Adaptable Network Striping

Flexible network striping l.jpg
1 / 26
0
0
1310 days ago, 394 views
PowerPoint PPT Presentation
. Restorative Emergencies. Because of the EMTs\' absence of medicinal preparing, specialists don't trust they can rely on upon EMTs for much past rapid transport.. . . Emergency911 callAmbulance arrivesEMTs\' introductory examinationTransport to hospitalDoctors start treatment. Remote determination and treatmentProvide propelled remote diagnostics capabilitiesAllow specialists to look at patients in-travel What we need to se

Presentation Transcript

Slide 1

Adaptable Network Striping Asfandyar Qureshi RQE: 21 st August 2006

Slide 2

Medical Emergencies Emergency 911 call Ambulance arrives EMTs' underlying examination Transport to doctor's facility Doctors start treatment Due to the EMTs' absence of therapeutic preparing, specialists don't trust they can rely on upon EMTs for much past rapid transport.

Slide 3

Mobile Telemedicine Remote determination and treatment Provide propelled remote diagnostics capacities Allow specialists to look at patients in-travel What we need to send Unidirectional VIDEO (~600 kbit/sec) Bidirectional AUDIO (~30 kbit/sec) Low-rate Physiological information (EKG, Heart-rate, and so forth)

Slide 4

System Deployment Plan to convey the framework so as to direct numerous therapeutic reviews Initial arrangement: Spring '07 Project accomplices Boston: Massachusetts General Hospital Orlando: Orange County Florida's Emergency Medical Services Economic requirements Limited/dynamic organization must be simple Initial medicinal review: two ambulances Cannot amortize the cost over numerous ambulances Use existing remote foundation

Slide 5

Talk Structure Motivation Wireless Wide Area Networks (WWANs) Performance diagram Horde: arrange striping middleware Network striping Horde API Horde internals Channel supervisors and transmission openings Packet planning

Slide 6

WWAN Performance Overview CDMA EV-DO interfaces US suppliers: Verizon and Sprint Low throughput Download < 400 kbits/sec Upload < 150 kbits/sec High and variable parcel RTTs 500 ms ± 200 ms Realistic WWAN execution information is difficult to find Lots of buildup/disinformation

Slide 7

WWAN Experiments We invested some energy running WWAN explores in Orlando and Boston Drove around in an auto Simultaneously utilized various interfaces Measured UDP throughput and RTTs Used GPS to infer scope maps Also logged 802.11 APs Not a thorough examination Nonetheless, gave a lot of knowledge into WWAN conduct

Slide 8

Throughput AGGREGATE mean = 306 kbps, stdev = 64 kbps SPRINT mean = 111 kbps, stdev = 36 kbps VERIZON-1 mean = 109 kbps, stdev = 21 kbps VERIZON-2 mean = 91 kbps, stdev = 32 kbps

Slide 9

High and variable RTTs Stationary pings (Boston)

Slide 10

Network Striping "Stripe" Application Data over Multiple Network Channels Simultaneously utilize many systems Wireless systems can be different and unsteady A B N organize stations M information streams M information streams

Slide 11

Striping: Related Work Wired systems e.g., n-ISDN lines as a virtual T1 line SIGCOMM '96: "A Reliable and Scalable Striping Protocol", ADISESHU, PARULKAR, AND VARGHESE. Generally worried with inspiring TCP to run well. Remote systems Globecom '99: "Versatile Inverse Multiplexing for Wide-Area Wireless Networks", SNOEREN. Mobisys 2004: "Deface: A Commuter Router Infrastructure for the Mobile Internet", RODRIGUEZ, CHAKRAVORTY, CHESTERFIELD, PRATT, AND BANERJEE. Multi-way video gushing 2001: "Lopsided Multiple Description Video Communication utilizing Path Diversity", APOSTOLOPOULOS AND WEE.

Slide 12

Challenges Network channels can be very disparate Channel QoS shifts essentially in time Number of channels changes Application Bandwidth constrained Different sorts of streams with divergent system QoS sensitivities Want applications to be free of the number and sorts of channels

Slide 13

Our approach: Horde Network striping middleware Exposes striping operation to applications Apps can conceptually adjust striping arrangement Flexible striping instrument Per-channel blockage control Does not bolster unmodified legacy applications Horde is focused at engineers of new portable applications Legacy support is generally simple to give… Virtual TUN interfaces TESLA (Salz et al, Usenix '03)

Slide 14

Forward Compatibility Applications don't have a short timeframe of realistic usability Depend on Horde's unique API The secluded outline of the middleware, permits our applications to be forward good Networks are probably going to develop. Sprint moving to WiMax Verizon moving to EV-DO rev An Our middleware is intended to all the while handle a wide range of sorts of systems and its usefulness can be broadened effortlessly.

