Phone, Tablet and Gizmo Thread #0x02

[SaiK]

as days of software engineers keep fighting for parallelism and mutual exclusions, you would not wait for long.. the hardware is going to handle that and compiled for soon. all you have to focus is on the business logic, while compiler and hardware will take care of the rest.

--

ps: matrimc, who would be the gizmo gal here?

[Raja Bose]

Anujan-ullah, what you are saying is valid but must also take cognizance of the fact that once such higher levels of abstraction are reached, as we are seeing now there will be essentially 2 classes (or more) of technical people in the ITvity field. One will be the core engineers who actually build the systems group up whether HW or SW and one will be the application oriented folks whom I would not consider engineers but today they are - everybody and their mother is a 'software engineer'. There is nothing wrong with being the latter vs the former coz the latter is where the value addition is happening, khota sikkas are pouring into coffers ka-ching ka-ching and the e-khan-o-my is flourishing. But those who are graduating as engineers (CS or EE) especially grad programs from good schools, if they also only have the level of knowledge as the latter when it comes to the systems at hand, we gotta big problem. For these bunch of people there is no excuse not to know their fundamentals back and forth, with their eyes closed and mijjile on phyrr. Yet that is what I see in a lot of folks whether from my batch or later. That needs to be remedied and that cannot be done when we have VCs/media/talking heads touting FB or eHarmony (the website/concept, not their superb backends) as cutting edge innovation but not someone designing a more power efficient RFIC.

[Anujan]

Saik-O

I am not sure about the hardware/compiler part. Automatic parallelization is a very hard task and has been attempted since the 70s without much success. ccNUMA machines are over 20 years old now and these problems have been attempted and tackled unsuccessfully for years. it is clear that you will not derive possible benefits without programmer intervention atleast for the next decade*.

*The reason I confidently make this prediction is because doing good-enough parallelization is hard not because "nobody took the time to write proper software for it". It is hard because the theoretical problems underlying it have not been solved yet. Take alias analysis for example. The very basic question of "do two pointers point to the same location" has not been solved in a general way in computer science. That means that you cannot take a sequential program and parallelize it so that its functions run in parallel - If the program uses pointers. So if you think that a programmer will write a program and the compiler will figure out which function calls can be run in parallel and will automatically do it, that is not happening any time soon. Atleast not for C++.

On the other hand, there is a some room -- Parallelization can be done successfully at some granularities -- for example hardware and compilers do a decent job parallelizing at the instruction level. However, there too if programmers write stupid code, the slow down can be in the order of 100x (not 100%, 100 times). Typical F-ups being using structure of arrays rather than array of structures and vice versa. That F-up is because programmers dont understand caching (caches can be 1000x faster than memory access) and dont understand cache line prefetching and whether arrays are stored in row major or column major order. I have personally hand optimized programs making 2 line changes to make them run 10 times faster. Yes 10 times.

At the task level, Threading toolkits and things like Hadoop do a good job parallelizing at task level but there too if programmers do a stupid job of not understanding the use of singletons for expensive one-time operations, the slowdown can be 1000x.

Ofcourse there is a set of computing tasks that abduls need not worry about. For example if you are writing excel macros you dont have to worry about threads and mutexes and parallelism. There is a lot of computing tasks that need to be done at this level and ease of doing such tasks is rapidly improving. OTOH, if you need to be any reasonable system type of person, knowing these tasks is absolutely essential. It could mean the difference between being able to do something vs not. Take cloud services for example -- if I told you that a system could be built which can handle billions of emails per second, scan them for viruses and search through them, what kind of programming and system building skills are needed for that? Increasingly software is moving towards that direction. Instagram for example (well maybe the service itself is pointless, but still) -- you have to run billions of image filters on billions of images per day. Store them, retrieve them and manipulate them. Can you tolerate a 10x slowdown there due to stupid programmers who dont know how to use a Mutex?

[SaiK]

do we need a new thread for CS & SE that does not fit in gizmo and IT?

