760 likes | 951 Views
Overview of Presentation. Background Historical evolution of Keck remote observing The Keck Remote Observing Model Remote observing from Waimea and CaliforniaRedirecting displays to remote locations Using X protocol Using VNC protocol Using a mix of both protocolsIssues and interactions betw
E N D
1. Optimizing the use of X and VNC protocols for support of remote observing Robert Kibrick and Steven L. Allen
University of California Observatories / Lick Observatory
Al Conrad, Terry Stickel, and
Gregory D. Wirth
W.M. Keck Observatory
Advanced Software, Control, and Communications Systems for Astronomy
2. Overview of Presentation Background
Historical evolution of Keck remote observing
The Keck Remote Observing Model
Remote observing from Waimea and California
Redirecting displays to remote locations
Using X protocol
Using VNC protocol
Using a mix of both protocols
Issues and interactions between X and VNC
Measurements of remote performance
Future directions
3. The Keck Telescopes
4. From 1993 to 1995, all Keck observing was done at the summit Observers at the summit work remotely from control rooms located adjacent to the telescope domes
5. Conducting observations involves coordinated effort by 3 groups
Telescope operator (observing assistant)
Responsible for telescope safety & operation
Keck employee; normally works at summit
Instrument scientist (support astronomer)
Expert in operation of specific instruments
Keck employee; works at summit or Waimea
Observers
Select objects and conduct observations
Employed by Caltech, UC, NASA, UH, or other
6. Keck 2 Control Room at the Mauna Kea Summit Telescope operator, instrument scientist, and observers work side by side, each at their own remote X Display
7. Observing at the Mauna Kea summit is both difficult and risky
Oxygen is only 60% of that at sea level
Lack of oxygen reduces alertness
Observing efficiency significantly impaired
Altitude sickness afflicts some observers
Some are not even permitted on summit:
Pregnant women
Those with heart or lung problems
8. Initiative to support remote observing from Keck Headquarters
1995: Remote control rooms built at Keck HQ
1996: Remote observing with Keck 1 begins
1997: >50% of Keck 1 observing done remotely
1999: remote observing >90% for Keck 1 and 2
2000: remote observing became the default mode
9. Keck 2 Remote Control Room at the Keck Headquarters in Waimea
Observer and instrument scientist in Waimea use video conferencing system to interact with telescope operator at the summit
10. The initial model for Keck Remote Observing All observing applications run on summit control computers
All displays are re-directed to display hosts at each site
11. Why did we choose this approach? Operational Simplicity
Operational control software runs only at the summit
All users run identical software on same computer
Simplifies management at each site
Allowed us to focus on commonality
Different sites / teams developed instrument software
Large variety of languages and protocols were used
BUT: all instruments used X-based GUIs
Legacy GUI applications (e.g., not web-based)
12. Limitations of Remote Observing from Keck HQ in Waimea
Most Keck observers live on the mainland.
Mainland observers fly > 3,200 km to get to Waimea
Collective direct travel costs exceed $400,000 U.S. / year
13. Keck Telescopes use Classical Scheduling
Kecks not designed for queue scheduling
Schedules cover a semester (6 months)
Approved proposals get 1 or more runs
Each run is between 0.5 to 5 nights long
Gaps between runs vary from days to months
Half of all runs are either half-night or 1 night long
14. Remote Observing from Waimea is not cost effective for short runs
Round trip travel time is 2 days
Travel costs > $1,000 U.S. per observer
About 50% of runs are for 1 night or less
Cost / run is very high for such short runs
Such costs limit student participation
17. Santa Cruz Remote Observing Facility Remote observer in California uses video conferencing system to interact with colleague in Waimea
18. Redirecting displays to remote sites using X protocol and ssh An ssh tunnel is used to relay X protocol packets and authentication across network firewalls to the remote site.
