Requirements

Application Oriented

Wide-area Data Services (WDS)

Superset of Current Categories

Solves Remote Office Problems

Storage-oriented

• Resources have been deployed to the edge to deliver performance–at the cost of information control.
• User Want high performance access to corporate data
• IT wants to cut costs and needs to control / protect data

Remote office backup

• Consolidate application servers and tape to primary data center
• WAN Accelerators facilitate remote office access to centralized application servers
• Backup process consolidated to main data center

Disaster recovery with “normal” WAN (DS-3 or slower)

• Asynchronous replication only
• “Virtual Bandwidth” through SDR
• Cost-effective, use your existing WAN

Disaster recovery with “high-speed” WAN (OC-3 to OC-12)

• For replicating >1TB of data
• Synchronous or asynch replication (depends on distance)
• Fill the pipe with high-speed TCP through put

Remote office data backup

• Protect data hosted on all servers and desktops at the remote site
• No professional IT staff at remote site
• Difficult to ensure consistent reliable backup execution

Data replication to DR site

• Ensure a single cataclysmic event will not destroy corporate data repository
• Create and maintain alternate copy of corporate data at DR site
• Fast data transfer speeds are critical
• Increases flexibility and frequency of replication transfers
• Allows greater amounts of data to be replicated

Datacenter

• Tape Backup
• Storage
• File Servers
• Web Servers
• Mail Servers
• Filers

Network-Centric

Problems:

• Need “More Bandwidth”
• Latency Kills Throughput
• Application Protocol Chattiness and Latency
• TCP Chattiness and Latency
• Limited WAN Bandwidth

Solutions Focus

• Network Compression
• QoS
• Measurement & Reporting
• TCP Optimization

Storage-Centric

• Site Consolidation – Servers/Storage
• Connect Islands of Storage
• Remote Site Backup

Solutions

• Copies of Files Locally
• “WAFS” (File caching)

Application-Centric

• Web, Email, FTP, Database, etc. slow Response Times
• Apps Often Not Used… Too Slow

Solutions

• Speed up Specific Apps
• “Caching or Protocol Accelerators”
• Web/Dynamic Caching
• Mail Caching

Networking-oriented

• Deployed in Remote Sites and Data Center—peer-to-peer architecture
  • Preserve client-server relationships
• Intercept and accelerate TCP traffic (CIFS, NFS, HTTP, FTP, MAPI, MS-SQL, etc.)
  • Byte-level caching
  • Best-in-class compression / redundancy elimination
  • Latency optimization
  • Predictive traffic analysis

Solution

• Eliminate WAN performance bottlenecks
• Accelerate applications designed for LANs to work well over high latency WAN links
• Enable site consolidation of servers & storage from RBOs to the datacenter
• Speed up backup & replication over WANs and help eliminate RBO backup
• Install completely transparent to clients, servers, and most importantly…users!

Features, Functions, and Benefits

Scalable Data Referencing (SDR)

• Bandwidth optimization
• An arbitrarily large amount of data can be represented by one reference.
• Central server controls access to files, hence consistency is guaranteed
• One segment can belong to many files
• Entire files can be represented by a single label
• Changes to files are easily incorporated into the label hierarchy

Virtual TCP window expansion

TCP latency optimization

• Send large file across the WAN (via WAN Friendly Protocol like FTP, HTTP, other)
• With unlimited bandwidth and cross country latency the data transfer would take 60 seconds due to the TCP based round trips

VWE — TCP with compression appliances

• With unlimited bandwidth and cross country latency the data transfer would still take 60 seconds due to the TCP based round trips
• Increases utilization of high-bandwidth / high-latency links

Compress data at the packet level, below TCP layer.

Each packet consumes less, bandwidth, but still bound by TCP latency

• TCP Window Size
• WAN Accelerator Virtual Window Size
• Latency impact before effect of transaction prediction
• (Compressing/Suppressing traffic before filling the TCP flow control window)
• 1000 Mbps, 8MB Bulk Transfer, Averaged over 5 tests, Warm WAN Accelerator

Time

• Long Fat Networks (LFNs) are very high bandwidth links that are often difficult to ‘fill’ due to existing TCP flow control
• “T” represents the actual average throughput – often a fraction of the full bandwidth
• Caused by slow start & congestion control algorithms
• Blue line represents throughput after EFS WAN Accelerator appliance deployment
• Congestion Control Protocol Dynamics are improved
• Slower fall back / faster ramp up

Transparent pre-Population of File Data

• SDR does great for data seen before, but less for data not seen, i.e. cold transfers
• RCU pre-populates new data to branch office WAN Accelerator before user ever asks for it
• End result: LAN-like performance even for data that is accessed the first time, always hot
• Transparent Pre-Production – CIFs File Sharing

Transaction prediction

• Application protocol latency optimization

Transparent pre-population

• Improve performance for “cold” data