--

what I meant was, i can declare mutex tags.. (i am sure some language capability should be there to empower the cores here - c or some parallel computing language) the compilers and execution should automatically parallelize it. going by the cores and hyperthreading- i am sure, easily we would be hitting soon at steps of 8 cores per cpu, and 32 core could very well be the starting point to say 1024 hyper cores on a box.

[Anujan]

The reason I mentioned ccNUMA is because you might have heard the marketing term for it these days "Multi Core"

ccNUMAs were like multi cores except the whole thing was not on a chip but on many chips. Miniaturization made ccNUMA possible on a single chip. But if you sold ccNUMAs branded as "here take a ccNUMA" nobody will buy it, but then if you sell people an "octa core" abduls will gladly fork money for it.

[SaiK]

what constitutes the parallel block closure? would not that direct how much of parallelization can be achieved.. and the reason, we need to identify what function we are trying to solve.?

[Anujan]

Saik-O

In theory it sounds fine, but in practice it can be notoriously difficult. One of the problems with automatic parallelization is that inferring run time behavior by inspecting static code is notoriously hard. Here there is a tension -- dynamic behavior like dynamic linkage makes programmers productive, they can write complex programs with minimum effort -- but makes compiling very hard. Lets look at an example.

Imagine you had 3 classes.
Class Parent with virtual method Foo()
Class Child inherits from parent has Foo()
Class Grandchild inherits from Child has Foo()

Imagine you had a function
Function Baz(Parent* Object) {
Object->Foo()
}

in C++, the following calls are valid

Class Parent Object;
Baz(&Object)

Class Child Object;
Baz(static_cast<Parent *>(&Object));

Class Grandchild Object;
Baz(static_cast<Parent *>(&Object));

In case 1, Foo() of parent is run by Baz, in Case 2, Foo() of Child is run by Baz and in Case 3 Foo() of Grandchild is run by Baz. This makes programmers productive. They dont need to write different versions of Baz based on what type of object is passed to it. Also understanding the program is easy for others.

The important thing is to note that while compiling the function Baz, you dont know which Foo() to run and that is resolved only at run time, depending on what object is passed to it! Now imagine Foo() of Child cannot be run in parallel with Foo() of Grandchild, but all other pairs are safe to run in Parallel. The question is -- can two invocations of Baz be run in parallel? Well you dont know at compile time!

Now you might say -- well that is easy, at compile time look at what object is passed and make copies of Baz specialized to each type. There are two problems with it
1. If Baz could take n classes, each of which has k specializations, you would need k^n specializations of Baz (k power n -- for 2 and 10, it is a 1024). It only gets worse with more methods, more classes and more specializations.

2. In general for a language like C++, this kind of static type inference cannot be done. There are languages where static type inference can be done like Ada, C++ is not one of them.

There are many many many such cases where run time behavior is unpredictable. Now comes the question -- how do you automatically parallelize it? In my previous incarnation as a nanha muj, I worked on such problems. In particular trying to automatically parallelize BLAS3*. Even under restricted set of conditions it is very very hard. Big bucks by three letter agencies are paid for doing such things.

* Here is a description of BLAS http://en.wikipedia.org/wiki/Basic_Line ... ubprograms

[ArmenT]

A few months ago, I did a presentation for our local enthusiasts group about implementing a scheme-like interpreter in perl.

[SaiK]

I was listening the graphics processor guru, now a prof at cmu, earlier from stan madrassa - fatahalian - you may find his jootube videos. he was highly speaking of compilers for parallel computing possibility.

you just write a function and a block code, with a specific language for parallel blocks.. what if kind of research. so, it would not be day too far that something on those will see a bright future. i may be wrong, but i strongly feel quantum computing will take shape the future, and before that we will see high demand parallel computing architectures making hay before being taken over by quantums.

when processor pipelines can execute with large l1 cache, things we are talking are different. ultimately it all boils down to what is our function, how much we can parallelize.

so, i am expecting the race conditions, mutex, et all would be only for those high dependency aspects of interprocess communications and l2 / l3 cache based logic.
--
anujan, there are modeling tools and languages to work on runtime behavior of high synchronous multi-threaded applications. so, i don't see any issues with race conditions, mutex being a botheration for aam thinkers like me. perhaps matrimc, may have a combinatorial analysis if my thoughts would yield higher chance of an outcome.

