BACKUP/MASTER: Building a Storage Wide Area Network (WAN) for Enterprise DR

1
BACKUP/MASTER Building a Storage Wide Area
Network (WAN) for Enterprise DR

• Dragon Slayer Consulting
• Marc Staimer, President CDS
• marcstaimer_at_earthlink.net
• 22 September 2004

2
Dragon Slayer background

• 7 yrs sales
• 7 yrs sales mgt
• 10 yrs mkting bus dev
• Storage SANs
• 6 years consulting
• Launched or participated
• 20 products

• Paid Consulting
• gt 70 vendors
• Unpaid Consulting
• gt 200 end users
• Known Industry Expert
• Speak 5 events/yr
• Write 3 trade articles/yr

3
Storage DR WAN level setting

• Storage WANs by definition
• Primarily for DR purposes
• Enterprise DR Characteristics
• Big blocks of data
• Can overwhelm standard IP routers
• Not limited to nights weekends
• Time is very relevant
• Time windows are getting smaller

4
Classifying DR data

• Mission-critical Crucial
• Organizations vital data
• Primary business processes, primary applications,
  SLAs
• Data access loss often means organizational death
• Essential Secret
• Very important to the organization
• Day-to-day business processes
• Instantaneous recovery preferred not required
• Important Valuable
• Many day-to-day organization ops apps
• Low-critical Nominal value
• Low organizational value

5
Prioritizing DR data

• Mission-critical Crucial
• eCommerce/order-entry/sales transactions
• Essential Secret
• Customer data/intellectual property/Email
• Important Valuable
• Employee records/marketing collateral
• Low-critical Nominal value
• Resumes/market data/competitive data

6
Recovery Point Objectives (RPO)

• RPO point-in-time which systems data
• Must be recovered TO within the DR facility

7
Recovery Time Objectives (RTO)

• RTO total time which systems apps
• Must be recovered AFTER an outage

8
Picking WAN DR options

• Mission critical
• Split mirror disk-to-disk
• Synch mirroring
• Hot standby at remote locations
• Servers and storage
• Continuous snapshots
• Essential
• Asynch remote mirroring
• Snapshot
• Distributed backup
• Volume copy

• Important
• Wide area backup
• Distributed directory journaling
• Low critical
• Electronic Tape vaulting

9
Wide Area Network (WAN) options

• ATM
• SONET
• TCP/IP
• WAN can gt 50 of DR OpEx costs

10
ATMAsynchronous Transfer Mode

• Pros
• High performance
• OC3 to OC48
• 155Mbps to 2.5Gbps
• Shared network (cell based)
• IP over ATM
• Excellent QoS
• Available from most telcos
• High bandwidth utilization

• Cons
• Bandwidth overhead
• Niche technology
• Out-of-favor
• Disappearing
• Appears to be Dead-End
• High cost

11
SONET/SDHSynchronous Optical Network/Synchronous
Digital Hierarchy

• Pros
• High performance
• OC3 to OC192
• 155Mbps to 10Gbps
• Preferred by most telcos
• Can be shared
• TCP/IP switch/routers
• CWDM technology
• DWDM technology
• POS (IP packet over SONET)
• Very high bandwidth utilization

• Cons
• Expensive
• Although declining
• Not shareable natively
• Not a LAN technology
• Separate mgt from LAN

12
TCP/IPTransmission Control Protocol/Internet
Protocol

• Pros
• Ubiquitous
• Available everywhere
• Well known mgt
• Large knowledge pool
• Shared network
• Std network for most orgs.
• Can piggyback on IP WAN
• DR WAN perceived as free

• Cons
• Designed for packet loss
• Typical 1
• Packet loss retransmissions
• Congestion
• Bit Error Rates
• Jitter
• Latency
• Router buffer overruns
• Packet loss low throughput

13
Calculating storage WAN bandwidth

• How much data between application sites?
• And the DR site
• Over what period of time to move the data?
• Will the bandwidth be shared?
• If so, how much bandwidth is available?
• What type of WAN?
• Native ATM
• Native SONET
• TCP/IP

14
Assumptions

• ATM
• gt 80 bandwidth utilization
• IP over ATM
• And native ATM end-to-end
• SONET/SDH
• a.k.a. clear channel
• gt 90 bandwidth utilization
• Primarily POS (IP over SONET)
• Or FCBB (FC over SONET)

• TCP/IP
• End-to-end
• Packet loss avg 1
• lt 30 bandwidth utilization
• Worsens w/distance
• Worsens w/gt packet loss
• For calculations 30

15
Market trends

• In a poll of over 200 end users
• From SMB, SME, Enterprise
• 61 DR over TCP WANs
• 3 DR over SONET
• 1 DR over ATM
• 24 No DR over WAN
• 11 Both TCP SONET or ATM

