Filter
Top Products
www.ti.com
B.2ReadPerformance
ReadPerformance
Sincereadsmustcompleteatransmit-remoteread-receivecyclebeforestartinganotherreadtransaction,
thedatathroughputislowerascomparedtowrites.Thereislatencyinvolvedinreadingthedatafromthe
remotedevice;andinsomecases,alocallatencyinwritingthereturneddatabeforethenextreadcan
start.
Themaxreadrateiscalculatedthesamewayasthemaxwriterate.Thepacketoverheadisasshown
below:
Read32-caaaaT
ReadBurst-claaaaT
ReadReturn-clddddddddT
Where
T-EndOfPacket
d-data,ddddrepresentsadditional32-bitwordsinburst,upto16words.
a-address
c-command
l-length
Thereare6bytesofoverheadforasingleread,7bytesforburstreads,and3bytesforreadreturns.The
timerequiredforareadisthetotalofthetimeforthereadrequest,remotelatency,readreturn,andlocal
latency.Thus,thethroughputcanbecalculatedasdatabytes/totaltransactiontime,wherethelatencyof
bothlocalandremotedevicesiscombined.
ReadThroughput=data/(((Read+ReadReturn+data)/maxreadrate)+Latency
=(data×maxreadrate)/((Read+ReadReturn+data)+Latency×max
readrate)
Forexample,witha4pin,99MHZVLYNQconnection,forasingle32-bitwordread:
ReadThroughput=32bits×316.8Mbps/(6×8+3×8+4×8+Latency×316.8Mbps)
=10137.6/(104+Latency×316.8Mbps)
Similarly,foraburstreadofsixteen32-bitwords,witha4pin,99MHZVLYNQconnection
ReadThroughput=16×32bits×316.8Mbps/(6×8+3×8+16×4×8+Latency×316.8Mbps)
=162201.6/(584+Latency×316.8Mbps)
Usingtheformulaabove,therelativeperformancewithvariouslatenciesisillustratedfora4pin,99MHZ
VLYNQclock,burstread(sixteen32-bitwords)throughputrate,asshowninTableB-3:
TableB-3.RelativePerformancewithVariousLatencies
NumberofVLYNQPinsThroughputThroughput
(99MHZ)BurstSizein32-bitWordsLatency(μsec)(Mbits/sec)(Mbytes/sec)
4160277.7434.72
1179.7022.46
1043.025.38
1005.000.62
ToefficientlyuseVLYNQbandwidth,itisdesirableforeachVLYNQdevicetowritefromthelocaldevice
totheremotedevice.Bursttransactionsaremoreefficientthansingleread/writetransactions.
SPRUE36A–September2007Write/ReadPerformance47
SubmitDocumentationFeedback