[ArmenT]

Anujun saar, why static_cast when calling Baz? Should work just fine without the static_cast (since Child and Grandchild are subclasses of Parent (is-a relationship)) and it will still call the correct method because Foo() is declared as a virtual method.

The fun really starts if the method Foo() is NOT declared virtual in Parent. In this case, the code will still compile whether you do:

Code: Select all

Child object1;
Baz(&object1);

or

Code: Select all

Child object2;
Baz(static_cast<Parent *>(&object2));

However, in both cases, it will not call Child::Foo(), but rather Parent::Foo() because the Foo() method is not declared virtual and hence no vtable lookup happens to locate the method for the actual argument type.

In languages like java and python, all methods are virtual by default.

[SaiK]

now i have to learn c++ to understand all these die hard castings.

[Vayutuvan]

One of my 3 year senior batch peechhaddi was asked about aliase detection in his quals and he seemed to have reduced halting problem to aliase detection. He left just a cryptic message a la Fermat saying that he did the reduction on the spot on one of our "internal to grad students" usenet groups dedicated to quals. He did not post any approach for the proof. So some of us got together and tried to recreate but never spent enough time as we were preparing for quals. He passed the quals so apparently this the claim and his proof were correct. That still remains a mystery to me as to how he did it.

So essentially Alias Detection is undecidable for general cases. gcc accepts --std=c99 and detects restrict keyword. This seems to help in making c programs as fast as fortran programs. I have several benchmarks for large saxpy, inner product, dgemm like ops, but blocked, unrolled etc.

[Vayutuvan]

ps: matrimc, who would be the gizmo gal here?

Not discounting the possibility (at the risk of KJo slotting me into Femi-Nazi bucket ).

[SaiK]

wouldn't reference /pointer counts be enough for alias detection? of course within oo sphere, you would have to qualify that further with all subclass references. dunno how when pointer is allocated memory space, something like a pointer counter could help or list of pointers to space. but then what is the use of such taknique?

[vina]

What are you doing wasting your time generating hot air in ppt-giri land.....time to come back to the trenches and soil your hands doing some real stuff.

Yes. I am not in PPT giri land anymore (last 3 years), but actually doing "Signal Processing" using Peechaddi/Pacchidee level stochastic models (did all the shoveling and trench digging myself, no employees , now have a great Fuhrer bunker that has withstood the nuclear armegeddons and winters in the past 3 years very well , pretty kicked about it) . And the "Signals" I process happen to be not electrical signals over a wire, but market signals over a feed!

After all, if the idea is to do "signal processing" to make money, who not do signal processing and make money directly , rather than do it , sell something through a lot of people to 400000000 abduls and then make money. Simpler, faster and better onree no ?

yeah we want rifle troops not "distinguished engineers" who attend only IETF/ITU/Nanog meetings and interact with "top tier management" of clients

Now meeting "top tier management" is now only to kiss upto them to scale (additional capital and funding onree). Kind of very different and inverted for me ,when I used to be on the other side and look over the Pince- Nez and subject the Abdul on the other side to a hazing session. Now being the Abdul subject to hazing is a different and humbling experience!

with low level code running on the metal for the entire semester in an attempt to make the students really map the textbook fundamentals to the way it is realized and implemented on the ground

Pah! I always knew that you were a liar. You are closer to 35 to have even thought of these things , instead of going rah-rah over HTML and Javascript like the 20 something you claim to be.

OTOH my generation never set dip switches in extension cards. We had "plug and play". We never used punched cards and assembly programming was on the wane

Pah!! What a bunch of wusses!

what is the status of hierarchical mutexes? Recursive mutexes? Alternately critical sections?

