文档库 最新最全的文档下载
当前位置:文档库 › A COMPREHENSIVE STOCHASTIC MODEL Abstract FOR TCP LATENCY AND THROUGHPUT

A COMPREHENSIVE STOCHASTIC MODEL Abstract FOR TCP LATENCY AND THROUGHPUT

A COMPREHENSIVE STOCHASTIC MODEL Abstract FOR TCP LATENCY AND THROUGHPUT
A COMPREHENSIVE STOCHASTIC MODEL Abstract FOR TCP LATENCY AND THROUGHPUT

A COMPREHENSIVE STOCHASTIC MODEL

FOR TCP LATENCY AND THROUGHPUT

D.Zheng,https://www.wendangku.net/doc/bf1114657.html,zarou

Electrical and Computer Engineering

Mississippi State University

Box9571

Mississippi State,MS39762,USA

glaz@https://www.wendangku.net/doc/bf1114657.html,

Abstract

Understanding the nature of TCP behavior is critical in order to properly engineer,operate,and evaluate the performance of the Internet,as well as to properly design and implement future networks.In this paper,we?rst develop a better and tractable model for the congestion window growth pattern in the slow-start https://www.wendangku.net/doc/bf1114657.html,ing this new slow-start phase model,we construct an accurate model for the short-lived TCP?ows and then an extended and more accurate TCP steady-state model.We validate our models with simulations and compare them against existing models.The results show that our model for the short-lived?ows yields more accurate performance predictions(up to)than the ones developed in[1]and[2].In addition,our extended steady-state model is up to more accurate than the model proposed in[3].

Key Words

TCP,performance evaluation,stochastic model

1.Introduction

A multitude of Internet applications,such as the world wide web,usenet news,?le transfer and remote login, have opted TCP as the transport mechanism.Thus,TCP greatly in?uences the performance of Internet[4],[5], and a well-designed TCP is of utmost importance to the level of satisfaction of Internet users.Several stochastic TCP models have been proposed[1]-[3],[6]for predicting its performance in terms of latency and throughput. Considerable emphasis has been given into better understanding of the dynamics of TCP and its sensitivity to network Parts of this paper were presented at ICC’03,c IEEE2003,and CIIT’03,c IASTED2003

parameters,such as the TCP round trip time and the packet loss rate.Understanding the impact of TCP dynamics on its performance is critical for optimizing TCP and the design of active queue management techniques[7],[8] and TCP-friendly multicast protocols[9],[10].However,TCP is a very complex protocol,and the fast-changing network conditions make the development of an accurate TCP stochastic model to be a very challenging task. Stochastic models of TCP can be classi?ed into three classes:(1)steady-state models for predicting the perfor-mance of bulk transfer?ows[11],[3],(2)models for short-lived?ows assuming low loss rates[12],[6],[13],and (3)models that combine the two above models[2],[1].

To the best of our knowledge,none of the steady-state models proposed so far account for the slow-start phase which begins at the end of every single time-out.The work in[3]assumes that the slow-start phase happens less frequently than the congestion-avoidance phase and the throughput in the slow-start phase is less than that in the congestion-avoidance phase,and that the slow-start phase can be ignored safely.While this could be the case with very small loss rates,the assumption does not hold in general.Empirical measurements show that of the packet losses lead to time-outs[3].Since TCP enters the slow-start phase when a time-out occurs,accurate TCP performance models must take into consideration of the aggregate effects of the slow-start phases.

All steady-state models assume the availability of unlimited data to send.Hence,the impact of the transient phase on performance is considered insigni?cant,and therefore is ignored.These models work well only for predicting the TCP send rate or the throughput of bulk data transfers,and are not applicable to predicting the performance of short-lived TCP?ows.

It is noted in[14]-[15]that the majority of TCP traf?c in the Internet consists of short-lived?ows,i.e.,the transmission comes to an end during the slow-start phase before switching to the congestion-avoidance phase. Hence,new models are needed that are capable of predicting the performance of short-lived TCP?ows;and this is one of the main subject of our study.

In this paper,we?rst develop a better and tractable model for the congestion window growth pattern in the slow-start https://www.wendangku.net/doc/bf1114657.html,ing this new slow-start phase model,we construct an accurate model for the short-lived TCP?ows and then an extended and more accurate TCP steady-state model.Major improvement in both models is achieved by relaxing key assumptions and enhancing critical approximations that have been made in existing popular models. The remainder of the paper is organized as follows.Section2presents an analysis in developing the improved and extended steady-state model.Section3builds the stochastic model for short-lived?ows.In section4,both models are validated with simulations and compared against existing models.Finally,Section5concludes the paper.

Figure1.The extended steady state model-evolution of congestion window size when loss indications are triple-duplicate ACK’s and time-outs.

2.Steady-state Model Incorporating the Slow-start Phase

2.1Assumptions

As in[3],we develop our models based on the BSD TCP Reno release[16].We assume that the link speed is very high,the round-trip time(RTT)remains fairly constant at all times,and the sender sends full-sized segments whenever the congestion window()allows.The advertised window is assumed to be always a constant and large.Thus,the congestion window evolution alone determines the send rate,which roughly can be described as .

We model the dynamics of TCP in terms of“rounds”as done in[3].A round starts when a window of packets is sent by the sender and ends when one or more acknowledgments are received for these packets.The effect of the delayed acknowledgment is taken into consideration,but neither the Nagle algorithm nor the silly window syndrome avoidance is considered.In addition,we assume that the packet losses are in accordance with the bursty loss model.The packet losses in different rounds are independent,but they are correlated within a single round; that is,if one packet in a round is lost,then the following back to back packets in the same round are also assumed to be lost.This is an idealization of the packet loss dynamics observed in the paths where FIFO drop-tail queues are used[1].Finally,we assume that the sender has unlimited data to send.

2.2Model Development

Fig.1depicts an instance of the congestion window’s evolution over time.As shown in the?gure,when a time-out occurs due to lost packets,TCP enters into the slow-start phase to recover from a perceived network congestion. Let TDP be the period between two triple-duplicate()losses,be the time spent in the slow-start phase, be the duration of the congestion-avoidance phase,and be the time interval of the time-out phase.Let

be the number of packets sent during the total time.Then,we have that:

(1)

(2) where is the number of packets sent during the slow-start phase,is the duration of the th TDP,is the total number of the TDPs in the interval,is the number of packets sent during the th TDP of interval ,and is the number of packets sent during the time-out phase.is the window size at the end of a slow start and?nally is the window size at the end of the TDP.

Assuming to be a sequence of independent and identically distributed(i.i.d.)random variables,we determine the send rate as:

Considering to be i.i.d.random variables and independent of and,we have:

(3) We next derive the closed form expressions for these expected values in the different TCP phases:the slow-start, the congestion-avoidance and the time-out phases.

1)The Slow-Start Phase:According to TCP Reno[17],[16],the current state of a TCP connection is determined based upon the values of the congestion window size()and the slow-start threshold().If is less than,TCP is in the slow-start phase,otherwise,it is in the congestion-avoidance phase.

Since TCP has no knowledge of the network conditions,during the slow-start phases,it probes for the available bandwidth“greedily”,i.e.,it increases the by one upon the receipt of a non-repeated acknowledgment.This algorithm can be formulated as:

(4) in which is the congestion window size for the round.(4)is due to the fact that assuming no loss,in round(),there is a total of packets sent to the destination,which,in turn,causes the receiver to generate acknowledgments1.According to the slow-start algorithm,upon receiving these ACKs,the sender increases the by the number of ACKs it has obtained,which is.

1the smallest integer bigger than.

5 Noting that the congestion window is an integer,we can simplify(4)as follows2:

(5) Rearranging,we get:

(6) Substituting this in(4),we get the following:

(7) In order to examine the accuracy of this approximation,a typical evolution of is given as follows:

Compared with the sequence generated by(7):

and the evolution of proposed by the model in[1]:

or calculated as:

The similarity between the two previous sequences and the discrepancy between the real evolution of with the proposed model in[1]show that(7)gives a better approximation of the slow-start phase.

Noting that(7)generates the Fibonacci sequence,we can therefore express as follows:

n=(8) where3:

(9)

and are determined by the initial value of.Assuming the initial value of is,we get:

(10) 2In deriving a model for the latency of the short-lived TCP?ows,(5)was approximated in[1]as:.

3is also called the golden number which will be denoted as in the later parts of this paper.

By knowing the evolution of the congestion window,we can calculate the total number of packets,,that are sent until the round,by summing the congestion window size during each round:

(11) The last approximation is due to the fact that:

Thus,from(11),the number of rounds,,can be computed as:

(12) Substituting(12)into(8),we can get the approximate relationship between the congestion window size and the total number of packets that have been sent,as follows:

(13) Taking the expectation of both sides of(13),we have:

(14)

in which is the expectation of.

