1. Improved data capacity using bandwidth acceleration in HiSeasNet Steve Foley, Jon Berger, John Orcutt, Frank Vernon Scripps Institution of Oceanography Steve Foley, Jon Berger, John Orcutt, Frank Vernon Scripps Institution of Oceanography December 14, 2009 AGU Fall Meeting

2. HiSeasNet Overview  Internet connectivity to University-National Oceanographic Laboratory System (UNOLS) research vessels via satellite  Low-bandwidth, high-latency, always-on  15 Intermediate-class and larger vessels  7 Full ocean for larger vessels (C-band)  8 North America coastal for smaller vessels (Ku- band)  Program began in 2002 with one ship, added earth station, ships, and bandwidth ever since  Program operated by Scripps Institution of Oceanography (SIO)  Earth station supports four satellite beams  Internet connectivity to University-National Oceanographic Laboratory System (UNOLS) research vessels via satellite  Low-bandwidth, high-latency, always-on  15 Intermediate-class and larger vessels  7 Full ocean for larger vessels (C-band)  8 North America coastal for smaller vessels (Ku- band)  Program began in 2002 with one ship, added earth station, ships, and bandwidth ever since  Program operated by Scripps Institution of Oceanography (SIO)  Earth station supports four satellite beams

3. HiSeasNet Technology  Uses marine-stabilized antennas  2.4m dishes for larger vessels (C-band)  1m-1.5m dishes for smaller vessels (Ku-band)  Connectivity is all IP based  64kbps to 96kbps ship-to-shore  Shared shore-to-ship links between 192kbps (3 ships) to 256kbps (5 ships)  Allows for flexibility of any type of traffic to be sent (e-mail, web, FTP, SSH, IM, VoIP, etc.)  Uses marine-stabilized antennas  2.4m dishes for larger vessels (C-band)  1m-1.5m dishes for smaller vessels (Ku-band)  Connectivity is all IP based  64kbps to 96kbps ship-to-shore  Shared shore-to-ship links between 192kbps (3 ships) to 256kbps (5 ships)  Allows for flexibility of any type of traffic to be sent (e-mail, web, FTP, SSH, IM, VoIP, etc.)

4. HiSeasNet Challenges  Users are demanding more services when they are at sea  Voice-over-IP  Video conferencing for outreach  Large data downloads of daily/hourly satellite images  Real-time collaboration with other scientists/ships  Streaming data to shore in real time  Satellite bandwidth can be expensive!  Coastal ship coverage is $750/mo, global ship coverage is $3000/mo  Bulk of the traffic (~90%) is shore-to- ship…the expensive direction  Users are demanding more services when they are at sea  Voice-over-IP  Video conferencing for outreach  Large data downloads of daily/hourly satellite images  Real-time collaboration with other scientists/ships  Streaming data to shore in real time  Satellite bandwidth can be expensive!  Coastal ship coverage is $750/mo, global ship coverage is $3000/mo  Bulk of the traffic (~90%) is shore-to- ship…the expensive direction

5. How can we increase value?  When the dollars run out, what have we done to improve user experience over a fixed sized, high-delay data pipe?  Use the right satellite encoding  Share the expensive (shore-to-ship) route between ships on-the-fly  Quality of Service (QoS) to keep shared link fair  QoS to manage critical/real-time traffic differently than bulk/non-real-time traffic  Cache common data (Web, DNS, FTP, etc.)  Compress data  Can we use network acceleration appliances?  When the dollars run out, what have we done to improve user experience over a fixed sized, high-delay data pipe?  Use the right satellite encoding  Share the expensive (shore-to-ship) route between ships on-the-fly  Quality of Service (QoS) to keep shared link fair  QoS to manage critical/real-time traffic differently than bulk/non-real-time traffic  Cache common data (Web, DNS, FTP, etc.)  Compress data  Can we use network acceleration appliances?

6. Network accelerator goals  Increase the amount of data that can be sent across an existing link  Ex: Push 384kbps across a 256kbps link  Allows for more data to be exchanged in a given amount of time  Decrease the round trip time across a link  Ex: Have a network response of 100ms instead of 550ms  Allows for quicker response times to applications/users  Faster connection setups mean better link efficiency and ultimately more data  Improve user experience  QoS to make applications get the data flows they need  Increase the amount of data that can be sent across an existing link  Ex: Push 384kbps across a 256kbps link  Allows for more data to be exchanged in a given amount of time  Decrease the round trip time across a link  Ex: Have a network response of 100ms instead of 550ms  Allows for quicker response times to applications/users  Faster connection setups mean better link efficiency and ultimately more data  Improve user experience  QoS to make applications get the data flows they need

7. Network accelerator methods  Open TCP sessions faster by buffering more setup instead of waiting for packet exchanges – link efficiency, user experience  Adjust TCP timers to efficiently pack long-fat satellite pipe (SCPS standard) – link efficiency  TCP packet caching (some do UDP, too) – data volume  Web / DNS caching – data volume  Header and payload compression on-the-fly – data volume  QoS filtering, and auto-fragmentation – user experience, link efficiency  NOTE: Does not improve compressed or encrypted packet payloads, just their headers  Open TCP sessions faster by buffering more setup instead of waiting for packet exchanges – link efficiency, user experience  Adjust TCP timers to efficiently pack long-fat satellite pipe (SCPS standard) – link efficiency  TCP packet caching (some do UDP, too) – data volume  Web / DNS caching – data volume  Header and payload compression on-the-fly – data volume  QoS filtering, and auto-fragmentation – user experience, link efficiency  NOTE: Does not improve compressed or encrypted packet payloads, just their headers

