 900GB 10k RPM SAS (2.5”) DriveSlide24
9
00GB @ 10k RPM HDD
24 disk (Data) + 8 disk (logs,
os
,
etc
)
9.5 Usable TB (RAID 10)
$10,176 Per Server (Disk Only)
40k Users
(>500 servers @ 2x HA)
$8.24/user
25GB - Just for Disk
!
$5
/
user
25 GB each
99.9% AvailabilitySlide25
OneDrive
For BusinessExpand
your footprint
today
g
o fast
unified search
s
ave moneySlide26
Know your business
Availability Requirements?Scale?Usage patterns?
When to use Remote Blob Storage (RBS)?
Lots of large files that are mostly “write-once, read rarely”
If SQL Storage costs are higher than the cumulative
hardware+ops cost of running the additional RBS system
Make sure the disaster recovery story meets your needs!
Lots of RBS Providers out there…Staying On-Prem?Slide27
StorSimple solution for SharePoint
Inactive Data
+
Backup Copies in Cloud
Speed of SSD + Elasticity of Cloud
SAS
Local Tier
Most
Active Data
on SSD
On-premises datacenter
Hybrid cloud storage
for SharePoint
StorSimple solution+
Windows
Azure
Addresses core SharePoint
challenges
iSCSI storage integrates transparently
Local performance + cloud
elasticity = good experience, lower cost, seamless scalability
Enterprise-grade security for cloud
Automated
snapshots +
rapid
DR =
better data protection, less mgmt time
Cloud storage
StorSimpleSlide28
In Summary
Realize the true cost of large scale storage
Know your organization
Prepare for Growth Management
Find the solution that fits your needsSlide29
MySPC
Sponsored by
connect.
reimagine.
transform.
Evaluate sessions
on
MySPC
using your
laptop or mobile device:
m
yspc.sharepointconference.comSlide30
Thank You!
Questions?
@
zrosenfield
Icons courtesy icons8.comSlide31
Erasure Coding References
https://www.usenix.org/conference/atc12/technical-sessions/presentation/huang
http://
blogs.msdn.com/b/windowsazure/archive/2012/06/13/usenix-best-paper-award-erasure-coding-in-windows-azure-storage.aspx
ReferencesSlide32
©
2014
Microsoft Corporation. All rights reserved. Microsoft, Windows,
and
other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.