If the slow-start phase is ended by a packet loss,the expected data that have been sent during this phase can be calculated as:

(15) where is the loss rate.

Substituting the value of in(14),we get:

(16) This is the expected value of the congestion window when the slow-start phase ends due to a lost packet.Observing that when is small,the expected value would be much bigger than the expected value of,i.e.,:

(17) where the last equality comes from the fact that after each time-out,the slow-start threshold is set to half of the current congestion window.

i

packets no of rounds

Figure 2.Packets sent during a TDP.Adopted from [3].

Thus,it is safe to assume that TCP enters the congestion-avoidance phase before a packet gets lost.That is,we assume that TCP always switches from the slow-start to congestion-avoidance phase when the congestion window reaches the value of .We show the proof of this in the next section after obtaining the closed-form solution

of

.

As a consequence,we have that the expected congestion window size at the end of the slow start be constrained by the limitation of the slow-start threshold:

(18)

Using (18)in (14)and rearranging,we obtain the expected number of packets sent during the slow-start phase:

(19)

The time spent in the slow-start phase is obtained by multiplying the number of rounds described in (12)with RTT:

(20)

2)The Congestion-Avoidance Phase:Let be the number of packets sent during the th TDP,be the

duration,and be the window size at the end of the TDP.With reference to Fig.2,we obtain the following

relations [3]4:

(21)(22)

(23)

4

For details see [3].Note that (23)captures more accurately the window size at the end of the TDP than the one presented in [3].

and:

(24) where is the penultimate round in the TDP which experiences packet losses,is the round trip time,is the number of packets sent in a TDP until the?rst loss happens,is the number of packets acknowledged by a received ACK,and is the number of packets sent in the fast retransmit phase,which is the last round[3]. Based on our assumptions,is obviously geometrically distributed.Hence:

(25) and therefore,we have that:

(26) In addition,based on(23)and(24),we also have that:

(27)

(28) where we assume and are mutually https://www.wendangku.net/doc/bf1114657.html,bining(26),(27),and(28),we get:

(29) Since is the number of packets sent when packets in the penultimate round are ACKed,its value equals to with probability:

(30) Therefore:

(31) for p https://www.wendangku.net/doc/bf1114657.html,ing(31)in(29)and rearranging,we get:

(32)

Inserting(32)in(27),we obtain:

(33) and:

(34) where we assume’s to be i.i.d.and.

In the previous subsection,we stated without proof that the slow-start phase will enter the congestion-avoidance phase before a packet loss happens.This can be proved if,the expected congestion window size at the end of the slow-start phase due to a packet loss,is bigger than the value of,which is the expected threshold at the beginning of the slow-start phase.In other words,we need to show that:

(35) where is given by(32).This is easily shown below,under the(normal)condition that is small:

The last inequality stands obviously.In fact,(35)is valid.

3)The Time-out Phases:The probability that a loss indication is a time-out under the current congestion window size,is given in[3]as:

(36) which gets simpli?ed when the loss rate,,is small:

Thus,,the expected probability that a loss leads to a time-out at the end of the congestion-avoidance phase, is approximated in[3]as follows:

(37) The traf?c traces collected in[3]indicate that the effect of the time-outs must always be captured by any TCP performance prediction model.In most of the traces,time-out events out-numbered the fast retransmit events,i.e.,

is around of the total loss.This value is larger than the value given by the formula of(37),as we further calculated that the is greater than,which,in turn,renders the to be less than.So,we believe that this approximation underestimates the real.As a matter of fact,the underestimation of in[3]is due to the approximation of by noting that:

The equality holds only when is a constant.

Now,using Taylor’s formula and expectation properties,we obtain the following5:

(38) Hence,to?nd a more accurate approximation of,we must?nd the variance of.

After a rigorous analysis6,we obtain the variance of,the congestion window size at the end of TDP,to be:

(39) Substituting(32)and(39)into(38),we get:

(40)

(40)gives a better,but still simple,estimation of.Then,,the probability that a loss detection is

a time-out(TO),can be found to be:

(41) The probability of,the number of TDPs,is derived according to:

This is due to the fact that,with probability,the packets lost at the end of the congestion control phase lead to a TO,and,with probability the TCP connection stays in TDP.By taking the expectation of,we get:

(42)

5See Appendix1for the derivation.

6See Appendix2for details.

The expressions for the number of packets sent in the time-out phase,and its duration,are given in[3]as:

(43)

(44)

where is de?ned as:

(45)

4)The Steady State Send Rate and Throughput:Substituting(19),(20),(26),(32),(34),(41),(42),(43)and (44)into(3),and taking into consideration the limitation of the window size[3],we?nally derive the send rate as:

when

(46)

when

This can be further simpli?ed as:

(47) To derive the throughput,we only need to change,the expected size of packets that have been sent in a TDP,to,the expected size of packets that have been received in a TDP.can be expressed as:

(48) where is and is given by(31).Also we substitute with,the expected number of packets received in the time out phase,where[3]:

Thus,the throughput can be formulated as:

(49)

or:

when

(50)

when

which,when p is small,can be simpli?ed as(47).This can be explained by noting that,if a loss seldom happens, then the send rate should just equal to the throughput.

Figure3.Our proposed model is compared with the one developed in[3]in terms of the predicted throughput difference versus the loss rate()for the case of:,,,,,.

Fig.3compares our model against the one proposed in[3].It shows the predicted throughput difference versus for the case of,bytes,segment,sec,segments,and .With both models,when,then.However,for,the model in[3] overestimates the throughput by up to a factor of2.5(at).Obviously,when,again both models obtain the same performance values.

3.Stochastic Model for Short-lived Flows

Our proposed model for the short-lived TCP?ows is partially based on our results given in Section2-2.1.In addition,it is composed of four parts according to a typical short-lived?ow evolution:the start of the connection (three-way-handshake),the initial slow-start phase,the?rst loss,and the subsequent losses.We?rst derive the latency a?ow experience in each part,and then sum them to obtain the total latency.

3.1The Connection Start-up Phase

Every TCP connection starts with the three-way-handshake process.Assuming that no ACK packets can get lost, this process can be well modeled as follows[1]:

(51) where is the duration of SYN time-out and is the packet loss rate.

We further assume that two or more time-outs within the three-way-handshake process is very rare.Otherwise, the slow-start threshold would get set to one,and therefore,the connection would get forced directly into the congestion-avoidance phase instead of into the slow-start phase.

3.2The Initial Slow-start Phase

After the three-way-handshake,the slow-start phase begins.In this phase,the sender’s congestion window() increases exponentially until either of the following two events occur:a packet gets lost or the reaches its maximum value.

In order to derive the latency for this phase,,the expected number of packets sent until a loss occurs is given by the following enhanced equation(based on the one given in[1]):

(52) where is the total?le size measured in packets that must be transmitted.

Substituting(52)in(14),we obtain the expected congestion window size at the end of the slow-start phase due to packet losses as:

(53) If,then the congestion window?rst grows to and then remains there while sending the rest of the packets.Thus,the whole procedure is divided into two parts[1].From(14),the number of packets sent when the grows to is given by:

(54) Substituting(54)into(12),we can obtain the duration of this step measured in rounds:

(55) In the second part:

(56) rounds are needed to transmit the remaining packets.

Combining the previous results together and using(12)for the case,the expected slow-start latency is computed as follows:

when

(57)

when

= w

= w+k

received packet lost packet W

W +ss ss

Figure 4.An illustration of a triple-duplicate (TD)event.

3.3The First Loss

The initial slow-start phase ends when a packet loss is detected with a probability of

.When a packet

gets lost,it could cause retransmission time-out (RTO)or lead to a triple duplicate ACKs,in which case TCP could recover in a round or two by using the fast retransmit and the fast recovery mechanism.We ?rst derive the probability that a packet loss leads to a time-out (TO).Due to the exponential growing pattern of

in the slow-start phase,

,the probability that a packet loss

leads to a TO is different from the probability that when the sender is in the congestion-avoidance phase.With reference to Fig.4,we derive the expression of as follows.

In the round with a TD event,let be the current size of ,which has a value

.In this round,

packets

were sent.Among them,

packets are assumed to be ACKed.Since the connection is still in the slow-start phase,

increases to

and another packets are sent in the next round 7.If more than three packets from these

packets get ACKed,then a TD would occur;otherwise,a TO would take place.Letting:

if

(58)

be the probability that no more than 2packets have been transmitted successfully in a round of packets,we then

obtain

to be:

for

otherwise

(59)

where

is as given by (30)and gives the probability that the ?rst

packets have been successfully transmitted

and ACKed in a round of

packets,provided that there might be one or more packets got lost.Simplifying (59),

7

The delayed acknowledgment concept is not applied here,but we show later that it does not affect the analysis of the .

we get to be equal to:

(60) As approaches zero,(60)reduces to:

(61) In case of delayed acknowledgment,successfully received packets generate8ACKs,and thus the size of the increases to and packets are sent.Therefore can be computed as:

for

otherwise

which is same as(60)since:

for

The expected time that TCP spends in the RTOs is given by(44).The time that TCP spends in the fast retransmit phase,,depends on where the loss would happen[2]:

if the lost packet is in the last three packets of the window

(62)

otherwise

Thus,when the congestion window size is bigger than three,the expected time,can be found to be:

(63) Finally,the expected latency that this loss would incur is:

(64) where is:

(65) 3.4Sending the Rest of the Packets

After the?rst packet loss,the transmission latency of the rest()packets is obtained by using our extended steady-state model as follows:

(66) where is as given by(49).