8. Accelerator appliances  Expand Networks 4830 and 4930 models  4930 units run squid web proxy on-board for smaller ships that don’t maintain one already  Supports TCP and UDP acceleration  Designed for low-bandwidth, high-latency sat links with SCPS compliance  Easy to install on-path  Appliance can be remotely managed from shore  Depending on data volume license, these are about $4k per box  Expand Networks 4830 and 4930 models  4930 units run squid web proxy on-board for smaller ships that don’t maintain one already  Supports TCP and UDP acceleration  Designed for low-bandwidth, high-latency sat links with SCPS compliance  Easy to install on-path  Appliance can be remotely managed from shore  Depending on data volume license, these are about $4k per box

9. Measuring accelerated data  Accelerators have two traffic categories:  “Raw” – what would be sent if there were no accelerator installed  “Accelerated” – what is actually sent on an interface, post acceleration  The difference between raw and accelerated data is the bandwidth improvement offered by the accelerator  Accelerators have two traffic categories:  “Raw” – what would be sent if there were no accelerator installed  “Accelerated” – what is actually sent on an interface, post acceleration  The difference between raw and accelerated data is the bandwidth improvement offered by the accelerator

10. Results  10 months of testing done on 2 C-band ships  Atlantis  Shore-to-ship: 13% improvement  Ship-to-shore: 17% improvement  Revelle  Shore-to-ship: 21% improvement  Ship-to-shore: 63% improvement  Bulk of total traffic (~90%) is shore-to-ship  Bulk of shore-to-ship traffic is web  Ship-to-shore traffic is scattered, mostly web requests, mail, data, IM  Roughly half of improvement is compression  Ships with web proxies have a 40-50% hit rate already  10 months of testing done on 2 C-band ships  Atlantis  Shore-to-ship: 13% improvement  Ship-to-shore: 17% improvement  Revelle  Shore-to-ship: 21% improvement  Ship-to-shore: 63% improvement  Bulk of total traffic (~90%) is shore-to-ship  Bulk of shore-to-ship traffic is web  Ship-to-shore traffic is scattered, mostly web requests, mail, data, IM  Roughly half of improvement is compression  Ships with web proxies have a 40-50% hit rate already

11. Results by application  HTTP from ship: 290%, to ship: 20%  Instant Messaging: 10-350%  Directional and based on how much video/audio gets passed along  General IM text chat seems ~35%  SSH: 0-35% (varies by content and crypto)  SMTP: 20-40%  POP3 from ship: 0%, to ship: 85%  IMAP from ship: 65%, to ship: 30%  Secure POP/IMAP/HTTP: 0% (encrypted)  FTP: 0% (usually compressed already)  DNS from ship: 15%, to ship: 40%  Streaming video from ship: 5%, to ship: 15%  Streaming data (uncompressed): 130%  HTTP from ship: 290%, to ship: 20%  Instant Messaging: 10-350%  Directional and based on how much video/audio gets passed along  General IM text chat seems ~35%  SSH: 0-35% (varies by content and crypto)  SMTP: 20-40%  POP3 from ship: 0%, to ship: 85%  IMAP from ship: 65%, to ship: 30%  Secure POP/IMAP/HTTP: 0% (encrypted)  FTP: 0% (usually compressed already)  DNS from ship: 15%, to ship: 40%  Streaming video from ship: 5%, to ship: 15%  Streaming data (uncompressed): 130%

12. Results caveats  Benefit depends heavily on traffic patterns, therefore it can vary with:  Use policies on a ship (imposed by operator)  Size of ship and science party  Science work being done  Cruise plan (weather/heading, port time, duration, etc.)  Human behavior  Traffic conditions vary between cruises and during cruises  HiSeasNet doesn’t inspect packets deeply  10 month average includes months when accelerator was still being tuned  Recent acceleration numbers are a bit better  Benefit depends heavily on traffic patterns, therefore it can vary with:  Use policies on a ship (imposed by operator)  Size of ship and science party  Science work being done  Cruise plan (weather/heading, port time, duration, etc.)  Human behavior  Traffic conditions vary between cruises and during cruises  HiSeasNet doesn’t inspect packets deeply  10 month average includes months when accelerator was still being tuned  Recent acceleration numbers are a bit better

13. Cost analysis  Initial investment:  15% of $3000 monthly C-band satellite lease against a $4k box is ROI of 9 months  Estimated: 30% of $750 monthly Ku-band satellite lease against a $4k box is ROI of 18 months  Hidden costs:  Earth station accelerator installation  Maintenance per year for ~20% for hardware and firmware support  Additional license needed with more bandwidth  Added hassle of one more box to troubleshoot/fail/blame  Benefits:  Retain improvement for years to come  Some added QoS for better experience  Initial investment:  15% of $3000 monthly C-band satellite lease against a $4k box is ROI of 9 months  Estimated: 30% of $750 monthly Ku-band satellite lease against a $4k box is ROI of 18 months  Hidden costs:  Earth station accelerator installation  Maintenance per year for ~20% for hardware and firmware support  Additional license needed with more bandwidth  Added hassle of one more box to troubleshoot/fail/blame  Benefits:  Retain improvement for years to come  Some added QoS for better experience

14. Future plans  Whole fleet is getting accelerators now  Installation is on-going over the next few months as ships go back to sea  So far data are encouraging, but most ships ships are not under normal operations right now  Smaller ships are getting accelerators with web proxies built-in  Tune accelerators better  Use extra bandwidth for:  More scientific data streams to shore in real-time  More web access  Smoother voice/video  Whole fleet is getting accelerators now  Installation is on-going over the next few months as ships go back to sea  So far data are encouraging, but most ships ships are not under normal operations right now  Smaller ships are getting accelerators with web proxies built-in  Tune accelerators better  Use extra bandwidth for:  More scientific data streams to shore in real-time  More web access  Smoother voice/video