Apple на wwdc 2012 |
Перевод с английского | ||
<< Theory W Software Management | На зоопарк фон >> |
Автор: Yang Richard Yang. Чтобы познакомиться с картинкой полного размера, нажмите на её эскиз. Чтобы можно было использовать все картинки для урока английского языка, скачайте бесплатно презентацию «Apple на wwdc 2012.ppt» со всеми картинками в zip-архиве размером 4756 КБ.
Сл | Текст | Сл | Текст |
1 | Mobile Software Development Framework: | 32 | maintains/monitors available resources no |
Adaptive Mobile Applications. 10/23/2012 | need to have each application re-implement | ||
Y. Richard Yang. 1. | the monitoring An application registers a | ||
2 | Outline. Admin Mobile cloud services | resource descriptor and an upcall event | |
Push notification service Track service | handler with the OS OS notifies the | ||
Storage service. 2. | application upon detecting resource | ||
3 | Admin. HW3 Project ideas. 3. | changes Application adjusts requests to | |
4 | Recap: Example Mobile Cloud Services. | the server. 32. | |
Push notification service Location based | 33 | Client System Model. 33. | |
service, e.g., Track service (supporting | 34 | Outline. Admin and recap Adaptive | |
location based services) Storage services, | mobile applications Client adaptation | ||
e.g., iCloud, Google Drive, Dropbox Proxy | Server adaptation. 34. | ||
service, e.g., Kindle Split Browser | 35 | Server/Content Adaptation: Examples. | |
Recognition/synthesis services. 4. | Objective: automating adaptation. 35. | ||
5 | Recap: Push Notification Service. A | 36 | Example: Adapting Audio Content. Send |
single persistent connection between Push | a lower resolution stream as the redundant | ||
Notification Service (PNS) and Mobile | information, e.g. nominal stream at 64 | ||
Device Authorization App register with PNS | kbps and redundant stream at 13 kbps (such | ||
Device app registers with PNS and forwards | as GSM). 2. 3. 36. | ||
registration to PNS App can send only to | 37 | Example: Adapting Fidelity of | |
registrations Scalability, fault | Video/Image Content? Potentially many | ||
tolerance, generality Send only | dimensions frame rate (for video) image | ||
notification, does not hold data. 5. | size quality of image Usage: e.g., data | ||
6 | Recap: Track Service. Creation. TC | acceleration offered by many carriers. 37. | |
MakeCollection(GroupCriteria criteria, | 38 | Frame Encoding: Block Transform | |
bool removeDuplicates). Manipulation. TC | Encoding. DCT. Quantize. Zig-zag. | ||
JoinTrackCollections (TC tCs[], bool | 011010001011101... 38. | ||
removeDuplicates) TC SortTracks (TC tC, | 39 | Discrete Cosine Transform. ? ? | |
SortAttribute attr) TC TakeTracks(TC tC, | 4C(u)C(v). (2j+1)up. (2k+1)vp. f(j,k) cos. | ||
int count) TC GetSimilarTracks (TC tC, | cos. F[u,v] =. n2. 2n. 2n. where. 1. for | ||
Track refTrack, float simThreshold) TC | w=0. ? 2. C(w) =. for w=1,2,…,n-1. 1. DCT | ||
GetPassByTracks (TC tC, Area[] areas) TC | is better at reducing redundancy than | ||
GetCommonSegments(TC tC, float | Discrete Fourier Transform but it is more | ||
freqThreshold). 6. | computationally expensive. n-1 n-1. j=0 | ||
7 | API Usage: Ride-Sharing Application. | k=0. 39. | |
// get user’s most popular track in the | 40 | Basis Functions of DCT. An image is a | |
morning TC myTC = MakeCollection(“name = | superposition of basis functions DCT | ||
Maya”, [0800 1000], true); TC myPopTC = | computes the contribution of each basis | ||
SortTracks(myTC, FREQ); Track track = | function - F[u,v]: for the basis function | ||
GetTracks(myPopTC, 0, 1); // find tracks | at position [u, v]. 40. | ||
of all fellow employees TC msTC = | 41 | Example: MPEG Block Encoding. | |
MakeCollection(“name.Employer = MS”, [0800 | Quantize. DCT. original image. zigzag. | ||
1000], true); // pick tracks from the | run-length and Huffman encoding of the | ||
community most similar to user’s popular | stream. 10011011100011... Discussion: how | ||
track TC similarTC = | to generate different encoding rates? DC | ||