8is the biggest integer small than.

Figure5.The model for short-lived TCP connections is compared with the steady-state model in terms of throughput versus loss rate for different?le sizes.Model parameter values:,,,,, .

3.5Total Latency

Grouping(51),(57),(63)and(66)together and considering the delay()caused by the delayed acknowledgment for the?rst packet(whose mean value is100ms for the BSD-derived implementations),we now have the total expected latency:

(67) Note that the last term is due to the fact that only half of a round is needed to send the last window of packets. In Fig.5,we compare this model for short-lived TCP connections against our steady-state model.Clearly,as the transferred?le size increases,the short-lived TCP connection model approaches the steady state model.This is because when a connection has a large amount of data to send,TCP spends most of its time in the steady-state.In addition,as the loss rate increases,the throughput predicted by the short-lived TCP connection model approaches the one predicted by the steady-state model.This is because as the connection loses its packets more frequently, the transient slow-start phase ends quickly and the remaining packets are sent in the steady-state phase.

4.Model Validation through Simulation

We validated our proposed analytical models with simulation experiments.We performed all experiments in ns-2[18]using the FullTCP agent.The FullTCP agent is modeled based on the4.4BSD TCP implementation and can simulate all the important features of TCP Reno.The ns-2simulation model used in all experiments is shown in Fig.6.

S1

S2

r2

r1

ftp1

FullTcp1Sink1DropTail

(Bursty loss model)

5ms

10Mb 5ms

10Mb 10Mb, 90ms

(MSS=536, Wm=20)Figure 6.The ns -2model that was used to validate our analytical TCP models.

Unlike in [1]where the Bernoulli loss model is used,in our experiments packets were getting lost according to the bursty loss model.Since ns-2does not have built-in bursty packet loss model,we added our own BurstyError Model,which was derived from the basic Error Model class.This BurstyError Model drops packets with probability

,which is a Bernoulli trial.After a packet is selected to be dropped with probability ,all the subsequent packets

in transit are also dropped.This emulates the DropTail queues behavior under congestion conditions.

We used FTP 9as the application for sending a controlled number of packets over a 10Mbps link.The experiments were designed such that the minimum RTT was 200ms.

4.1The Steady-state Model

Using the same system parameter values that were used to generate Fig.3,we performed 1000simulation experiments for each value of ,where

was varied from 0.005to 0.1.The ?le size was set to 10MB.Fig.

7compares the simulation results against the analytical results obtained from our proposed steady-state model (Full:(50),and Approximate:(47))and the one developed in [3].Clearly,the results match our expectations.The predicted throughput values at each value of

obtained from our model much closer to the simulation values.

To quantify the accuracy of our model relative to the simulation data,we computed the average error using the following expression taken from [3]):

where

is the throughput predicted by the models and

is the throughput observed from the

simulation experiments.A smaller average error value indicates a better model accuracy.We plotted these average errors against loss rates in Fig.8.It shows that in most cases the average error is under

for our proposed full

9

FTP is a major Internet application that is used to remotely transfer ?les.

Figure7.Predicted throughput obtained by our proposed steady-state model and the one developed in[3]are compared against simulation results for the case of,,,,.

model(i.e.,(50))and above for the one from[3].Approximately,in most cases,our model is more accurate than the model proposed in[3].This supports our claim that by including the slow-start phase into the steady-state model more accurate predictions can be obtained.

In addition,Fig.8depicts the following:the average error in predicted throughput from both analytical models increases as approaches zero.Let say that and the initial slow-start threshold is set to the maximum window size.Then,the initial slow-start phase is extended until the congestion window reaches the maximum window size.Since there are no packet losses,TCP never switches to the congestion avoidance phase,but rather continues transmitting packets at its maximum sending rate allowed by the maximum window size.For these cases that,our short-lived TCP?ow model should be used instead of the steady-state model.

4.2Short-lived Flows Model:Latency versus Transferred File Size

Fig.9shows the relationship between the latency and the transferred?le size under no loss conditions.It compares the latency predictions given by our proposed short-lived TCP model((67))and the ones obtained by the short-lived TCP models developed in[1]and[2]against the simulation results.Obviously,our model’s prediction values match the simulated values better that the values obtained by the other models.Our model resulted in average error,compared to and obtained by the models in[1]and[2],respectively.

Analyzing the results,we also observed that all prediction errors resulted from our model are within[-RTT/2, RTT/2].We believe that this is because our model accounts for the delayed acknowledgment mechanism.Thus,

Figure8.Our proposed steady-state model is compared with the one developed in[3]in terms of the average error for the case of ,,,,.

TABLE I

O UR SHORT-LIVED TCP CONNECTION MODEL IS COMPARED AGAINST THE ONE PROPOSED IN[1]IN TERMS OF THE AVERAGE ERROR.

Loss Rate

File Size0.526KB2KB6KB11KB

[CSA00]9.40% 4.08% 6.43%8.38%

Proposed 5.83%0.59%7.54%7.64%

for the cases where is small,the prediction errors are insigni?cant.This is not valid for the other models propose by[1]and[2].

4.3Short-lived Flows Model:Throughput versus Loss Rate and File Size

Fig.10,11and12compares the throughput versus loss rate predictions given by our proposed short-lived TCP model and the one obtained by the short-lived TCP models developed in[1]against the simulation results for the cases of2KB,6KB,and11KB?le sizes.Table I compares the two models in terms of the average error.

As can be observed,when the transferred?le size is small and the loss rate is low,our model yields more accurate predictions than the model from[1].Again,this is because our model accounts for the delay acknowledgment mechanism and uses(golden number)instead of(see[1]).However,for large?le sizes and loss rates,both models yield similar predictions and in agreement with our steady-state model,as expected.

Figure9.Predicted latency versus transferred?le size obtained by our short-lived TCP connection model and the ones developed in[1]

and[2]are compared against simulation results for the case of,,,,

,

.

中考必会几何模型:8字模型与飞镖模型

8字模型与飞镖模型模型1:角的8字模型 如图所示,AC 、BD 相交于点O ,连接AD 、BC . 结论:∠A +∠D =∠B +∠C . O D C B A 模型分析 证法一: ∵∠AOB 是△AOD 的外角,∴∠A +∠D =∠AOB .∵∠AOB 是△BOC 的外角, ∴∠B +∠C =∠AOB .∴∠A +∠D =∠B +∠C . 证法二: ∵∠A +∠D +∠AOD =180°,∴∠A +∠D =180°-∠AOD .∵∠B +∠C +∠BOC =180°, ∴∠B +∠C =180°-∠BOC .又∵∠AOD =∠BOC ,∴∠A +∠D =∠B +∠C . (1)因为这个图形像数字8,所以我们往往把这个模型称为8字模型. (2)8字模型往往在几何综合题目中推导角度时用到. 模型实例 观察下列图形,计算角度: (1)如图①,∠A +∠B +∠C +∠D +∠E =________; 图图① F D C B A E E B C D A 图③ 2 1O A B 图④ G F 12 A B E 解法一:利用角的8字模型.如图③,连接CD .∵∠BOC 是△BOE 的外角, ∴∠B +∠E =∠BOC .∵∠BOC 是△COD 的外角,∴∠1+∠2=∠BOC . ∴∠B +∠E =∠1+∠2.(角的8字模型),∴∠A +∠B +∠ACE +∠ADB +∠E =∠A +∠ACE +∠ADB +∠1+∠2=∠A +∠ACD +∠ADC =180°. 解法二:如图④,利用三角形外角和定理.∵∠1是△FCE 的外角,∴∠1=∠C +∠E .

