Datacentric Networking and System Design
Datacentric Networking and System Design
09/09/2022 2
Data Centers
Data center (DC) is a physical facility that enterprises use to house computing and storage
infrastructure in a variety of networked formats.
Data Centers
Traditional data centers
• Host a large number of relatively small- or medium-sized applications, each running on a
dedicated hardware infrastructure that is decoupled and protected from other systems in the same
facility
• Usually for multiple organizational units or companies
Things to consider:
• Ownership
• Responsibility
• Access
• Up-front costs
09/09/2022 13
Outsource if:
• You have a short-term need for a server environment,
perhaps until a permanent Data Center is constructed
• You want a standby facility ready to take over for a
primary Data Center in the event of a catastrophic event.
09/09/2022 15
Cross-Functional Support
• Responsibility for a company's Data Center is typically
shared among multiple departments and personnel.
• Example:
• Security manager typically governs physical access into
the Data Center.
• IT manager coordinates where servers are physically
deployed.
• Each one has different point of view with regards to
security access.
• Solution: Foster communication and seek compromise
09/09/2022 20
Make It Robust
• Above all, your Data Center has to be reliable. Its overarching reason
for existence is safeguarding your company's most critical equipment
and applications. Regardless of what catastrophes happen you want
your Data Center up and running so your business continues to operate.
• Data Center infrastructure must have depth: standby power supplies to
take over when commercial electricity fails, and redundant network
stations to handle the communication needs if a networking device
malfunctions.
• The infrastructure must be configured so there is no single component
or feature that makes it vulnerable. It does little good to have multiple
standby power systems if they are all wired through a single circuit, or to
have redundant data connections if their cable runs all enter the building
at one location.
• In both examples, a malfunction at a single point can bring the entire
Data Center offline.
09/09/2022 22
Make It Modular
• Your Data Center must not only have a depth of infrastructure, it
must also have breadth. You want sufficient power, data, and
cooling throughout the room so that incoming servers can be
deployed according to a logical master plan, not at the mercy of
wherever there happens to be enough electrical outlets or data
ports to support them.
• To achieve this uniform infrastructure, design the room in
interchangeable segments. Stock server cabinet locations with
identical infrastructure and then arrange those locations in identical
rows. Modularity keeps your Data Center infrastructure simple and
scalable. It also provides redundancy, on a smaller scale, as the
standby systems mentioned previously. If a component fails in one
section of the Data Center, users can simply plug in to the same
infrastructure in another area and immediately be operational again.
09/09/2022 23
Make It Flexible
• It is safe to assume that routers, switches, servers, and
data storage devices will advance and change in the
coming years. They may become smaller or bigger.
• Data Centers are not static, so their infrastructure should
not be either. Design for flexibility. Build infrastructure
systems using components that are easily changed or
moved.
• Inflexible infrastructure invariably leads to more expense
down the road.
• Part of a Data Center's flexibility also comes from whether
it has enough of a particular type of infrastructure to
handle an increased need in the future.
09/09/2022 24
Standardize
• Make the Data Center a consistent environment. This provides
stability for the servers and networking equipment it houses, and
increases its usability.
• When building a new facility, it might be tempting to try something
different, to experiment with an alternate design philosophy or
implement new technology. If there are new solutions that truly
provide quantifiable benefits, then by all means use them. Do not
tinker with the design just to tinker, though.
• Once you find a design model or infrastructure component that
provides the functions and features you are looking for, make it
your standard. Avoid variety for variety's sake. The more complex
the environment, the greater the chance that someone will
misunderstand the infrastructure and make a mistake, most likely
in an emergency.
09/09/2022 25
• Life Span
• How long it is expected to support your company's needs
without having to be expanded or retrofitted, or otherwise
undergo major changes.
• The most effective strategy is to design a Data Center
with a projected life span of a few years.
09/09/2022 38
• Budget Decisions
• It is no good to spend millions of dollars on a server
environment to protect your company's assets if that cost
drives your business into bankruptcy.
• The most obvious costs for a Data Center are labor and
materials associated with its initial construction, which,
even for a room smaller than 1000 square feet or 100
square meters, normally runs into hundreds of thousands
of dollars. This includes:
• Initial construction
• Consulting fees
• Real estate
• Ongoing operational expenses
09/09/2022 39
• Budget Decisions
• It depends on the downtime cost
• The cost of a generic employee at your business and then multiply this by the
length of the outage and by how many employees are unable to work during
downtime/
• For example, a generic employee costs your company a total of $150,000 a year.