19. Benefits of using ssh to forward X displays Tunnels through firewalls
Secure forwarding of X protocol
Transparent forwarding of X authentication
Transparent redirection of the X DISPLAY
Efficient in-stream data compression
20. Limitations of using X to redirect displays to the mainland Sluggish performance for some operations
very slow application startup compared to Waimea
slow creation of new windows and pulldown menus
guider display update rate is too low
Inability to share “single-user” applications
figdisp realtime image display (LRIS, HIRES, ESI)
various data reduction packages
Some applications sensitive to inter-site font variations
21. Factors contributing to sluggish performance at mainland sites Lower inherent bandwidth (Mauna Kea to Oahu)
High round trip time (RTT) between Keck and California
Yields lower effective bandwidth for un-tuned TCP
Slows any functions requiring multiple round trips
High RTT limits benefit of tuning or compression
Propagation delays not likely to improve over time
As networks grow, hops and RTTs may increase
Keck/California link: bandwidth / hops / RTT
1998: 1.5 Mbs / 12 hops / 70 millisecond average RTT
2004: 35 Mbs / 22 hops / 100 millisecond average RTT
22. Strategies for coping with lower bandwidth and long RTTs Use compression to compensate for lower bandwidth
Low-bandwidth X (LBX)? Inferior to ssh
Differential X protocol converter (dxpc) Insecure
ssh with compression enabled authentication
Use TCP tuning to increase effective bandwidth
Minimize operations that require multiple round trips
Start up all needed X clients at start of observing session
Pop up and then iconify all GUI subpanels at session start
Rely on backing store in user’s X server for rapid re-instantiation
Use “tear off” pulldowns for commonly used pulldown menus
Look for ways to eliminate long round trip transactions
23. Virtual Network Computing (VNC) VNC server provides shareable virtual desktop
VNC clients (viewer) offer remote access to that desktop
All clients share that desktop (application sharing)
All state is retained in the server, none in the clients
Clients can connect/disconnect without affecting session
VNC protocol works at the framebuffer level
VNC available on most OS / windowing systems
24. Redirecting displays to remote sites using VNC protocol & ssh An ssh port-forwarding tunnel is used to relay VNC protocol packets and authentication across network firewalls to the remote site.
25. Benefits of using VNC Single-user applications can be shared between sites
Shared desktop promotes training and collaboration
Shared desktop persists even if remote VNC client crashes
Optional read-only sharing for “look but don’t touch”
X clients connect to a local X server (short RTTs)
X client applications run on control computer at summit
VNC server runs on computer at the summit
X clients see the VNC server as their local X server
Speeds up client functions that require multiple X transactions
Application startup and initial painting of displays
Creation of pop-up windows and sub-panels
Instantiation of pull-down menus
Loading of fonts
26. Disadvantages of using VNC Some operations are slower across a low bandwidth link
iconify / de-iconify operations are very slow (no backing store)
color map scrolling is slower than under a native X connection
large image pan / zoom operations twice as slow as native X
Ssh has no built-in capability for forwarding VNC packets
must start port-forwarding tunnel (ssh –L) before VNC viewer
several-minute TCP timeout if tunnel must be re-established
Shared desktop is sometimes a source of user confusion
Keyboard / mouse input from all connected clients are merged
Users at different sites could type or move mouse at same time
Potential for multiple users to create conflicting inputs
27. An alternative topology for using VNC with ssh and X The VNC server and VNC viewer are both run on the same computer at the summit. The X display generated by the VNC viewer is re-directed to the remote site via a standard ssh tunnel.
28. Benefits of this topology
Retains most benefits of the first VNC topology:
Single-user applications can be shared between sites
Observing X clients connect to a local X server (short RTTs)
Speeds up functions that require multiple X transactions
Improved performance for ds9 zoom in / out
VNC not needed at remote sites; only at the summit
Avoids ssh port-forward complexity and TCP timeout
29. Neither protocol is optimal for all applications & sites
Some functions work better under X
image display functions (pan, zoom in / out, color map scroll)
iconifying / de-iconifying windows (if backing store enabled)
any functions more sensitive to bandwidth than to RTT
Some functions work better under VNC
creation of new windows, pop-ups, and sub-panels
instantiation of pull-down menus
any applications that are RTT sensitive (Keck guider eavesdrop)
Most output-only applications work OK using either
Using a mix of both protocols
30. Multiple monitors facilitates a mix of X and VNC protocols For example, one monitor can display a shared VNC desktop while the others carry the redirected X displays of X clients running at the summit.