Is this in Inglees ? Me YumBeeYea no understand anything here. Mutts, yes, Exes - yes, Hierarchy absolutely (YumbeeYea giri is a lot about that) , but Hierarchical Mutexes is too much for ppt giri!

[Raja Bose]

vina mullah, the most hardcore embedded guy I know is 22 years old, not 42 He is wasting his time in Hickory Pork right now just becoz he wants to schmooze this motorhama at his workplace.

Signal is signal, all come thru wire whether its a trace or market wire!

[negi]

Vina saar I thought you were a Mech engg. isn't it considered blasphemous in your circles to talk about 'Soft-ware' ?
Anyways this talk about older generation > new generation actually reminds me of the usual phenomenon where a hardware/system Engg would think Software jockeys are useless, similarly blokes in solid state physics would think that all these embedded/VLSI/CHIp design guys are a bunch of show offs and if that was not enough particle physicists would look down upon the entire bunch for latter don't think there is anything to be learnt in playing around with stuff where everything has been already discovered they instead get a high in chasing stuff like Higgs bosons or tachyons.

[suryag]

RB mullah "every generation thinks it is more intelligent than its predecessor and wiser that its successor" - something of that sort but i totally agree with you when you find kids who dont know how

x=a+b; is realised in a processor, they take a lot of things for granted

[vina]

Vina saar I thought you were a Mech engg. isn't it considered blasphemous in your circles to talk about 'Soft-ware' ?.

As a Generalist YumBeeYea ( a certified IBEE League, Easht Koasht one to boot), I can talk with authority on anything and everything, including things I have absolutely no clue of! ( I absolutely have the upper hand in Technology Strategy! )

We after all are the cousins of the other uber Generalists, the IAS, who one day can be in charge of taking care of susoo and potty facilities for Mantriji,be the district administrator the next, run large cities like Bangalore the third, develop roads, bridges and ports the next, set up manufacturing plants, run "policy" from Dilli after that and finally act holier than thou and also spout "subaltern studies" and "sustainable development" and "energy security" the day after.

YumBeeYeas, IAS Monkeys and Dilli Billis are all joined in the hip in that respect. We can shoot our mouths off about anything and everything. That does take some serious amount of Chutzpah, which we have in spades!

[vina]

More Egg Samples of YumBeeYea giri leading to "Soup-e-rear Upper Hand in Technology Strategy"

Take Mickey S*ts , price cut, and inventory write down on its Surface tablets. A kwik back of the envelope calculation based on the price cuts and quantum of write down leads to the following anal-y-sis

Mickey sold roughly 1million Surface Dabbas(per what RB and other say) and had roughly 6 million in inventory . That is the impairment value of that close to a 900 million. Some very smart YumBeeYeas at Mickey must have looked at Fruit's numbers and what Mahdi did with the iPad and decided to make and inventory that kind of volumes!

Like Doc Shiv Ji always maintained (there are more "parliamentary" versions) "If my aunt had a d*ck , she would be my uncle " , "If Ballmer was Mahdi, Surface would have been an iPad!"

[Vayutuvan]

what is the status of hierarchical mutexes? Recursive mutexes? Alternately critical sections?

Is this in Inglees ? Me YumBeeYea no understand anything here. Mutts, yes, Exes - yes, Hierarchy absolutely (YumbeeYea giri is a lot about that) , but Hierarchical Mutexes is too much for ppt giri!

Think of threads spawning threads - say resulting in a tree like thread hierarchy. Peer threads at the same level should be and are excluded from the CS at that level but child threads should not be excluded. pthreads does not allow for spawning a tree like thread structure. In other words all threads are at the same level and are excluded from the CS. Some patterns of computations fit a DAG. To construct a DAG I can have a bunch of threads already started and waiting for work - also called a thread pool - from which I can build a DAG if the scope rules are dynamic rather static.

For example, in the example given by Anujan what if the recursive call in fact was started in a thread? There would be a deadlock even if recursive mutexes were possible.