(Remember, this is all costs combined, not just salary.) That is about $60 an hour,
assuming the employee works a traditional 40-hour work week, and 52-week calendar
year. If your Data Center goes offline for two hours and stops the work of 100 employees
at that site, that is $12,000 for that single outage.
• Or by calculating the revenue:
• Assume that your company typically brings in $1 million a year in online business. If the
website accepts orders around the clock, then divide $1 million by 8760, the number of
hours in a year. That works out to $114 an hour, which means that the four hours of
downtime also disrupted about $500 in sales.
• Most difficult value of all to quantify comes from when a server crashes and
data is destroyed.
• For example, intellectual property has been destroyed.
09/09/2022 40
CCN Nodes
Processing an Interest:
– Matching Data is found in the Content Store
=> send it and consume Interest
– Pending Interest in PIT
=> add this face to RequestingFaces list
– Use FIB to forward Interest on outgoing faces, add to PIT
Processing Data:
Data follows a chain if PIT entries back to the source
Duplicate and unsolicited Data is discarded
09/09/2022 47
09/09/2022 48
Naming Content
Hierarchical content names with a flexible format
Individual name consists of a number of components
Names can be relative to some known name, e.g.
next/previous
Same content can have multiple names! Problems with
caching?
A source of data performs a Register operation for a
prefix
09/09/2022 50
Routing
Routing between CCN nodes can occur over
unmodified OSPF.
Incremental deployment of CCN nodes is possible
Integration with BGP is also possible
Routers do not construct spanning trees
Loops are not possible anyway
Multiple paths can be used
09/09/2022 51
Network Security
Sending a malicious packet to a host is difficult
because CCN talks only about content, not to hosts
Data based DoS attacks are impossible because only
one Data packet is forwarded per Interest
Interest flooding:
Multiple Interests for the same content are combined
Limit the forwarding of unsuccesful interests
What if sender and receiver collude?
09/09/2022 53
Evaluation
Transfer time vs Number of Sinks
09/09/2022 54
Evaluation
Failover
09/09/2022 55
DOT
DOT provides an API and a plugin architecture
Transfer Plugins: eg. Multi-path, portable storage
Storage Plugins: access to local data, divide data into
chunks, compute hashes
Basic API:
Sender calls put with data, gets back an OID
Receiver uses OID to get data
09/09/2022 58
Evaluation
Multipath Plugin: Using two 100 Mbit/s Ethernet links,
transfer time went down from 3.59 seconds to 1.90
seconds
Modified Postfix mail server to use DOT
Minimal modification: 184 LoC
DOT saves 20% of total message bytes transferred
Duplicated messages
Partial redundancies in messages
09/09/2022 59
Outline
Motivation
What is content distribution ?
Schemes for content distribution
• Web Caching
• Content Distribution Networks
• Peer-to-Peer File sharing (not covered)
CDN Internetworking
What content is/is not suitable for CDNs?
CDNs vs. Caches
09/09/2022 60
CDN
• A content delivery network, or content distribution
network (CDN), is a geographically distributed network of
proxy servers and their data centers.
• The goal is to provide high availability and performance by
distributing the service spatially relative to end users. CDNs came
into existence in the late 1990s as a means for alleviating the
performance bottlenecks of the Internet[1][2] as the Internet was
starting to become a mission-critical medium for people and
enterprises.
• Since then, CDNs have grown to serve a large portion of the Internet
content today, including web objects (text, graphics and scripts),
downloadable objects (media files, software, documents),
applications (e-commerce, portals), live streaming media, on-demand
streaming media, and social media sites.[3]
09/09/2022 61
CDN
• CDNs are a layer in the internet ecosystem. Content
owners such as media companies and e-commerce
vendors pay CDN operators to deliver their content to
their end users. In turn, a CDN pays
Internet service providers (ISPs), carriers, and network
operators for hosting its servers in their data centers.
• CDN is an umbrella term spanning different types of
content delivery services: video streaming, software
downloads, web and mobile content acceleration,
licensed/managed CDN, transparent caching, and
services to measure CDN performance, load balancing,
Multi CDN switching and analytics and cloud intelligence.