31.
X Visuals:
Depth: 8-, 16-, and 24-bit
PseudoColor, StaticColor, and TrueColor
Many X servers support multiple X visuals
VNC server supports only 1 visual at a time
VNC server must satisfy least capable client
Legacy X clients need 8-bit PseudoColor Issues and Interactions between VNC and X
32.
Colormap issues
slower scrolling if pixel retransmit needed
colormap flashing if insufficient colors
private color maps
Whitepixel and blackpixel conflicts
Fonts are supplied by the X server
Using X model, X server is local to observer
Using VNC model, X server is at summit Issues and Interactions between VNC and X
33. Local ds9 X client and server: 6.6 seconds
Summit ds9 display redirected to mainland
direct to mainland X server: 72.0 (76.0*)
via mainland VNC viewer: 11.0 (27.5*)
via VNC viewer at summit: 10.6 (26.2*)
* values in ( ) are without ssh compression
In-stream compression helps in all cases
Measurements of remote performance: ds9 startup
34. Local ds9 X client and server: 3.0 seconds
Summit ds9 display redirected to mainland
direct to mainland X server: 10.3 (11.9*)
via mainland VNC viewer: 3.0 ( 7.7*)
via VNC viewer at summit: 3.0 ( 6.8*)
* values in ( ) are without ssh compression
In-stream compression helps in all cases
Measurements of remote performance: file chooser popup
35. Local ds9 X client and server: 0.7 seconds
Summit ds9 display redirected to mainland
direct to mainland X server: 4.0 (10.2*)
via mainland VNC viewer: 9.0 (17.0*)
via VNC viewer at summit: 6.5 (11.5*)
* values in ( ) are without ssh compression
In-stream compression helps in all cases
Measurements of remote performance: ds9 zoom in
36. Local ds9 X client and server: 0.0 seconds
Summit ds9 display redirected to mainland
direct to mainland X server: 0.1 ( 3.0*)
via mainland VNC viewer: 4.0 (10.0*)
via VNC viewer at summit: 4.5 (13.5*)
* values in ( ) are without ssh compression
In-stream compression helps in all cases
Measurements of remote performance: ds9 draw cut for plot
37.
Can redirect displays of legacy X clients:
using X protocol
using VNC protocol
using a combination of both protocols
Choice depends on functionality of each X client
Good remote performance for most X GUI clients
Need better remote performance for ds9
Summary
38. Future directions Explore distributed ds9-like image displays:
“ds9server”: an image / image section server
Runs on the instrument control computer on summit
Interfaces to multi-HDU FITS images on disk or in shmem
Extracts subset of pixels requested by display client(s)
Efficiently transmits pixels and WCS-info to display client(s)
“ds9viewer”: an image / image section client
Runs at remote site and provides local GUI to observer
Convert GUI events into requests to transmit to ds9server
Receives pixel / WCS-info stream transmitted by ds9server
Displays pixel stream locally as a bitmap image or plot
39. Future directions Explore distributed ds9-like image displays:
design image server / viewer protocol to:
Minimize round trips between client(s) and server
Support progressive/lossy transmission of image sections
Enable support of a web-based client/viewer
Enable support of a X-based client/viewer
challenges: ds9 live plots, magnifier window
Test protocol with NIST Network Emulation Tool
Simulates network latency, bandwidth, jitter
http://www.antd.nist.gov/tools/nistnet/index.html
40. Mainland remote observing (MRO) is operational
For all Keck optical optical instruments
From 3 mainland sites (UCSC, UCSD, CIT)
A mix of X and VNC protocols is used
Specific mix depends on instrument being used
Provides competitive performance for most GUIs
Provides tolerable performance for image displays
MRO efficiency would be enhanced by:
A distributed image display server / client
More optimal TCP tuning between sites Conclusion
41. U.S. National Science Foundation
University of Hawaii
Gemini Telescope Consortium
University Corp. for Advanced Internet Development (UCAID)
Corporation for Education Network Initiatives in California (CENIC) Acknowledgments
42.