16
What WAN do you use today for data protection?

• TCP/IP
• SONET
• ATM
• TCP SONET or ATM
• None of the above

17
Why TCP/IP WANs are so prevalent with DR

• Perception that the bandwidth is free
• Or at least very inexpensive
• Piggyback on IP WAN networks
• Evenings and Weekends

18
Most DR apps primarily USE IP

• Asynch mirroring
• Snapshot
• Volume replication
• Distributed backup
• Incremental
• Replication or Backup

• Tape Vaulting
• Continuous replication
• Continuous snapshot
• Fibre Channel over WAN
• FCIP
• iFCP
• Although there is FCBB
• FC over SONET

TCP/IP
19
Reality checkWhy TCP/IP WAN throughput is dismal

• TCP/IP
• Byte-streaming protocol moving data in small
  packets
• Retransmits the data from the last point of the
  error
• Immediately reduces the rate
• Backs down to slow start mode
• Additional ramp-up packet loss causes further
  rate reduction
• During periods of lossy conditions
• Application performance never has a chance to
  recover
• Why packet loss is so detrimental to TCP
  throughput

20
TCP 1985 Designed for LANs

• TCP Slow Start
• Packet rate 2X
• Per successful R-T
• Per Loss Event
• Sending rate cut 1/2
• TCP Congestion Control
• Sending rate gt 1
• Per successful R-T

21
TCP 2004 LAN protocol over the WAN

• Same internal logic
• Since 1985!
• High BW Loss events
• large packet losses
• High Latency
• Slower recovery
• During congestion control
• Infrequent feedback
• Changing route conditions
• Based on packet loss events

22
TCP resource contention on shared linksreduces
data protection throughput

• Sporadic packet loss
• Short long distance sessions
• Contend for same resources
• Router queues change dynamically
• From traffic bursts

3
23
TCP What really happens to long distance
sessions?

• Packet loss events
• Frequent for shared nets
• Loss events
• Router buffer overruns
• Affect other sessions
• Lots of lost packets
• LD sessions beat down
• By SD sessions
• Results
• Low throughput
• Random delays

24
The DR TCP WAN disconnect

• As distance gt, performance lt
• Worse with higher bandwidth

25
DR TCP/IP conclusion

• Perception reality do not match
• Must be taken into account
• When building a Storage WAN for Enterprise DR
• Increasing the bandwidth doesnt solve the problem

26
Now what?

• What happens when
• Throughput is much less than usable bandwidth?
• Time windows cant be met?
• The IP WAN is insufficient?
• Throwing more bandwidth at it fails to resolve
  problem?

27
There are 2 choices

• Go ATM end-to-end
• Not very palatable to most end users
• OR TCP enhancers
• Proxies
• Compressors
• Caching/spoofing
• Accelerators

28
Different clever technologies

• TCP/IP performance enhancing proxy
• Eliminates TCP packet loss latency issues
• Compression
• Increases payloads per packet
• Compression increases from 2X to 400X
• Caching (a.k.a. spoofing)
• Acknowledges packets locally
• Accelerators
• Resequencing, QoS, concatenation, duplication
  elimination
• Chatty protocol elimination

29
TCP/IP network shielding
Bit Error Rates
Network Jitter

• Shields TCP/IP network
• Bit error rates
• Congestion
• Jitter
• Latency
• Buffer overflows
• Much gt BW utilization!
• Before compression

TCP/IP Latency
Router buffer Overflows
Network Congestion
Data protection packets in a TCP/IP network
30
gt DS3 TCP/IP performance enhancements

• NetEx - HyperIP
• Orbital Data - IP Express

31
lt DS3 TCP/IP performance enhancements

• Expand - IP Accelerators
• 1800/4800/6800/9000 series
• Peribit
• SR20/50/55/80
• Net Celera
• T Series
• River Bed - Steel Head
• 500/1K/2K/3K/5K
• Orbital Data
• IP Express LC

32
Caching appliances

• River Bed - Steel Head 500/1K/2K/3K/5K
• CIFS MAPI (NFS coming)
• Tacit - Ishared Server
• CIFS NFS
• Kashya - KBX4000
• Includes volume replication, snapshot,
  mirroring
• File block replication

33
Storage WAN for enterprise DR TCP enhancement
caveats

• Needs TCP enhancement
• Packet loss is an issue
• Long distance
• Big bandwidth
• Large amounts of data
• Data migration
• Volume replication
• Snapshots
• High IOPS
• Bulk data transfers

• May not need it
• Incremental data
• Only changed data
• Short time for net new data
• Asynchronous mirroring
• Short distance
• Small bandwidth

34
Other issues to weigh

• Shared WAN
• Dedicated WAN

• Shared cost mgt.
• VLANs important
• Dedicated cost mgt.
• More flexibility

35
Summary and conclusions

• Build your DR foundation 1st
• Calculate DR throughput requirements
• Pick WAN technology of choice
• If TCP determine need for enhancement
• Implement
• Reassess quarterly