Slide 15

Horde API: Sessions Each host runs a neighborhood Horde daemon Exposes a RPC-like interface to applications Before any information is sent, peering applications must arrange a named session. Different sessions (remarkable names) are permitted "Tavarua" "Tavarua" Application An Application B N organize channels Horde A Horde B Host B Host A

Slide 16

Horde API: Streams & ADUs Within a session, at least one streams can be arranged. Streams are (casual) bi-directional FIFOs of Application Data Units (ADUs). Most extreme reordering  (greatest number of channels) ADUs can be divided, in part lost, and so on. "Tavarua" "Tavarua" "video" "sound" Application An Application B "information" Horde A Horde B Host B Host A

Slide 17

API: Data Send Timeline Host A: send information App asks for that Horde send an ADU Data is planned (ADU sections → parcels) at least one ADU pieces for each bundle Host B: get information Data is unloaded (bundles → ADU parts) App told about got sections ACKs sent for got bundles Host A: get ACKs mapped to ADU section data Losses recognized from ACK stream App advised about ACK'd or Lost ADU parts

Slide 18

Horde API: QoS Objectives ADUs can be "labeled" with Quality-of-administration targets Tags are indications to bundle scheduler A conceptual approach to balance the parcel booking arrangement Example labels: ADU need Correlation assemble Minimize the joint misfortune likelihood P(X lost ∧ Y lost), if the two ADUs X and Y are in a similar relationship aggregate. Components in a similar connection gathering are more averse to encounter related disappointments. E.g., President and Vice president would have a similar gathering.

Slide 19

Scheduler and system layer can have per-session arrangements. Different usage of these. Swarm Internals Applications APPLICATION IPC API ADUs Session ctrl MUX Packet Scheduler (iMux) Horde Daemon Network Channel Management Packets O/S NETWORK SERVICES

Slide 20

Channel Managers A solitary channel supervisor example for every dynamic system interface Pool director makes and crushes the M i The M i actualize blockage control Multiple executions of the channel administrator interface Pool chief picks one in light of per-interface settings Packet Scheduler Network Management Layer M 1 M 2 M N O/S NETWORK SERVICES

Slide 21

Channel Managers Primary administrations gave by M X Throttle parcel sends on interface X. Produce ACK and LOSS warnings for bundles sent on X. Every M X runs a clog control session Congestion control rationale has a place beneath the striping layer Multiple autonomous blockage spaces, require numerous free sessions. Channel-particular advancements conceivable Pick clog control calculation in view of the channel sort (e.g., 802.11, EV-DO, WiMax) Implementations: AIMD, CBR, AIMD_EVDO, DCCP * , …

Slide 22

Transmission Slots TxSlot items are the interface between the scheduler and the system layer For the scheduler, each channel director is a wellspring of TxSlots. Each TxSlot gives a bundle TX capacity. At the point when a space gets to be distinctly accessible, the scheduler maps information to that opening. A TxSlot gives anticipated that QoS would that TX E.g., expected RTT and expected misfortune likelihood. The scheduler can utilize the normal QoS fields to decide the best information for that opening.

Slide 23

D i T i TxSlot Life Cycle ADU schedule_tx_slot(… ) TxSlot → Data mapping ADU at least one ADU sections pressed together Packet Scheduler T i Manager k generate_tx_slot(… ) TxSlot distribution consume_tx_slot(… ) TxSlot deallocation Packet

Slide 24

Hoard Scheduler » schedule_tx_slot ( opening ): streams = set of streams with unsent information while ( space not full) and ( streams not vacant) : stream = most elevated need from streams adu = ADU at the head of stream f = biggest piece that will fit in opening if test_constraints ( space , f ) is alright : perused f from adu into opening if (not any more unsent information in adu ): pop adu from stream if ( stream exhaust): expel from streams if ( opening has information) : consume_tx_slot ( opening )

Slide 25

Currently Supported Objectives Optimization Static ADU need Constraints Stream FIFO requesting ADU misfortune likelihood limit ADU idleness edge Stream inactivity change edge ADU relationship bunches Other Resilient ADU sends

Slide 26

Conclusion Mobile telemedicine arrangement Horde is a major some portion of a framework that is being sent soon. Will permit specialists to attempt things they wouldn't have possessed the capacity to strive for another 5-10 years. Bolster the advancement of requesting versatile applications Transparently abuse all accessible remote assets. Capable deliberations (inner and API) Objectives, transmission spaces, and so on. Permitted us to separately explore different avenues regarding distinctive blockage control plans and booking methodologies.

SPONSORS