Robert Kibrick, UCO/Lick Observatory
University of California, Santa Cruz
California 95064, U.S.A.
E-mail: kibrick@ucolick.org
WWW: http://www.ucolick.org/~kibrick
Phone: +1-831-459-2262
FAX: +1-831-459-2298 Author Information
43. END OF PRESENTATION !!!
SPARE SLIDES FOLLOW
44. Keck 2 Remote Observing Room as seen from the Keck 2 summit
Telescope operators at the summit converse with astronomer at Keck HQ in Waimea via the videoconferencing system.
45. Videoconferencing has proved vital for remote observing from Waimea Visual cues (body language) important!
Improved audio quality extremely valuable
A picture is often worth a thousand words
Troubleshooting: live oscilloscope images
“Cheap” desktop sharing (LCD screens)
Chose dedicated versus PC-based units:
Original (1996) system was PictureTel 2000
Upgrading to Polycom Viewstations
46. Interaction between video-conferencing and type of monitors Compression techniques motion sensitive
“Moving” scene requires more bandwidth
CRT monitors cause “flicker” in VC image
Beating of frequencies: camera .vs. CRT
CRT phosphor intensity peaking, persistence
CRT monitor “flicker” causes problems:
Wastes bandwidth and degrades resolution
Visually annoying / nausea inducing
Use LCD monitors to avoid this problem
47. The Keck Headquarters in Waimea
Most Keck technical staff live and work in Waimea. Allows direct contact between observers and staff. Visiting Scientist’s Quarters (VSQ) located in same complex.
48. Motivations for Remote Observing from the U.S. Mainland
Travel time and costs greatly reduced
Travel restrictions accommodated
Sinus infections and ruptured ear drums
Late stages of pregnancy
Increased options for:
Student participation in observing runs
Large observing teams with small budgets
Capability for remote engineering support
49. Santa Cruz Remote Observing Video Conferencing Remote observer’s colleague in Waimea as seen from Santa Cruz remote ops
50. The Weather in Waimea Remote observer in California points remotely-controlled camera at the window in Waimea remote ops
51. The Remote Observing Facility at Keck Headquarters in Waimea
Elevation of Waimea is 800 meters
Adequate oxygen for alertness
Waimea is 32 km NW of Mauna Kea
45 Mbps fiber optic link connects 2 sites
A remote control room for each telescope
Videoconferencing for each telescope
On-site dormitories for daytime sleeping
52. Mainland remote observing goals and implementation strategy
Goals:
Target mainland facility to short duration runs
Avoid duplicating expensive Waimea resources
Avoid overloading Waimea support staff
Strategy:
No mainland dormitories; observers sleep at home
Access existing Waimea support staff remotely
Restrict mainland facility to experienced observers
Restrict to mature, fully-debugged instruments
53. Mainland remote observing facility is an extension of Keck HQ facility Only modest hardware investment needed:
Workstations for mainland remote observers
Network-based videoconferencing system
Routers and firewalls
Backup power (UPS) – especially in California!!!