GetSimilarTracks(msTC, track, 0.8); | component. AC components. coded bitstream | ||
Track[] similarTracks = | < 10 bits (0.55 bits/pixel). 41. | ||
GetTracks(similarTC, 0, 20); // Verify if | 42 | Examples. 42. | |
each track is frequently traveled by its | 43 | Outline. Admin and recap Adaptive | |
respective owner User[] result = | mobile applications Client adaptation | ||
FindOwnersOfFrequentTracks(similarTracks); | Server adaptation Proxy and job partition. | ||
8 | Outline. Admin and recap Mobile Cloud | 43. | |
Services Push notification service | 44 | Example Benefit of Cloud (RTT -> | |
Location based service Storage service. 8. | Server). 44. | ||
9 | Storage/Sync Service. Store content in | 45 | Example: MAUI. Maui server. |
cloud to be accessible by multiple | Smartphone. Application. Application. Maui | ||
selected devices Deployed services, e.g., | Controller. 45. | ||
iCloud Google Drive | 46 | How Does a Programmer Use MAUI? Goal: | |
https://developers.google.com/drive/integr | make it dead-simple to MAUI-ify apps Build | ||
te-android-ui DropBox. 9. | app as a standalone phone app Add .NET | ||
10 | Storage/Sync Service Example: iCloud. | attributes to indicate “remoteable” Follow | |
Backend Hosted by Windows Azure and Amazon | a simple set of rules. | ||
AWS Uses HTTPS to send to servers Client | 47 | MAUI Proxy. Maui server. Smartphone. | |
storage models Key Value Store UIDocument | Application. Application. Handles Errors. | ||
Core data More details: see iCloud | Maui Controller. | ||
sessions | 48 | MAUI Profiler. Profiler. Annotated | |
https://developer.apple.com/videos/wwdc/20 | Callgraph. CPU Cycles. State size. Device | ||
2/?id=209. 10. | Profile. Execution Time. Network Latency. | ||
11 | Problems Caused by Mobile Storage. | Callgraph. Network Bandwidth. | |
Read miss stalls progress (user has to | 49 | MAUI Solver. A sample callgraph. C | |
wait for data) Synchronization/consistency | 5000 mJ 3000 ms. 10000 mJ. B 900 mJ 15ms. | ||
user may see outdated data user | 1000mJ. 25000 mJ. D 15000 mJ 12000 ms. | ||
modification may generate conflicts. 11. | Energy and delay for state transfer. A. | ||
12 | Approaches. Read miss explicit user | Computation energy and delay for | |
file selection automatic | execution. | ||
hoarding/prediction, e.g., CODA, SEER | 50 | Is Global Program Analysis Needed? | |
Synchronization/consistency keep | InitializeFace Recognizer 5000 mJ. 10000 | ||
modification logs and develop merge tools, | mJ. FindMatch 900 mJ. User Interface. | ||
e.g., Bayou efficient file comparisons and | 1000mJ. 25000 mJ. DetectAndExtract Faces | ||
merging, e.g., rsync, LBFS. 12. | 15000 mJ. Cheaper to do local. | ||
13 | Amazon Silk Split-Browser. Dynamic | 51 | Is Global Program Analysis Needed? |
split browsing Intelligently partition | InitializeFace Recognizer 5000 mJ. 10000 | ||
work between local and Amazon cloud. 13. | mJ. FindMatch 900 mJ. User Interface. | ||
http://www.extremetech.com/mobile/97587-am | 1000mJ. 25000 mJ. DetectAndExtract Faces | ||
zon-silk-bridging-the-gap-between-desktop- | 15000 mJ. Cheaper to do local. Cheaper to | ||
nd-tablet-web-browsers. | do local. | ||
14 | Outline. Admin and recap Adaptive | 52 | Is Global Program Analysis Needed? |
mobile applications. 14. | 20900mJ. InitializeFace Recognizer. User | ||
15 | Adaptive Mobile App: Bigger Picture. | Interface. FindMatch. 1000mJ. | |
in-net proxy. device. in-net service. | DetectAndExtract Faces. Cheaper to | ||
On-device app/sys adaptation. Device-aware | offload. | ||
service delivery. Service partition. 15. | 53 | MAUI Implementation. Platform Windows | |
16 | Outline. Admin and recap Adaptive | Mobile 6.5 .NET Framework 3.5 HTC Fuze | |
mobile applications Client adaptation. 16. | Smartphone Monsoon power monitor | ||
17 | Example: iphone 5. | Applications Chess Face Recognition Arcade | |
http://www.anandtech.com/show/6324/the-iph | Game Voice-based translator. | ||
ne-5-performance-preview. 17. | 54 | Questions. How much can MAUI reduce | |
18 | CPU Power Consumption Model. The power | energy consumption? How much can MAUI | |