09/09/2022 62
CDN
• CDN nodes are usually deployed in multiple
locations, often over multiple Internet backbones.
Benefits include reducing bandwidth costs,
improving page load times, or increasing global
availability of content.
• The number of nodes and servers making up a
CDN varies, depending on the architecture, some
reaching thousands of nodes with tens of
thousands of servers on many remote
points of presence (PoPs). Others build a global
network and have a small number of
geographical PoPs
09/09/2022 63
Why CDN?
• Over the last few years there has been a huge
increase in the number of Internet users.
YouTube alone has 2 Billion users worldwide,
while Netflix has over 160 million users.
Streaming content to such a wide demographic of
users is no easy task.
• One can think that a straightforward approach to
this can be building a large data center, storing all
the content in the servers, and provide it to the
users worldwide. But there are issues that arise
when this approach is followed-
09/09/2022 64
Why CDN?
• Firstly if the data center is in the USA and the user is in
India there will be slower delivery of content.
• Secondly, a single data center represents a single point of
failure.
• Thirdly, if some content is being accessed frequently from
a remote area then it is likely to follow the same links, and
this, in turn, results in wastage of bandwidth.
09/09/2022 65
What is a CDN?
•
A CDN is essentially a group of servers that are
strategically placed across the globe with the purpose of
accelerating the delivery of web content. A CDN-
• Manages servers that are geographically distributed over
different locations.
• Stores the web content in its servers.
• Attempts to direct each user to a server that is part of the
CDN so as to deliver content quickly.
09/09/2022 66
With CDN
09/09/2022 69
Example
• Suppose you are hosting a website, wherein your origin
server(server containing the primary source of your
website’s data, where website files are hosted) is located
in Australia and a company XYZ provides you the CDN
service.
When a user in India clicks on a video on your website,
the request goes to the user’s local DNS server(See DNS
), which relays the request to the authoritative DNS server
of your website.
09/09/2022 70
CDN Security
• CDN providers profit either from direct fees paid by
content providers using their network, or profit from the
user analytics and tracking data collected as their scripts
are being loaded onto customer's websites inside their
browser origin.
• As such these services are being pointed out as potential
privacy intrusion for the purpose of behavioral targeting[6]
and solutions are being created to restore single-origin
serving and caching of resources
09/09/2022 72
CNN.com
congested link
Server Farm
Server-1 Server-2 Server-n
Requests = R/n
L4-L7 Switch
Requests = R
Internet
09/09/2022 79
Total delay =
Internet delay + Internet delay=2 sec
Access delay
Δ – traffic intensity
09/09/2022 80
Web server
GET
Object present ?
No-> Fetch Object
Yes-> Send Object RESPONSE
RESPONSE RESPONSE
GET GET
Cache
Client 1
09/09/2022 81
Web Cache
Total delay =
(2 + .01) x 0.6 = 1.2 Sec delay = tens of milliseconds
ΔAL = 0.6 1.5 Mbps access link
Institutional
cache
Hit rate = 0.4 100 Mbps LAN
Δ – traffic intensity
09/09/2022 82
Parent
Caching proxies serve only their clients, not all users on the Internet
Replica
congested
Replica
Not congested
Client
09/09/2022 85
Terminology
• Content: Any publicly accessible combination of text, images,
applets, frames, MP3, video, flash, virtual reality objects, etc.
ers
v
Ins
ser
s
tal
S ell
ls
erv
Cisco,
er s
Lucent, H/W and S/W
Vendor Exodus
Inktomi, Hosting
CacheFlow Provider
09/09/2022 88
CDN: Distribution
Origin server in
North America
push content
Akamai CDN
CDN distribution node
push content
push content
CDN server in South
America
push
CDN server in Asia
content
CDN server in
Europe
09/09/2022 89
CDN
Request Distribution
Routing and
Infrastructure Accounting
Infrastructure
Surrogate
Surrogate
Client
09/09/2022 91
t D NS
o c lien www.cnn.com
s ure t
Mea e sults
R Akamai DNS
www.cnn.com
32.22
63.251.132.22
surrogate 63.210.135.39
surrogate
pi
63.251.1
ng
g
pin
Session
www.cnn.com
63.251.132.22
Content Modification
CNN.com PUT Authoritative DNS server for cdn.com
/ima
g es/*
.g if
GET www.cnn.com/index.html
?
