08-12-2012, 03:33 PM
LANS, performance and Client/Server design issues
LANS.ppt (Size: 112 KB / Downloads: 53)
Basic performance definitions
Bandwidth
Raw data rate of links
Capacity
Theoretical limit of data transfer
Measured over the network, sub-net or link
Throughput
Actual data transmitted (e.g. packets per second)
Limited by protocol overhead, delays, latency etc
Basic performance definitions
Latency
End-to-end delay, comprising
propagation delay (near speed of light),
transmission delay (media speed),
store-and-forward delay (bridge/switch/router buffering),
processing delay (action on protocol elements)
Sensitivity to delay is application dependent
video is very sensitive and
virtual terminal (Telnet) is medium sensitive (user-dependent)
Jitter
The variability of latency
Buffering can smooth the delay
Media access delay
LAN access delay depends on
Access scheme used
No. of contending devices
Accuracy
Data corruption
Bit error rate on WAN links
< 1 in 106 on LANs
Key performance relationships
Payload (TCP/IP over Ethernet)
Payload = MTU – (TCPOverhead + IPOverhead+ MACOverhead)
MTU is maximum transmission unit
Overheads are: TCP 20 bytes; IP 20 Bytes; MAC 18 bytes
Maximum packet rate
PPSmax =Channel Speed
(8 bits x PDUsize )
For example at 64 kbps with 128 byte PDUs
PPSmax =64000/(8 x 128) = 62.5 pps
Performance issues
Different network types have different maximum packet/frame sizes
Overlarge packets need fragmentation and re-assembly to be transmitted
limits throughput
reduces performance
Compression can be used to improve performance on slower speed links
Scalable Client Server
For the single User
Client, middleware and most of the business services on a single machine
For the SME
Use of small LAN
Often involves multiple clients talking to a local server
For the Enterprise
Connection of multiple servers across a network
To utilise fully requires low cost, high speed bandwidth