[SaiK]

when you say excluded from the CS, do you mean allowed to be executed or wait?

rbullah, and may we know which motraham are we talking?

[Raja Bose]

^^^motorham is not famous like Marissa bibi so her identity is of no consequence to anybody but that geek.

------

Mean-e-while,...

Samsung unlikely to pull away from Google at October developer event

I wonder if the author knows that Sammy already has its own parallel App Store. But then its Chris Ziegler....

[Singha]

imo the galaxy design style has been cloned across 20 models by Samsung and 2000 models by agile chipanda types. just looking at one makes me slightly ill these days.
android too seems to have a dated look trying to replicate a desktop icon system and doesnt really appeal to women's tastes in simplicity , smaller size, fashionable "look", not having to dig deep into options ....

the world awaits something new .. MSFT has a opportunity here if they can follow through on the Lumia range and not slip on a banana peel. having the same core HW from 520 for a long way up is advantage .... and its not the rabbit breeding cortex A7 for once...the infamous QUAD core munna whose performance is apparently well below "proper" dual cores from the likes of arm or qualcomm.

[SaiK]

There are these tier 2 players like HTC and LG. I am sure, they can produce something different for the competition.

[Vayutuvan]

when you say excluded from the CS, do you mean allowed to be executed or wait?

rbullah, and may we know which motraham are we talking?

SaiK corrected my post to make it a little clear. See if you can parse it and make sense. If not more later

[Mort Walker]

Samsung unlikely to pull away from Google at October developer event

I wonder if the author knows that Sammy already has its own parallel App Store. But then its Chris Ziegler....

Sammy is looking to monetize ASAP and will throw enough crap out there that some of it may stick. One day the universe will have many Galaxys integrated together such as the Galaxy phone, tablet, laptop/desktop, TV, fridge, microwave oven, oven, cook top, dishwasher, and washer+dryer. Perhaps Sammy will work on this. Right now, the top of the line Samsung 30+ cu.ft. fridge, which by the way is great fridge, has a touchscreen display that is running a stripped down Android OS and the only way to install apps is through the Samsung app store.

[SaiK]

okay, let me ask the question differently: when you say child thread should not be excluded, do you mean they are allowed to execute the CS?
If so, then what happens to the data integrity?

[nachiket]

imo the galaxy design style has been cloned across 20 models by Samsung and 2000 models by agile chipanda types. just looking at one makes me slightly ill these days.
android too seems to have a dated look trying to replicate a desktop icon system and doesnt really appeal to women's tastes in simplicity , smaller size, fashionable "look", not having to dig deep into options ....

the world awaits something new .. MSFT has a opportunity here if they can follow through on the Lumia range and not slip on a banana peel. having the same core HW from 520 for a long way up is advantage .... and its not the rabbit breeding cortex A7 for once...the infamous QUAD core munna whose performance is apparently well below "proper" dual cores from the likes of arm or qualcomm.

My mom finally bought the Lumia 720 which I recommended. She hasn't used any kind of smartphone before, so I am very curious to find out how long it takes for her to figure out the ins and outs of WP8 and get familiar with it. As per popular opinion, including that of our very own Boseullah, WP8 is more intuitive for non tech-savvy people than Android or iOS.

As for the processor performance, WP8 does not seem to need cutting edge hardware to work smoothly unlike Android. I would any day choose Android over WP if I was choosing between top of the line handsets like HTC One/GS4 and Lumia 920. But in the low and mid ranges the WP handsets score over Android in their smoothness and fluidity and general bug free experience. FruitCo of course has no offerings in this range. Even the obsolete iPhunwa 4 is probably costlier than 720 in India.

[BhairavP]

I got my mom a 520 - she's very happy Whatsapping away to glory with her younger siblings

[Singha]

520 has no flash. does the camera still work indoors? thats the only downside of 520 I could think of. 620 has a smaller screen.

[BhairavP]

^Not very well. But that's fine for mom, she doesn't take too many snaps anyway.
Nice phone otherwise, for 9k.