consumption rate P of a CMOS processor | improve performance? Can MAUI Run | ||
satisfies where k is a constant, C the | Resource-Intensive Applications? | ||
capacitance of the circuit, f the CPU | 55 | How much can MAUI reduce energy | |
frequency, and V the voltage When the | consumption? Face Recognizer. Big savings | ||
supply voltage V is lower, | even on 3G. An order of magnitude | ||
charging/discharging time is longer; thus | improvement on Wi-Fi. | ||
frequency should be lower => P ~ O(V3). | 56 | How much can MAUI improve performance? | |
18. | Face Recognizer. Improvement of around an | ||
19 | CPU Power Model. Discussion: what | order of magnitude. | |
voltage to operate on? throughput. 19. | 57 | Latency to server impacts the | |
20 | Dynamic Voltage Scaling. Basic idea: | opportunities for fine-grained offload. | |
determining voltage according to program | Arcade Game. Solver would decide not to | ||
response time requirement For normal | offload. Up to 40% energy savings on | ||
applications, give reasonable response | Wi-Fi. | ||
time For multimedia applications, use the | 58 | Can MAUI Run Resource-Intensive | |
deadline to determine voltage. 20. | Applications? Translator. Can be run on | ||
21 | Architecture. multimedia applications. | the phone with MAUI. CPU Intensive even on | |
monitoring. requirements. scheduling. | a Core 2 Duo PC. | ||
scheduler. profiler. speed adaptor. CPU. | 59 | Can MAUI Adapt to Changing Conditions? | |
demand distribution. time constraint. | 11KB + missiles. 11KB + missiles. 11KB + | ||
speed scaling. 21. | missiles. missiles. Required state is | ||
22 | Demand Prediction. Online profiling | smaller. Complexity increases with # of | |
and estimation: count number of cycles | missiles. *Missiles take around 60 bytes | ||
used by each job. CDF F(x) = P [X ? x]. 1. | each. | ||
cumulative probability. Cmin=b0. br=Cmax. | 60 | Case 1. Zero Missiles Low latency (RTT | |
br-1. 22. | < 10ms). HandleEnemies. DoFrame. | ||
23 | Observations. Demand distribution is | DoLevel. HandleBonuses. Offload starting | |
stable or changes slowly. 23. | at DoLevel. HandleMissiles. Computation | ||
24 | CPU Resource Allocation. How many | cost is close to zero. *Missiles take | |
cycles to allocate to a multimedia job? | around 60 bytes each. | ||
Application should meet ? percent of | 61 | Case 2. 5 Missiles Some latency (RTT = | |
deadlines ? each job meets deadline with | 50ms). HandleEnemies. DoFrame. DoLevel. | ||
probability ? ? allocate C cycles, such | HandleBonuses. Very expensive to offload | ||
that F (C ) =P [X ? C ] ? ? 24. | everything. Little state to offload. | ||
25 | How Fast to Run the CPU? Assume the | HandleMissiles. Only offload Handle | |
strategy is to run job i at a fix (also | Missiles. Most of the computation cost. | ||
called uniform) speed Si Assume it needs | *Missiles take around 60 bytes each. | ||
Ci cycles during a time duration of Pi | 62 | Bayou Write Operation: An Example. 62. | |
Fact: since power is a convex function of | 63 | Can MAUI Adapt to Changing Conditions? | |
frequency, if a job needs C cycles in a | Adapt to: Network Bandwidth/Latency | ||
period P, then the optimal frequency is | Changes Variability on method’s | ||
C/P, namely the lowest constant frequency. | computational requirements Experiment: | ||
25. | Modified off the shelf arcade game | ||
26 | Why Not Uniform Speed? Intuitively, | application Physics Modeling (homing | |
uniform speed achieves - minimum energy if | missiles) Evaluated under different | ||
use the allocated exactly However, jobs | latency settings. | ||
use cycles statistically - often complete | 64 | Bayou Write Operation: An Example. 64. | |
before using up the allocated - potential | 65 | Motivation. The CODA system assumes | |
to save more energy ? stochastic DVS. 26. | that modifications are kept as logs (CML) | ||
27 | Stochastic DVS. For each job find | a user sends the logs to the servers to | |
speed Sx for each allocated cycle x time | update If the storage of a client is | ||