∵∠2是△GBD 的外角,∴∠2=∠B +∠D . ∴∠A +∠B +∠C +∠D +∠E =∠A +∠1+∠2=180°. (2)如图②,∠A +∠B +∠C +∠D +∠E +∠F =________. 图② F D C B A E 312图⑤ P O Q A B F C D 图⑥ 2 1 E D C F O B A (2)解法一: 如图⑤,利用角的8字模型.∵∠AOP 是△AOB 的外角,∴∠A +∠B =∠AOP . ∵∠AOP 是△OPQ 的外角,∴∠1+∠3=∠AOP .∴∠A +∠B =∠1+∠3.①(角的8字模型),同理可证:∠C +∠D =∠1+∠2.② ,∠E +∠F =∠2+∠3.③ 由①+②+③得:∠A +∠B +∠C +∠D +∠E +∠F =2(∠1+∠2+∠3)=360°. 解法二:利用角的8字模型.如图⑥,连接DE .∵∠AOE 是△AOB 的外角, ∴∠A +∠B =∠AOE .∵∠AOE 是△OED 的外角,∴∠1+∠2=∠AOE . ∴∠A +∠B =∠1+∠2.(角的8字模型) ∴∠A +∠B +∠C +∠ADC +∠FEB +∠F =∠1+∠2+∠C +∠ADC +∠FEB +∠F =360°.(四边形内角和为360°) 练习: 1.(1)如图①,求:∠CAD +∠B +∠C +∠D +∠E = ; 图 图① O O E E D D C C B B A A 解:如图,∵∠1=∠B+∠D ,∠2=∠C+∠CAD , ∴∠CAD+∠B+∠C+∠D+∠E=∠1+∠2+∠E=180°. 故答案为:180° 解法二:

美国常青藤名校的由来

美国常青藤名校的由来 以哈佛、耶鲁为代表的“常青藤联盟”是美国大学中的佼佼者,在美国的3000多所大学中,“常青藤联盟”尽管只是其中的极少数,仍是许多美国学生梦想进入的高等学府。 常青藤盟校(lvy League)是由美国的8所大学和一所学院组成的一个大学联合会。它们是:马萨诸塞州的哈佛大学,康涅狄克州的耶鲁大学,纽约州的哥伦比亚大学,新泽西州的普林斯顿大学,罗德岛的布朗大学,纽约州的康奈尔大学,新罕布什尔州的达特茅斯学院和宾夕法尼亚州的宾夕法尼亚大学。这8所大学都是美国首屈一指的大学,历史悠久,治学严谨,许多著名的科学家、政界要人、商贾巨子都毕业于此。在美国,常青藤学院被作为顶尖名校的代名词。 常青藤盟校的说法来源于上世纪的50年代。上述学校早在19世纪末期就有社会及运动方面的竞赛,盟校的构想酝酿于1956年,各校订立运动竞赛规则时进而订立了常青藤盟校的规章,选出盟校校长、体育主任和一些行政主管,定期聚会讨论各校间共同的有关入学、财务、援助及行政方面的问题。早期的常青藤学院只有哈佛、耶鲁、哥伦比亚和普林斯顿4所大学。4的罗马数字为“IV”,加上一个词尾Y,就成了“IVY”,英文的意思就是常青藤,所以又称为常青藤盟校,后来这4所大学的联合会又扩展到8所,成为现在享有盛誉的常青藤盟校。 这些名校都有严格的入学标准,能够入校就读的学生,自然是品学兼优的好学生。学校很早就去各个高中挑选合适的人选,许多得到全国优秀学生奖并有各种特长的学生都是他们网罗的对象。不过学习成绩并不是学校录取的惟一因素,学生是否具有独立精神并且能否快速适应紧张而有压力的大一新生生活也是他们考虑的重要因素。学生的能力和特长是衡量学生综合素质的重要一关,高中老师的推荐信和评语对于学生的入学也起到重要的作用。学校财力雄厚,招生办公室可以完全根据考生本人的情况录取,而不必顾虑这个学生家庭支付学费的能力,许多家境贫困的优秀子弟因而受益。有钱人家的子女,即使家财万贯,也不能因此被录取。这也许就是常青藤学院历经数百年而保持“常青”的原因。 布朗大学(Brown University) 1754年由浸信会教友所创,现在是私立非教会大学,是全美第七个最古老大学。现有学生7000多人,其中研究生近1500人。 该校治学严谨、学风纯正,各科系的教学和科研素质都极好。学校有很多科研单位,如生物医学中心,计算机中心、地理科学中心、化学研究中心、材料研究实验室、Woods Hole 海洋地理研究所海洋生物实验室、Rhode 1s1and反应堆中心等等。设立研究生课程较多的系有应用数学系、生物和医学系、工程系等,其中数学系海外研究生占研究生名额一半以上。 布朗大学的古书及1800年之前的美国文物收藏十分有名。 哥伦比亚大学(Columbia University) 私立综合性大学,位于纽约市。该校前身是创于1754年的King’s College,独立战争期间一度关闭,1784年改名力哥伦比亚学院,1912年改用现名。

网络仿真技术文献综述

成绩:

网络仿真文献综述 摘要:网络仿真技术是一种通过建立网络设备和网络链路的统计模型, 并模拟网络流量的传输, 从而获取网络设计或优化所需要的网络性能数据的仿真技术。网络仿真技术以其独有的方法能够为网络的规划设计提供客观、可靠的定量依据,缩短网络建设周期,提高网络建设中决策的科学性,降低网络建设的投资风险。 网络仿真技术是一种通过建立网络设备和网络链路的统计模型, 并模拟网络流量的传输, 从而获取网络设计或优化所需要的网络性能数据的仿真技术。由于仿真不是基于数学计算, 而是基于统计模型,因此,统计复用的随机性被精确地再现。 关键词:网络仿真;统计模型;仿真技术

1.前言 目前,数据网络的规划和设计一般采用的是经验、试验及计算等传统的网络设计方法。不过,当网络规模越来越大、网元类型不断增多、网络拓扑日趋复杂、网络流量纷繁交织时,以经验为主的网络设计方法的弊端就越来越显现出来了。网络规划设计者相对来说缺乏大型网络的设计经验,因此在设计过程中主观的成分更加突出。 数学计算和估算方法对于大型复杂网络的应用往往是非常困难的,得到的结果的可信性也是比较低的,特别是对于包交换、统计复用的数据网络,情况更是如此。因此,随着网络的不断扩充,越来越需要一种新的网络规划和设计手段来提高网络设计的客观性和设计结果的可靠性,降低网络建设的投资风险。网络仿真技术正是在这种需求拉动下应运而生的。网络仿真技术以其独有的方法能够为网络的规划设计提供客观、可靠的定量依据,缩短网络建设周期,提高网络建设中决策的科学性,降低网络建设的投资风险。 网络仿真技术是一种通过建立网络设备和网络链路的统计模型, 并模拟网络流量的传输, 从而获取网络设计或优化所需要的网络性能数据的仿真技术。由于仿真不是基于数学计算, 而是基于统计模型,因此,统计复用的随机性被精确地再现。它以其独有的方法为网络的规划设计提供客观、可靠的定量依据,缩短网络建设周期,提高网络建设中决策的科学性,降低网络建设的投资风险。 2.网络仿真软件比较分析 网络仿真软件通过在计算机上建立一个虚拟的网络平台,来实现真实网络环境的模拟,网络技术开发人员在这个平台上不仅能对网络通信、网络设备、协议、以及网络应用进行设计研究,还能对网络的性能进行分析和评价。另外,仿真软件所提供的仿真运行和结果分析功能使开发人员能快速、直观的得到网络性能参数,为优化设计或做出决策提供更便捷、有效的手段。因此运用网络仿真软件对网络协议、算法等进行仿真已经成为计算机网络通信研究中必不可少的一部分。 2.1 OPNET仿真软件介绍

