©2000 Pål HalvorsenMIS 2000, Chicago, October 2000 Network Level Framing in INSTANCE Pål...

Preview:

Citation preview

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Network Network LevelLevel Framing in INSTANCE Framing in INSTANCE

Pål Halvorsen, Thomas Plagemann, and Vera GoebelUniversity of Oslo, UniK- Center for Technology at Kjeller

Norway

Overview:• Application Scenario

• The INSTANCE project:– Zero-Copy-One-Copy Memory Architecture

– Integrated Error Management

– Network Level FramingNetwork Level Framing

• Conclusions

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Media-on-Demand server:Applicable in applications like News- or Video-on-Demand provided by city-wide cable or pay-per-view companies

Application Scenario

Network Network

Multimedia Storage Server

Project goals:Optimize performance within a single server:• Reduce resource requirements • Maximize number of clients

Retrieval is the bottleneck:Some important factors:• Memory management• Error management• Communication protocol processing

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Zero-Copy-One-Copy IZero-Copy Delay-Minimized Broadcast

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Zero-Copy-One-Copy II

Integrated Zero-Copy and Delay-Minimized Broadcasting

Zero-Copy Delay-Minimized Broadcast

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Traditional Error Management

Encoder Decoder

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Integrated Error Management

Encoder Decoder

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Traditional Storage

TRANSPORT

NETWORK

LINK

TRANSPORT

NETWORK

LINK

TRANSPORT

NETWORK

LINK

TRANSPORT

NETWORK

LINK

Upload to serverFrequency: low (1)

Download from serverFrequency: very high

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Network Level Framing (NLF)

TRANSPORT

NETWORK

LINK

TRANSPORT

NETWORK

LINK

NETWORK

LINK

NETWORK

LINK

TRANSPORT TRANSPORTTRANSPORT TRANSPORT

Upload to serverFrequency: low (1)

Download from serverFrequency: very high

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

NLF:

When to Store Packets

IP

Transport Layer

Network Layer

Link Layer

IP

UDP

UDP

IP

TCP

IP

TCP

IP

UDP

IP

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

NLF:

Splitting the UDP Protocol

UDP

IP

UDP

Prepend UDP and IP headers

Prepare pseudo header for checksum

Calculate checksum

Fill in some other IP header fields

Hand over datagram to IP

udp_output()

Temporarily connect

Disconnect connectedsocket

udp_output()

Prepend UDP and IP headers

Prepare pseudo header for checksum, clear unknown fields

Prealculate checksum

udp_PreOut()

Update UDP and IP headers

Update checksum, i.e., only add checksum of prior unknown fields

Fill in other IP header fields

Hand over datagram to IP

udp_QuickOut()

UDP

IP

UDP

UDP

IP

UDP

UDP

IP

UDP

UDP

IP

UDP

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

NLF:

Checksum Operations I• The UDP checksum covers three fields:

– A 12 byte pseudo header containg fields from the IP header

– The 8 byte UDP header

– The UDP data

• Checksum calculation function (in_cksum):

u_16int_t *w;for each mbuf in packet {

w = mbuf->m_data;while data in mbuf {

checksum += w;w++;

}}

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

NLF:

Checksum Operations II• Traditional checksum operation:

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

NLF:

Checksum Operations III• NLF checksum operation:

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

FEC

FEC

Integrated Error Management and NLF

UDP - PreOut

Network

UDP - QuickOut

UDP - Input

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Expected Performance Gain

• Traditionally, 60% of the UDP/IP processing is spent on checksum calculation [Kay et al. 93]

• Performance gain using NLF (packet size dependent):– The checksum operation is greatly reduced

• E.g., 4KB packet size checksum operation on only 0.2% of data

• Speed-up factor of about 2

– Throughput gain of about 10%

©2000 Pål HalvorsenMIS 2000, Chicago, October 2000

Conclusion

The INSTANCE project aims at optimizing the data retrieval in an MoD server:– Zero-Copy-One-Copy memory architecture– Integrated Error Management

– Network Level FramingNetwork Level Framing Saving time for checksum calculation where expected

performance gain is:– About 99% reduction of the checksum calculation time

(UDP/IP processing speed-up of 2)– About 10% increase in throughput

– Future work

Recommended