Index.html
om
64.236.24.28
n.c
cd
28 .
ry:
Index.html
.24
g if
q ue
.
236
s/1
...
g e
<img a
64.
S
im
DN
src="http://www.cdn.com
n n/
c
T/ if
/cnn/images/1.gif”>
. g
... GE 1
64.236.24.28
Client Local DNS server
09/09/2022 94
Metrics
• Network Proximity (Surrogate to Client):
•Network hops (traceroute)
•RTT
•Internet mapping services (NetGeo, IDMaps)
•…
• Surrogate Load:
• Number of active TCP connections
• HTTP request arrival rate
• Other OS metrics
• …
• Bandwidth Availability
09/09/2022 95
CDI: Architecture
Digital Island
ATT
Akamai
comcast
Content
• Images
Suitable for CDNs
• High-volume e-commerce transactions (thanksgiving sale)
• Streaming media (audio and video) (media events)
• Java Applets
• Virtual Reality Objects
• Flash content
References:
• Michael Rabinovich and Oliver Spatsheck, “Web Caching and
Replication “, Addison-Wesley 2001.
• PPT slides by Janardhan Iyengar on “Overlay Networks”
• PPT slides by Brad Cain on “Interconnection of Content Delivery Networks”
• http://www.cis.udel.edu/~girish/856/cdn-bib.pdf
09/09/2022 101
Proxy deployments
• Non-transparent
Explicit client configuration
Browser auto configuration
Proxy auto discovery
• Transparent
Connection “Hijacking” or interception.
09/09/2022 102
Internet
Other traffic
Proxy
09/09/2022 103
Client IP = a1 Proxy IP = a2 Origin Server IP = a3
SYN(a1 to a3)
to a1)
SYN/ACK(a3
ACK/HTTP re
quest (a1 to a3
)
SY
N(a
2 to a
3)
to a2)
CK(a3
/A
SYN
ACK
/HT
TP
requ
est
(a2
to a
3)
to
ta (a3
Da
) a2)
Data (a3 to a1
09/09/2022 104
Disk-to-Tape
Disk-to-Cloud
• Disk-to-cloud solutions facilitate the recovery of individual files which may have
been corrupted or accidentally erased – a process that’s unduly time-consuming
with simple tape backups. Recovery times with cloud backup may be in the order of
seconds, rather than the hours or more it might take to sort through tapes.
• Encryption of data at the source helps resolve any nagging issues about the
security of what’s essentially a third-party storage medium. However, the risk
remains of accidental or deliberate corruption of information which is held online.
• Named as a leader in Gartner’s 2016 Magic Quadrant for Data Center Backup &
Recovery for the upper-end mid-market and larger enterprise environments, Veeam
Backup & Replication provides in-place recovery for fast recovery times, in what the
vendor describes as an offering of ‘Availability for the Always-On Enterprise’. It’s
capable of protecting and recovering an entire virtual machine (VM), as well as
individual files.
09/09/2022 129
• The ease with which files, folders and volumes may be recovered is the
litmus test for recovery software. Key metrics to watch for here include:
• RTO (Recovery Time Objective): How long it should take to get
everything back in order
• RPO (Recovery Point Objective): How much data the enterprise can
stand to lose, and still remain viable
• From a user perspective, the software should have controls for
allocating and managing access rights and privileges.
• For virtual desktops, mobile workspaces, or web-server farms and other
systems requiring heavy replication, facilities for compression and
deduplication, together with the capturing of roll-back snapshots
representing stable configurations are also a plus.
09/09/2022 133
• In order to supply cooling air to the places of high heat in front of the
enclosure, perforated tiles are installed. Tiles for service under the raised
floors systems can be removed from the floor and they can be shifted to
other places.
• Close to the particularly strong heating equipment, the devices that
guides the air flow into the holes of perforated raised floor tiles can be
installed. When the premises have to change something, tiles are
removed from the floor and they are moved to another location.
• In the initial arrangement of the room they all want to ensure optimal, as it
seems, the balance of computer and communications equipment. But
people often do not know at what point they will need this technique in the
future, therefore they do not know what changes need to be made in the
scheme of its deployment. It is therefore desirable to have a solution that
allows you to quickly cost-shift the equipment, introduce new
technologies to replace the equipment. If you are completely sure that you
will never make any changes and do not need to introduce new
technologies, raised floor is not for you.
09/09/2022 139