第四章 景观模型制作

第四章景观模型制作 第一节主要工具的使用方法 —、主要切割材料工具的使用方法 (—)美术刀 美术刀是常用的切割工具,一般的模型材料(纸板,航模板等易切割的材料)都可使用它来进行切割,它能胜任模型制作过程中,从粗糙的加工到惊喜的刻划等工作,是一种简便,结实,有多种用途的刀具。美术刀的道具可以伸缩自如,随时更换刀片;在细部制作时,在塑料板上进行划线,也可切割纸板,聚苯乙烯板等。具体使用时,因根据实际要剪裁的材料来选择刀具,例如,在切割木材时,木材越薄越软,刀具的刀刃也应该越薄。厚的刀刃会使木材变形。 使用方法:先在材料商画好线,用直尺护住要留下的部分,左手按住尺子,要适当用力(保证裁切时尺子不会歪斜),右手捂住美术刀的把柄,先沿划线处用刀尖从划线起点用力划向终点,反复几次,直到要切割的材料被切开。 (二)勾刀 勾刀是切割切割厚度小于10mm的有机玻璃板,ABS工程塑料版及其他塑料板材料的主要工具,也可以在塑料板上做出条纹状机理效果,也是一种美工工具。 使用方法:首先在要裁切的材料上划线,左手用按住尺子,护住要留下的部分,右手握住勾刀把柄,用刀尖沿线轻轻划一下,然后再用力度适中地沿着刚才的划痕反复划几下,直至切割到材料厚度的三分之二左右,再用手轻轻一掰,将其折断,每次勾的深度为0.3mm 左右。 (三)剪刀 模型制作中最常用的有两种刀:一种是直刃剪刀,适于剪裁大中型的纸材,在制作粗模型和剪裁大面积圆形时尤为有用;另外一种是弧形剪刀,适于剪裁薄片状物品和各种带圆形的细部。 (四)钢锯 主要用来切割金属、木质材料和塑料板材。 使用方法:锯材时要注意,起锯的好坏直接影响锯口的质量。为了锯口的凭证和整齐,握住锯柄的手指,应当挤住锯条的侧面,使锯条始终保持在正确的位置上,然后起锯。施力时要轻,往返的过程要短。起锯角度稍小于15°,然后逐渐将锯弓改至水平方向,快钜断时,用力要轻,以免伤到手臂。 (五)线锯 主要用来加工线性不规则的零部件。线锯有金属和竹工架两种,它可以在各种板材上任意锯割弧形。竹工架的制作是选用厚度适中的竹板,在竹板两端钉上小钉,然后将小钉弯折成小勾,再在另一端装上松紧旋钮,将锯丝两头的眼挂在竹板两端即可使用。 使用方法:使用时,先将要割锯的材料上所画的弧线内侧用钻头钻出洞,再将锯丝的一头穿过洞挂在另一段的小钉上,按照所画弧线内侧1左右进行锯割,锯割方向是斜向上下。 二、辅助工具及其使用方法 (一)钻床 是用来给模型打孔的设备。无论是在景观模型、景观模型还是在展示模型中,都会有很多的零部件需要镂空效果时,必须先要打孔。钻孔时,主要是依靠钻头与工件之间的相对运动来完成这个过程的。在具体的钻孔过程中,只有钻头在旋转,而被钻物体是静止不动的。 钻床分台式和立式两种。台式钻床是一种可以放在工台上操作的小型钻床,小巧、灵活,使

1第一章 8字模型与飞镖模型(1)

O D C B A 图12图E A B C D E F D C B A O O 图12图E A B C D E D C B A H G E F D C B A 第一章 8字模型与飞镖模型 模型1 角的“8”字模型 如图所示,AB 、CD 相交于点O , 连接AD 、BC 。 结论:∠A+∠D=∠B+∠C 。 模型分析 8字模型往往在几何综合 题目中推导角度时用到。 模型实例 观察下列图形,计算角度: (1)如图①,∠A+∠B+∠C+∠D+∠E= ; (2)如图②,∠A+∠B+∠C+∠D+∠E+∠F= 。 热搜精练 1.(1)如图①,求∠CAD+∠B+∠C+∠D+∠E= ; (2)如图②,求∠CAD+∠B+∠ACE+∠D+∠E= 。 2.如图,求∠A+∠B+∠C+∠D+∠E+∠F+∠G+∠H= 。

D C B A M D C B A O 135E F D C B A 105O O 120 D C B A 模型2 角的飞镖模型 如图所示,有结论: ∠D=∠A+∠B+∠C 。 模型分析 飞镖模型往往在几何综合 题目中推导角度时用到。 模型实例 如图,在四边形ABCD 中,AM 、CM 分别平分∠DAB 和∠DCB ,AM 与CM 交于M 。探究∠AMC 与∠B 、∠D 间的数量关系。 热搜精练 1.如图,求∠A+∠B+∠C+∠D+∠E+∠F= ; 2.如图,求∠A+∠B+∠C+∠D = 。

O D C B A O D C B A O C B A 模型3 边的“8”字模型 如图所示,AC 、BD 相交于点O ,连接AD 、BC 。 结论:AC+BD>AD+BC 。 模型实例 如图,四边形ABCD 的对角线AC 、BD 相交于点O 。 求证:(1)AB+BC+CD+AD>AC+BD ; (2)AB+BC+CD+AD<2AC+2BD. 模型4 边的飞镖模型 如图所示有结论: AB+AC>BD+CD 。

新整理描写常青藤优美句段 写常青藤作文散文句子

描写常青藤优美句段写常青藤作文散文句子 描写常青藤优美句段写常青藤作文散文句子第1段: 1.睁开朦胧的泪眼,我猛然发觉那株濒临枯萎的常春藤已然绿意青葱,虽然仍旧瘦小,却顽强挣扎,嫩绿的枝条攀附着窗格向着阳光奋力伸展。 2.常春藤是一种常见的植物,我家也种了两盆。可能它对于很多人来说都不足为奇,但是却给我留下了美好的印象。常春藤属于五加科常绿藤本灌木,翠绿的叶子就像火红的枫叶一样,是可爱的小金鱼的尾巴。常春藤的叶子的长约5厘米,小的则约有2厘米,但都是小巧玲珑的,十分可爱。叶子外圈是白色的,中间是翠绿的,好像有人在叶子上涂了一层白色的颜料。从叶子反面看,可以清清楚楚地看见那凸出来的,一根根淡绿色的茎。 3.渴望到森林里探险,清晨,薄薄的轻雾笼罩在树林里,抬头一看,依然是参天古木,绕着树干一直落到地上的常春藤,高高低低的灌木丛在小径旁张牙舞爪。 4.我们就像马蹄莲,永不分开,如青春的常春藤,紧紧缠绕。 5.我喜欢那里的情调,常春藤爬满了整个屋顶,门把手是旧的,但带着旧上海的味道,槐树花和梧桐树那样美到凋谢,这是我的上海,这是爱情的上海。 6.当我离别的时候,却没有你的身影;想轻轻地说声再见,已是人去楼空。顿时,失落和惆怅涌上心头,泪水也不觉悄悄滑落我伫立很久很久,凝望每一条小路,细数每一串脚印,寻找你

的微笑,倾听你的歌声――一阵风吹过,身旁的小树发出窸窸窣窣的声音,像在倾诉,似在安慰。小树长高了,还有它旁边的那棵常春藤,叶子依然翠绿翠绿,一如昨天。我心头不觉一动,哦,这棵常春藤陪伴我几个春秋,今天才惊讶于它的可爱,它的难舍,好似那便是我的生命。我蹲下身去。轻轻地挖起它的一个小芽,带着它回到了故乡,种在了我的窗前。 7.常春藤属于五加科常绿藤本灌木,翠绿的叶子就像火红的枫叶一样,是可爱的小金鱼的尾巴。常春藤的叶子的长约5厘米,小的则约有2厘米,但都是小巧玲珑的,十分可爱。叶子外圈是白色的,中间是翠绿的,好像有人在叶子上涂了一层白色的颜料。从叶子反面看,可以清清楚楚地看见那凸出来的,一根根淡绿色的茎。 8.常春藤是多么朴素,多么不引人注目,但是它的品质是多么的高尚,不畏寒冷。春天,它萌发出嫩绿的新叶;夏天,它郁郁葱葱;秋天,它在瑟瑟的秋风中跳起了欢快的舞蹈;冬天,它毫不畏惧呼呼作响的北风,和雪松做伴常春藤,我心中的绿色精灵。 9.可是对我而言,回头看到的只是雾茫茫的一片,就宛如窗前那株瘦弱的即将枯死的常春藤,毫无生机,早已失去希望。之所以叫常春藤,可能是因为它一年四季都像春天一样碧绿,充满了活力吧。也许,正是因为如此,我才喜欢上了这常春藤。而且,常春藤还有许多作用呢!知道吗?一盆常春藤能消灭8至10平

多领域建模理论与方法

XXX理工大学 CHANGSHA UNIVERSITY OF TECHNOLOGY&TECHNOLGY 题目:多领域建模理论与方法 学院: XXX 学生: XXX 学号: XXX 指导教师: XXX 2015年7月2日

多领域建模理论和方法 The theories and methods of Multi-domain Modeling Student:XXX Teacher:XXX 摘要 建模理论和方法是推动仿真技术进步和发展的重要因素,也是系统仿真可持续发展的基础[1]文中综述了多领域建模主要采用的四种方法,并重点对基于云制造的多领域建模和仿真进行了叙述,并对其发展进行了展望。 关键词:多领域建模仿真;云制造;展望 Abstract:The theory and method of system model building is not only the key factor to stimulate the development and improvement of simulation technique but also the base of system simulation. This paper analysis four prevails way in Multi-domain Modeling, especially to the Multi-domain Modeling and Simulation in cloud manufacturing environment. We give a detail on its development and future. Keywords: Multi-domain Modeling and simulation; Cloud manufacturing; Future development 一引言 随着科学技术的发展进步和产品的升级需求,对产品提出了更高的要求,使得建模对象的组成更加复杂,涉及到各个学科、进程的复杂性以及设计方法的多元化。这些需求都是以前单领域建模方案无法满足的,因此,必须建立一个建模方式在设计过程中完成对繁杂目标的多领域建模、结构仿真、多元化分析等。 多领域建模是将机械、控制、电子等不同学科领域的模型“组装”成一个更大的模型进行仿真。根据需要的不同,实际建模过程中,可以将模型层层分解。将不同领域的仿真模型“零件”组装成“部件”,“子系统”则是由不同学科下的部件装配而成,与此同时装配完成的不同学科的分子系统还能再装配成为一个全面仿真模型,称之为“系统”,由此可见多领域建模技术在繁杂产品设计过程中具有出众的优势。 本文对多领域建模常用的四种方法:基于各领域商用仿真软件接口的建模方法;基于高层体系结构的建模方法;基于统一建模语言的多领域建模方法和基于云制造环境下多领域建模的方法进行了分析并对基于云制造环境下多领域建模方法进行了展望。