is 1/Sx and energy is (1 - F(x))S3x. 27. | limited, it may not be able to save logs | ||
28 | Example Speed Schedule. cycle: speed: | then upon reconnection, the cache manager | |
100 MHz. 200 MHz. 400 MHz. Observation: | needs to find the difference between the | ||
speed up the processor with increasing | stored file and its local cached copy same | ||
clock cycles. 28. | problem exists for the rsync tool ! | ||
29 | DVS. A1. A2. A1. context switch Store | Question: how to efficiently compare the | |
speed for switched-out New speed for | differences of two remote files (when the | ||
switched-in. speed up within job. switch | network connection is slow)? 65. | ||
back. speed. execution. 29. | 66 | LBFS: Low-Bandwidth File System. Break | |
30 | Implementation. Hardware: HP N5470 | Files into chunks and transfer only | |
laptop Athlon CPU (300, 500, 600, 700, | modified chunks Fixed chunk size does not | ||
800, 1000MHz) round speed schedule to | work well why? 66. | ||
upper bound. system call. process control | 67 | Flexible Chunk Size. Compute hash | |
block. DVS modules PowerNow speed scaling | value of every 48 byte block if the hash | ||
Soft real-time scheduling. standard Linux | value equals to a magic value, it is a | ||
scheduler. Extension to Linux kernel | chunk boundary. 67. | ||
2.4.18 716 lines of C code. 30. | 68 | Bayou: Managing Update Conflicts. | |
31 | Evaluation: Normalized Energy. Reduces | Basic idea: application specific conflict | |
power consumption However, limited due to | detection and update Two mechanisms for | ||
few speed options. 31. | automatic conflict detection and | ||
32 | Odyssey: An Example Client | resolution dependency check merge | |
Architecture. Application indicates | procedure. | ||
resource capabilities in its request to | http://zoo.cs.yale.edu/classes/cs422/2011/ | ||
service Operating system | ib/terry95managing.pdf. 68. | ||
Apple на wwdc 2012.ppt |
«Переводчик с английского на русский» - Основные этапы. Плюсы: минимальная компьютерная подготовка переводчика. Ваш Проводник в Мире Многоязычной Информации. Выравнивание памяти переводов. Перевод с использование памяти переводов. Минимизация работы по подвёрстке ПО. Минусы: поддержка памяти переводов. Услуги. Минусы: большие затраты на вычитку.
«Перевод названий фильмов» - Функции заголовка. Некоторые любят погорячее. Интересные примеры. Наиболее употребляемый тип перевода. Результаты опроса. Особенности перевода. Трансформация. Название фильма в оригинале. Особенности перевода названий англоязычных фильмов. Замена названия. Типы перевода названий фильмов. Название. Перевод названия фильма.
«Профессиональный перевод» - Одобрение — ОК Неодобрение, осуждение — ОК. УПС базируется на сокращениях, аббревиатурах как широко распространенных, так и индивидуальных. Как записывать? Желательно владение переводческой нотацией, или сокращенной записью. Сокращения в УПС. Сокращенная буквенная запись; Аббревиация; Цифровое обозначение; Символизация.
«Перевод текстов» - Область применения PROMT Translation Suite. Преимущества. Революция в индустрии перевода - уникальный продукт PROMT Translation Suite. Перевод сегментов текста. Сохранение результатов редактирования и перевода в базе. Машинный перевод. Технология Translation Memory ™. Пополнение базы переводов. Общая схема работы с PROMT Translation Suite.
«Перевод рекламных текстов» - Advertising. Особенности использования. Сравнения. Особенности перевода рекламных текстов. Актуальность проблемы. Значение слова «реклама». Средства художественной выразительности. Соединения. Проблемы, связанные с переводом рекламных текстов. Функции рекламных текстов. Прилагательные и наречия. Универсальность.
«Технический английский» - Упражнения. Электронное пособие по изучению курса технического перевода. Конкурсы. Перевод интернационализмов. Префиксы – приставки. Классификация перевода. Требования к практическому владению навыками перевода. Русские и английские «Кулибины». Классификация по качеству. Полисемия – случай, когда слово имеет более одного значения.