Backup network path to Mauna Kea and Waimea
Avoids expensive duplication of resources
Share existing resources wherever possible
Internet-2 link to the mainland
Keck support staff and operational software
54. The initial model for Keck Remote Observing The control computers at the summit:
Each telescope and instrument has its own computer
All operational software runs only on these computers
All observing data written to directly-attached disks
Users access data disks remotely via NFS or ssh/scp
The display workstations
Telescopes and instruments controlled via X GUIs
All users access these X GUIs via remote X displays
X Client software runs on summit control computers
Displays to X server on remote display workstation
55. Operational simplifications
Only one copy of operational software to maintain
Only “vanilla” hardware / software needed at remote site
Simplifies sparing and swapping of equipment
Simplifies system maintenance at remote site
Simplifies authentication / access control
56. Focus effort on X standardization and optimization over long links Maintain consistent X environment between sites
Optimize X performance between sites
Eliminates need to maintain:
Diverse instrument software at multiple sites
Diverse telescope software at multiple sites
Coordinate users accounts at multiple sites
Fewer protocols for firewalls to manage
57. Remote observing differences: Waimea versus the mainland System Management:
Keck summit and HQ share a common domain
Mainland sites are autonomous
Remote File Access:
Observers at Keck HQ access summit data via NFS
Observers on mainland access data via ssh/scp
Propagation Delays:
Summit to Waimea round trip time is about 1 ms.
Summit to mainland round trip time is about 100 ms.
58. Shared access and control of instruments Most software for Keck optical instruments provides native multi-user/multi-site control
All users have consistent view of status and data
Instrument control can be shared between sites
Multipoint video conferencing key to coordination
Some single-user applications can be shared via X-based application sharing environments:
XMX http://www.cs.brown.edu/software/xmx
VNC http://www.uk.research.att.com/vnc
59. Increased propagation delay to mainland presents challenges Initial painting of windows is much slower
But once created, window updates fast enough
All Keck applications display to Waimea OK
A few applications display too slowly to mainland
System and application tuning very important
TCP window-size parameter (Web100 Initiative)
X server memory and backing store
Minimize operations requiring round trip transactions
60. Shared access and control of instruments Most software for Keck optical instruments provides native multi-user/multi-site control
All users have consistent view of status and data
Instrument control can be shared between sites
Multipoint video conferencing key to coordination
Some single-user applications can be shared via X-based application sharing environments:
XMX http://www.cs.brown.edu/software/xmx
VNC http://www.uk.research.att.com/vnc
61. Fast and reliable network needed for mainland remote observing 1997: 1.5 Mbps Hawaii -> Oahu -> mainland
1998: 10 Mbps from Oahu to mainland
1999: First phase of Internet-2 upgrades:
45 Mbps commodity link Oahu -> mainland
45 Mbps Internet-2 link Oahu -> mainland
2000: Second phase upgrade:
35 Mbps Internet-2 link from Hawaii -> Oahu
Now 35 Mbps peak from Mauna Kea to mainland
2002: 155 Mbs from Oahu to mainland
62. End-to-end reliability is critical to successful remote operation Keck Telescope time is valued at $1 per second
Observers won’t use facility if not reliable
Each observer gets only a few nights each year
What happens if network link to mainland fails?
Path from Mauna Kea to mainland is long & complex
At least 14 hops crossing 6 different network domains
While outages are rare, consequences are severe
Even brief outages cause session collapse & panic
Observing time loss can extend beyond outage
The real question is, what happens if the link to the mainland fails?The real question is, what happens if the link to the mainland fails?