关于美国常青藤

一、常青藤大学 目录 联盟概述 联盟成员 名称来历 常春藤联盟(The Ivy League)是指美国东北部八所院校组成的体育赛事联盟。这八所院校包括:布朗大学、哥伦比亚大学、康奈尔大学、达特茅斯学院、哈佛大学、宾夕法尼亚大学、普林斯顿大学及耶鲁大学。美国著名的体育联盟还有太平洋十二校联盟(Pacific 12 Conference)和大十联盟(Big Ten Conference)。常春藤联盟的体育水平在美国大学联合会中居中等偏下水平,远不如太平洋十校联盟和大十联盟。 联盟概述 常春藤盟校(Ivy League)指的是由美国东北部地区的八所大学组成的体育赛事联盟(参见NCAA词条)。它们全部是美国一流名校、也是美国产生最多罗德奖学金得主的大学联盟。此外,建校时间长,八所学校中的七所是在英国殖民时期建立的。 美国八所常春藤盟校都是私立大学,和公立大学一样,它们同时接受联邦政府资助和私人捐赠,用于学术研究。由于美国公立大学享有联邦政府的巨额拨款,私立大学的财政支出和研究经费要低于公立大学。 常青藤盟校的说法来源于上世纪的50年代。上述学校早在19世纪末期就有社会及运动方面的竞赛,盟校的构想酝酿于1956年,各校订立运动竞赛规则时进而订立了常青藤盟校的规章,选出盟校校长、体育主任和一些行政主管,定期聚会讨论各校间共同的有关入学、财务、援助及行政方面的问题。早期的常青藤学院只有哈佛、耶鲁、哥伦比亚和普林斯顿4所大学。4的罗马数字为"IV",加上一个词尾Y,就成了"IVY",英文的意思就是常青藤,所以又称为常青藤盟校,后来这4所大学的联合会又扩展到8所,成为如今享有盛誉的常青藤盟校。 这些名校都有严格的入学标准,能够入校就读的学生,必须是品学兼优的好学生。学校很早就去各个高中挑选合适的人选,许多得到全国优秀学生奖并有各种特长的学生都是他们网罗的对象。不过学习成绩并不是学校录取的惟一因素,学生是否具有独立精神并且能否快速适应紧张而有压力的大一新生生活也是他们考虑的重要因素。学生的能力和特长是衡量学生综合素质的重要一关,高中老师的推荐信和评语对于学生的入学也起到重要的作用。学校财力雄厚,招生办公室可以完全根据考生本人的情况录取,而不必顾虑这个学生家庭支付学费的能力,许多家境贫困的优秀子弟因而受益。有钱人家的子女,即使家财万贯,也不能因

角色模型制作流程

幻想之旅角色模型制作流程 1.拿到原画后仔细分析角色设定细节,对不清楚的结构、材质细节及角色身高等问题与 原画作者沟通,确定对原画理解准确无误。 2.根据设定,收集材质纹理参考资料。 3.开始进行低模制作。 4.制作过程中注意根据要求严格控制面数(以MAX为例,使用Polygon Counter工具查 看模型面数)。 5.注意关节处的合理布线,充分考虑将来动画时的问题。如有疑问与动作组同事讨论咨 询。 6.由于使用法线贴图技术不能使用对称复制模型,可以直接复制模型,然后根据具体情 况进行移动、放缩、旋转来达到所需效果。 7.完成后,开始分UV。 分UV时应尽量充分利用空间,注意角色不同部位的主次,优先考虑主要部位的贴图(例如脸,前胸以及引人注意的特殊设计),为其安排充分的贴图面积。使用Relax Tool 工具确保UV的合理性避免出现贴图的严重拉伸及反向。 8.低模完成后进入法线贴图制作阶段。 现在我们制作法线贴图的方法基本上有三种分别是: a.在三维软件中直接制作高模,完成后将低模与高模对齐,然后使用软件工具生成法 线贴图。 b.将分好UV的低模Export成OBJ格式文件,导入ZBrush软件。在ZB中添加细节 制作成高模,然后使用Zmapper插件生成法线贴图。 c.在Photoshop中绘制纹理或图案灰度图,然后使用PS的法线贴图插件将灰度图生成 法线贴图。 (具体制作方法参见后面的制作实例) 建议在制作过程中根据实际情况的不同,三种方法结合使用提高工作效率。 9.法线贴图完成后,将其赋予模型,查看法线贴图的效果及一些细小的错误。 10.进入Photoshop,打开之前生成的法线贴图,根据其贴在模型上的效果对法线贴图进行 修整。(例如边缘的一些破损可以使用手指工具进行修补,或者在绿色通道中进行适当的绘制。如需加强某部分法线贴图的凹凸效果可复制该部分进行叠加可以起到加强

什么是美国常青藤大学

https://www.wendangku.net/doc/bf1114657.html, 有意向申请美国大学的学生,大部分听过一个名字,常青藤大学联盟。那么美国常青藤大学盟校到底是怎么一回事,又是由哪些大大学组成的呢?下面为大家介绍一下美国常青藤大学联盟。 立思辰留学360介绍,常青藤盟校(lvy League)是由美国的七所大学和一所学院组成的一个大学联合会。它们是:马萨诸塞州的哈佛大学,康涅狄克州的耶鲁大学,纽约州的哥伦比亚大学,新泽西州的普林斯顿大学,罗德岛的布朗大学,纽约州的康奈尔大学,新罕布什尔州的达特茅斯学院和宾夕法尼亚州的宾夕法尼亚大学。这8所大学都是美国首屈一指的大学,历史悠久,治学严谨,许多著名的科学家、政界要人、商贾巨子都毕业于此。在美国,常青藤学院被作为顶尖名校的代名词。 常青藤由来 立思辰留学介绍,常青藤盟校的说法来源于上世纪的50年代。上述学校早在19世纪末期就有社会及运动方面的竞赛,盟校的构想酝酿于1956年,各校订立运动竞赛规则时进而订立了常青藤盟校的规章,选出盟校校长、体育主任和一些行政主管,定期聚会讨论各校间共同的有关入学、财务、援助及行政方面的问题。早期的常青藤学院只有哈佛、耶鲁、哥伦比亚和普林斯顿4所大学。4的罗马数字为“IV”,加上一个词尾Y,就成了“IVY”,英文的意思就是常青藤,所以又称为常青藤盟校,后来这4所大学的联合会又扩展到8所,成为现在享有盛誉的常青藤盟校。 这些名校都有严格的入学标准,能够入校就读的学生,自然是品学兼优的好学生。学校很早就去各个高中挑选合适的人选,许多得到全国优秀学生奖并有各种特长的学生都是他们网罗的对象。不过学习成绩并不是学校录取的惟一因素,学生是否具有独立精神并且能否快速适应紧张而有压力的大一新生生活也是他们考虑的重要因素。学生的能力和特长是衡量学生综合素质的重要一关,高中老师的推荐信和评语对于学生的入学也起到重要的作用。学校财力雄厚,招生办公室可以完全根据考生本人的情况录取,而不必顾虑这个学生家庭支付学费的能力,许多家境贫困的优秀子弟因而受益。有钱人家的子女,即使家财万贯,也不能因此被录取。这也许就是常青藤学院历经数百年而保持“常青”的原因。

模型制作方法

动画精度模型制作与探究 Animation precision model manufacture and inquisition 前言 写作目的:三维动画的制作,首要是制作模型,模型的制作会直接影响到整个动画的最终效果。可以看出精度模型与动画的现状是随着电脑技术的不断发展而不断提高。动画模型走精度化只是时间问题,故精度模型需要研究和探索。 现实意义:动画需要精度模型,它会让动画画面更唯美和华丽。游戏需要精度模型,它会让角色更富个性和激情。广告需要精度模型,它会让物体更真实和吸引。场景需要精度模型,它会让空间更加开阔和雄伟。 研究问题的认识:做好精度模型并不是草草的用基础的初等模型进行加工和细化,对肌肉骨骼,纹理肌理,头发毛发,道具机械等的制作更是需要研究。在制作中对于层、蒙版和空间等概念的理解和深化,及模型拓扑知识与解剖学的链接。模型做的精,做的细,做的和理,还要做的艺术化。所以精度模型的制作与研究是很必要的。 论文的中心论点:对三维动画中精度模型的制作流程,操作方法,实践技巧,概念认知等方向进行论述。 本论 序言:本设计主要应用软件为Zbrsuh4.0。其中人物设计和故事背景都是以全面的讲述日本卡通人设的矩阵组合概念。从模型的基础模型包括整体无分隔方体建模法,Z球浮球及传统Z球建模法(对称模型制作。非对称模型制作),分肢体组合建模法(奇美拉,合成兽),shadow box 建模和机械建模探索。道具模型制作,纹理贴图制作,多次用到ZBURSH的插件,层概念,及笔刷运用技巧。目录: 1 角色构想与场景创作 一初步设计:角色特色,形态,衣装,个性矩阵取样及构想角色的背景 二角色愿望与欲望。材料采集。部件及相关资料收集 三整体构图和各种种类基本创作 2 基本模型拓扑探究和大体模型建制 3 精度模型大致建模方法 一整体无分隔方体建模法 二Z球浮球及传统Z球建模法(对称模型制作。非对称模型制作) 三分肢体组合建模法(奇美拉,合成兽) 四shadow box 建模探索和机械建模 4 制作过程体会与经验:精度细节表现和笔刷研究 5 解剖学,雕塑在数码建模的应用和体现(质量感。重量感。风感。飘逸感)

2019年美国常春藤八所名校排名

2019年美国常春藤八所名校排名享有盛名的常春藤盟校现在是什么情况呢?接下来就来为您介绍一下!以下常春藤盟校排名是根据2019年美国最佳大学进行的。接下来我们就来看看各个学校的状态以及真实生活。 完整的常春藤盟校名单包括耶鲁大学、哈佛大学、宾夕法尼亚大学、布朗大学、普林斯顿大学、哥伦比亚大学、达特茅斯学院和康奈尔大学。 同时我们也看看常春藤盟校是怎么样的?也许不是你所想的那样。 2019年Niche排名 3 录取率5% 美国高考分数范围1430-1600 财政援助:“学校选择美国最优秀的学生,想要他们来学校读书。如果你被录取,哈佛会确保你能读得起。如果你选择不去入学的话,那一定不是因为经济方面的原因。”---哈佛大三学生2019年Niche排名 4 录取率6% 美国高考分数范围1420-1600 学生宿舍:“不可思议!忘记那些其他学校的学生宿舍吧。在耶鲁,你可以住在一个豪华套房,它更像是一个公寓。一个公寓有许多人一起住,包括一个公共休息室、洗手间和多个卧室。我再不能要求任何更好的条件了。这个套房很大,很干净,还时常翻修。因为学校的宿舍深受大家喜爱,现在有90%的学生都住在学校!”---耶鲁大二学生

2019年Niche排名 5 录取率7% 美国高考分数范围1400-1590 综合体验:“跟任何其他学校一样,普林斯顿大学有利有弊。这个学校最大的好处也是我选择这个学校的主要原因之一就是它的财政援助体系,任何学生想要完成的计划,它都会提供相应的财政支持。”---普林斯顿大二学生 2019年Niche排名 6 录取率9% 美国高考分数范围1380-1570 自我关心:“如果你喜欢城市的话,宾夕法尼亚大学是个不错的选择。这里对于独立的人来说也是一个好地方,因为在这里你必须学会自己发展。要确保进行一些心理健康的训练,因为这里的人通常会过量工作。如果你努力工作并且玩得很嗨,二者都会使你精疲力尽,所以给自己留出点儿时间休息。”---宾夕法尼亚大一学生 2019年Niche排名7 录取率7% 美国高考分数范围1410-1590 综合体验:“学校的每个人都很关心学生,包括我们的身体状况和学业成绩。在这里,你可以遇到来自世界各地的多种多样的学生。他们在学校进行的安全防范教育让我感觉受到保护。宿舍生活非常精彩,你会感觉跟室友们就像家人一样。总之,能成为学校的一员我觉得很棒,也倍感荣幸!”---哥伦比亚大二学生2019年Niche排名9 录取率9% 美国高考分数范围1370-1570 学术点评:“新的课程培养学术探索能力,在过去的两年中我

NPT型IGBT电热仿真模型参数提取方法综述_徐铭伟

电力自动化设备 Electric Power Automation Equipment Vol.33No.1Jan.2013 第33卷第1期2013年1月 0引言 近年来,绝缘栅双极型晶体管IGBT (Insulated Gate Bipolar Transistor )因其不断改善的电压、电流承受能力和工作频率、功率损耗等性能指标而被广泛应用到机车牵引、开关电源、新能源发电等电能变换和处理领域中[1],因此IGBT 的可靠性受到国内外科研工作者的广泛关注。研究表明,与IGBT 器件结温(T j )相关的热循环过程和器件封装材料热膨胀系数不一致是致其故障的主要诱因[2-3],IGBT 的电热仿真模型可以估计结温的变化情况,从而可用于IGBT 可靠性的评估。国内外对IGBT 的电热仿真模型开展了大量研究工作[4-6],其中基于半导体物理并考虑自热效应(Self -heating )的IGBT A.R.Hefner 器件模型[6] 和反映其封装传热过程的Cauer 网络[7-9]联合组成的IGBT 电热模型准确度较高,并已在Saber 、Pspice 等电路仿真软件中得到应用[10-11],但是,仿真软件有限的器件模型库无法满足仿真需要,同时出于技术保密的缘故,半导体制造商并不会提供建立电热模型需要的模型参数,因此如何建立一种有效并准确的参数提取方法就显得十分必要。 IGBT 电热仿真模型参数同半导体物理、器件以 及封装结构直接相关,无法直接测量,只能通过一定 的技术方法和手段获取。一个有效的参数提取过程是获得有效的电热模型的前提条件;此外,实现模型参数的准确提取对于分析IGBT 的性能、优化驱动电路的设计、指导其应用以及选型都具有重要意义。在参数提取之后,有效性验证也至关重要,可以让使用者合理选择器件的工作范围。由于非穿通(NPT )型 IGBT 目前在工业领域中已获得了广泛而成熟的应 用[12],本文将以其作为参数提取的研究对象。本文从NPT 型IGBT 电热仿真模型的工作原理出发,首先将模型参数分为电参数和热参数两大类。然后对近年来模型参数提取方法的研究情况进行讨论,依据提取手段的不同将文献中出现的IGBT 电参数提取方法归纳为4类:仿真提取[13];经验估计,如利用经验公式[12,14-18]、数据手册[15-16]或者参数典型范围[12];参数隔离[19-27];参数优化,包括直接搜索技术[14]、模拟退火算法[28-29]、变量轮换法[30-32]等。同时归纳Cauer 网络的参数提取可以从IGBT 的封装结构[8-9,33-34]和封装瞬态热阻曲线[7,35-36]2个方向出发,并列表给出了提取电参数和热参数的不同方法之间的优缺点。最后对各种提取方法进行了总结,并讨论了一个模型电参数提取步骤,以增强参数提取工作的有序性和可靠性,这对于提高IGBT 电热仿真模型的应用水平,扩大其使用范围起到了积极的作用。 1IGBT 电热仿真模型及其参数 IGBT 的电热仿真模型是建立在考虑了半导体 自热效应的Hefner 物理模型基础之上,耦合了受结温影响的器件模型及与散热路径相关的动态热模型。在分析器件损耗特性、辅助电力电子设计以及研究因器件老化衰退引起的变换器端口特性等方面, NPT型IGBT电热仿真模型参数提取方法综述 徐铭伟,周雒维,杜 雄,沈 刚,杨 旭 (重庆 大学输配电装备及系统安全与新技术国家重点实验室,重庆400044) 摘要:对NPT 型IGBT 电热仿真模型的工作原理进行了概述,并将模型参数分为电参数(即基于半导体物理的Hefner 器件模型参数)和热参数(即反映器件封装传热的Cauer 网络参数)两大类,然后对近年来模型参数提取方法的研究情况进行讨论。依据提取技术手段的不同将IGBT 电参数提取方法归纳为仿真提取、经验估计、参数隔离和参数优化4类,并从时效性、准确性、复杂性等方面对各种方法进行了比较和评价;从IGBT 的封装结构和封装瞬态热阻曲线2个方向出发讨论了Cauer 网络参数的提取。最后讨论了一个模型电参数的提取步骤。 关键词:绝缘栅双极型晶体管;电热;仿真;模型;参数提取;热网络;电参数;热参数中图分类号:TM 322 文献标识码:A DOI :10.3969/j.issn.1006-6047.2013.01.026 收稿日期:2011-08-09;修回日期:2012-10-19 基金项目:科技部国际合作项目(2010DFA72250);国家自然科学基金资助项目(51077137);输配电装备及系统安全与新技术国家重点实验室重点资助项目(2007DA10512711101);中央高校基本科研业务费资助项目(CDJXS11150022) Project supported by the International Cooperation Project of the Minister of Science and Technology of China (2010DFA -72250),the National Natural Science Foundation of China (51077137),the Key Program in State Key Laboratory of Power Transmission Equipment &System Security and New Tech -nology (2007DA10512711101)and the Fundamental Research Funds for the Central Universities of China (CDJXS11150022)