63. Keck Observatory policy on mainland remote observing
If no backup data path is available from mainland site, at least one member of observing team must be in Waimea
Backup data path must be proven to work before mainland remote observing is permitted without no team members in Waimea
64. Mitigation plan: install end-to-end ISDN-based fall-back path Install ISDN lines and routers at:
Each mainland remote observing site
Keck 1 and Keck 2 control rooms
Fail-over and fall-back are rapid and automatic
Toll charges incurred only during network outage
Lower ISDN bandwidth reduces efficiency, but:
Observer retains control of observations
Sessions remain connected and restarts avoided
Prevents observer panic
65. Summary of ISDN-based fallback path Install 3 ISDN lines (6 B channels) at each site
Install Cisco 2600-series routers at each end
Quad BRI interfaces
Inverse multiplexing
Caller ID (reject connections from unrecognized callers)
Multilink PPP with CHAP authentication
Dial-on-demand (bandwidth-on-demand)
No manual intervention needed at either end
Fail-over occurs automatically within 40 seconds
Uses GRE tunnels, static routes, OSPF routing
66. Running OSPF routing over aGRE tunnel
On each router, we configure 3 mechanisms:
A GRE tunnel to the other endpoint
A floating static route that routes all traffic to the other endpoint via the ISDN dialer interface
A private OSPF domain that runs over the tunnel
OSPF maintains its route through the tunnel only if the tunnel is “up”
OSPF dynamic routes take precedence over floating static route
67. Fail-over to ISDN backup data path
If the Internet-2 path is “up”, OSPF “hello” packets flow across the tunnel between routers
As long as “hello” packets flow, OSPF maintains the dynamic route, so traffic flows through tunnel
If Internet-2 path is “down”, OSPF “hello” packets stop flowing, and OSPF deletes dynamic route
With dynamic route gone, floating static route is enabled, so traffic flows through ISDN lines
68. The hardest problem was the lodging! LRIS operated from Santa Cruz all 5 nights
ISDN backup path activated several times
Observing efficiency comparable to Waimea
Lodging was the biggest problem
Motel check-in/check-out times incompatible
Required booking two motels for the same night
Motels are not a quiet place for daytime sleep
69. OSPF keeps trying to send “hello” packets through the tunnel, even with Internet is down
As long as Internet-2 path remains down the “hello” packets can’t get through
Once the Internet-2 path is restored, “hello” packets flow between routers
OSPF re-instates dynamic route through tunnel
All current traffic gets routed through the tunnel
All ISDN calls are terminated Fall-back to the normal Internet-2 path
70. Operational costs of ISDN backup data path
Fixed leased cost is $70 per line per month
Three lines at each site -> $2,500 per site/year
Both sites -> $5,000/year
Long distance cost (incurrent only when active)
$0.07 per B-channel per minute
If all 3 lines in use:
$0.42 per minute
$25.20 per hour
71. Recent operational experience Remote observing science from Santa Cruz:
Low Resolution Imaging Spectrograph (LRIS)
Echellete Spectrograph and Imager (ESI)
Remote engineering and instrument support
ESI
High Resolution Echelle Spectrometer (HIRES)
Remote Commissioning Support
ESI
DEIMOS (see SPIE paper 4841-155 & 4841-186)
72. Unplanned use of the facility during week of Sept. 11, 2001 All U.S. commercial air traffic grounded
Caltech astronomers have a 5-day LRIS run on Keck-I Telescope starting September 13
No flights available
Caltech team leaves Pasadena morning of 9/13
Drives to Santa Cruz, arriving late afternoon
Online with LRIS well before sunset in Hawaii
73. Extending mainland remote observing to other sites Other sites motivated by Santa Cruz success
Caltech remote facility is nearly operational
Equipment acquired
ISDN lines and router installed
Will be operational once routers are configured
U.C. San Diego facility being assembled
Equipment specified and orders in progress
Other U.C. campuses considering plans
74. Administrative challenges: scheduling shared facilities Currently only one ISDN router at Mauna Kea
Limits mainland operation to one site per night
Interim administrative solution
Longer term solution may require:
Installation of additional ISDN lines at Mauna Kea
Installation of an additional router at Mauna Kea
75. Remaining challenges TCP/IP tuning of end-point machines
Needed to achieve optimal performance
Conflicts with using “off-the-shelf” workstations
Conflict between optimal TCP/IP parameters for the normal Internet-2 path .vs. the ISDN fall-back path
Hoping for vendor-supplied auto-tuning
Following research efforts of Web100 Project
Administrative challenges
Mainland sites are currently autonomous
Need to develop coordination with Keck
76. Internet-2 makes mainland operation feasible
Backup data path protects against interruptions
Keck HQ is the central hub for remote operation
Mainland remote observing model is affordable:
Mainland sites operate as satellites of Keck HQ
Leverage investment in existing facilities and staff
Leverage investment in existing software
Share existing resources wherever feasible
Avoid expensive and inefficient travel for short runs
Model is being extended to multiple sites
Summary