美国常青藤大学研究生申请条件都有哪些

我国很多学子都想前往美国的常青藤大学就读于研究生,所以美国常青藤大学研究生申请条件都有哪些? 美国常青藤大学研究生申请条件: 1、高中或本科平均成绩(GPA)高于3.8分,通常最高分是4分,平均分越高越好; 2、学术能力评估测试I(SAT I,阅读+数学)高于1400分,学术能力评估测试II(SAT II,阅读+数学+写作)高于2000分; 3、托福考试成绩100分以上,雅思考试成绩不低于7分; 4、美内国研究生入学考试(GRE)成绩1400分以上,经企管理研究生入学考试(GMAT)成绩700分以上。 大学先修课程(AP)考试成绩并非申请美国大学所必需,但由于大学先修课程考试对于高中生来说有一定的挑战性及难度,美国大学也比较欢迎申请者提交大学先修课程考试的成绩,作为入学参考标准。

有艺术、体育、数学、社区服务等特长者优先考容虑。获得国际竞赛、辩论和科学奖等奖项者优先考虑,有过巴拿马国际发明大赛的得主被破例录取的例子。中国中学生在奥林匹克数、理、化、生物比赛中获奖也有很大帮助。 常春藤八所院校包括:哈佛大学、宾夕法尼亚大学、耶鲁大学、普林斯顿大学、哥伦比亚大学、达特茅斯学院、布朗大学及康奈尔大学。 新常春藤包括:加州大学洛杉矶分校、北卡罗来纳大学、埃默里大学、圣母大学、华盛顿大学圣路易斯分校、波士顿学院、塔夫茨大学、伦斯勒理工学院、卡内基梅隆大学、范德比尔特大学、弗吉尼亚大学、密歇根大学、肯阳学院、罗彻斯特大学、莱斯大学。 纽约大学、戴维森学院、科尔盖特大学、科尔比学院、瑞德大学、鲍登学院、富兰克林欧林工程学院、斯基德莫尔学院、玛卡莱斯特学院、克莱蒙特·麦肯纳学院联盟。 小常春藤包括:威廉姆斯学院、艾姆赫斯特学院、卫斯理大学、斯沃斯莫尔学院、明德学院、鲍登学院、科尔比学院、贝茨学院、汉密尔顿学院、哈弗福德学院等。

计算机仿真技术概述及其在交通仿真领域的应用

计算机仿真技术简介 计算机仿真技术是一门综合性信息技术,它通过专用软件,整合图像、声音、动画等,将三维的现实环境、物体模拟成多维表现形式的计算机仿真,再由数字媒介作为载体传播给人们。当人们通过该媒体浏览观赏时就如身临其境一般。并且可以选择任意角度,观看任意范围内的场景或选择观看物体的任意角度。正是由于对身临其境的真实感和对超越现实的虚拟性,以及建立个人能够沉浸其中、超越其上、进出自如、具有交互作用的多维信息系统的追求,推动了计算机仿真技术在各个领域中的应用与发展。并且,因其有效性、经济性、安全性、直观性等特点而受到广泛的应用。它是在计算机图形学基础上发展起来的一种仿真应用技术。 计算机仿真已成为系统仿真的一个重要分支,系统仿真很大程度上指的就是计算机仿真。计算机仿真技术的发展与控制工程、系统工程及计算机工程的发展有着密切的联系。一方面,控制工程、系统工程的发展,促进了仿真技术的广泛应用;另一方面,计算机的出现以及计算机技术的发展,又为仿真技术的发展提供了强大的支撑。工业方面,计算机仿真一直作为一种必不可少的工具,在减少损失、节约经费开支、缩短开发周期、提高产品质量等方面发挥着重要的作用。 综上所述,计算机仿真技术是以数学理论、相似原理、信息技术、系统技术及其应用领域有关的专业技术为基础,以计算机和各种物理效应设备为工具,利用系统模型对实际的或设想的系统进行试验研究的一门综合性技术。它集成了计算机技术、网络技术、图形图象技术、面向对象技术、多媒体、软件工程、信息处理、自动控制等多个高新技术领域的知识。 计算机仿真技术原理 对于需要研究的对象,计算机一般是不能直接认知和处理的,这就要求为之建立一个既能反映所研究对象的实质,又易于被计算机处理的数学模型。关于研究对象、数学模型和计算机之间的关系,可以用图1来表示。

初中数学优质专题:8字模型与飞镖模型

1 O D C B A 图1 2图E A B C D E F D C B A O O 图12图E A B C D E D C B A 第一章 8字模型与飞镖模型 模型1 角的“8”字模型 如图所示,AB 、CD 相交于点O , 连接AD 、BC 。 结论:∠A+∠D=∠B+∠C 。 模型分析 8字模型往往在几何综合 题目中推导角度时用到。 模型实例 观察下列图形,计算角度: (1)如图①,∠A+∠B+∠C+∠D+∠E= ; (2)如图②,∠A+∠B+∠C+∠D+∠E+∠F= 。 热搜精练 1.(1)如图①,求∠CAD+∠B+∠C+∠D+∠E= ; (2)如图②,求∠CAD+∠B+∠ACE+∠D+∠E= 。

2 H G E F D C B A D C B A M D C B A O 135 E F D C B A 2.如图,求∠A+∠B+∠C+∠D+∠E+∠F+∠G+∠H= 。 模型2 角的飞镖模型 如图所示,有结论: ∠D=∠A+∠B+∠C 。 模型分析 飞镖模型往往在几何综合 题目中推导角度时用到。 模型实例 如图,在四边形ABCD 中,AM 、CM 分别平分∠DAB 和 ∠DCB ,AM 与CM 交于M 。探究∠AMC 与∠B 、∠D 间的数量关系。

3 105O O 120 D C B A O D C B A 热搜精练 1.如图,求∠A+∠B+∠C+∠D+∠E+∠F= ; 2.如图,求∠A+∠B+∠C+∠D = 。 模型3 边的“8”字模型 如图所示,AC 、BD 相交于点O ,连接AD 、BC 。 结论:AC+BD>AD+BC 。

2021中考数学易错题飞镖模型8字模型探究试题

2021中考数学易错题飞镖模型8字模型探究试题模型一:角的飞镖模型基础 结论:C + ∠ ∠ = ∠ B + A BDC∠ 解答: ①方法一:延长BD交AC于点E得证 ②方法二:延长CD交AB于点F得证 ③方法三:延长AD到在其延长方向上任取一点为点G得证 总结: ①利用三角形外角的性质证明

模型二:角的8字模型基础结论:D ∠ ∠ = + + C B A∠ ∠

解答: ①方法一:三角形内角和得证 ②方法二:三角形外角【BOD 】的性质得证总结: ①利用三角形内角和等于 180证明 推出 ②利用三角形外角的性质证明

角的飞镖模型和8字模型进阶 【例1】如图,则= ∠E D B A + C + + ∠ ∠ ∠ + ∠ 解答: ①方法一:飞镖ACD得证 ∠E + D C A B ∠ ∠ = 180 ∠ + + ∠ +

②方法二:8字BECD得证 + ∠ ∠E B A + C D ∠ = + 180 + ∠ ∠ 【例2】如图,则= E ∠F + D C A B ∠ ∠ ∠ + + ∠ ∠ + + 解答:飞镖ABF+飞镖DEC得证 ∠F + ∠ E D B + A C ∠ = ∠ + 210 ∠ ∠ + + 【例3】如图,求= E D ∠F B A + C ∠ + ∠ + ∠ ∠ + ∠ + 解答:8字模型得证 ∠F + ∠ E D A B C + 360 + = ∠ ∠ ∠ + ∠ + 【例4】如图,求= ∠D C A + B ∠ + ∠ + ∠

解答:连接BD得飞镖BAD+飞镖DBC得证 + ∠D A ∠ C B = + ∠ 220 + ∠ 【例5】如图,求= ∠H G ∠ F + D A C + E B + ∠ + ∠ ∠ + + ∠ + ∠ ∠ 解答:飞镖EHB+飞镖FAC得证 ∠H ∠ + + ∠ G F A B C D E ∠ + + = 360 ∠ ∠ ∠ + + ∠ + 模型三:边的飞镖模型基础 结论:CD + > AC BD AB+

相关文档
相关文档 最新文档