[Singha]

my spider feel says android is heading into trouble atleast for samsung cash cow pov. everyone and their brother are loading up on 10-15k range panda phones featuring poor cam & internal mem but similar hw and screen .
the plastic cases look a bit cheaper but on a 2 yr ownership cycle people dont really care.

@ 6.3 inch the galaxy mega has about reached max size. they need to provide something compelling in 4.5" real phone form factor @ 15K with superia Nexus4 type specs and HTC type build quality to demand a premium over the crowd of pandas camped there.

it must blow away the pandas in looks, speed, function to attract back the younger gen. capturing younger gen 10-15k first purchase is highly important to capture their next move into the 20-25k range.

[negi]

Speaking of Android how is the old, cranky and stubborn Sony doing with it's Experia series afaik they have an entire range of android devices in 8-35k range including dual sim ones.

[prahaar]

Android is emerging as a punar-janma of the Old-Symbian devices in terms of the demography and price points. Also the segmentation of different android versions reminds me of the Old-Symbian issues. Android has plateaued in terms of novelty, it is anybody's guess who will bring in the next non-IOS segment. The Life-Cycle seems strangle similar.

[Singha]

well finally ordered the lumia 820 for wife. its bigger screen size and more premium look was the clincher over 620 and its chunkier form factor and heavier weight over the 720 which is kind of a odd one (520,620,820 are kind of similar in body). the 720 felt too light and a bit flimsy in my hand and the side keys didnt feel so premium or durable. the camera is better than 720 in MP though not in widest aperture and hopefully the amoled will satisfy her photo and video interests vs the ips lcd on 720.

have to admit the tile UI is very pleasing even on a 620 and its lightning fast....I tried loading some xls and doc and it was quick..cant imagine my samsung galaxy duos being that fast at anything.

fiddled with some 5" panda phablets ... ugly, heavy, felt like a brick and not a phone. a crowd of munnas on lunch break were gathered there (in croma) store when I went to have a look before ordering online.

I paid on up agreement of NOT wrapping the gorgeous yellow body in a crappy plastic cover and other favours that cannot be talked of here

[prahaar]

well finally ordered the lumia 820 for wife. its bigger screen size and more premium look was the clincher over 620 and its chunkier form factor and heavier weight over the 720 which is kind of a odd one (520,620,820 are kind of similar in body). the 720 felt too light and a bit flimsy in my hand and the side keys didnt feel so premium or durable. the camera is better than 720 in MP though not in widest aperture and hopefully the amoled will satisfy her photo and video interests vs the ips lcd on 720.

have to admit the tile UI is very pleasing even on a 620 and its lightning fast....I tried loading some xls and doc and it was quick..cant imagine my samsung galaxy duos being that fast at anything.

fiddled with some 5" panda phablets ... ugly, heavy, felt like a brick and not a phone. a crowd of munnas on lunch break were gathered there (in croma) store when I went to have a look before ordering online.

I paid on up agreement of NOT wrapping the gorgeous yellow body in a crappy plastic cover and other favours that cannot be talked of here

Did you choose the wireless charging option? In my personal experience, it is quite addictive once you invest in it. It is quite bad that wireless charging cover is not part of the default package.

[Singha]

in india it doesnt come with wireless charger in the box. let my bandaged musharaff recover from this hit first.

[Raja Bose]

Android is emerging as a punar-janma of the Old-Symbian devices in terms of the demography and price points. Also the segmentation of different android versions reminds me of the Old-Symbian issues. Android has plateaued in terms of novelty, it is anybody's guess who will bring in the next non-IOS segment. The Life-Cycle seems strangle similar.

Arrey bhaiyya hum kya aise hi last do saal se bol rahein hain ki Android is the naya Symbian. It exactly mirrors Symbian's life cycle - the only difference is now its on touchscreen (vs non-touch) devices and the threat to Android is from a disruption in the services business model rather than purely HW shift. Regardless the underlying fundamentals are pretty much the same.

WP suffers from one major disadvantage right now which is the lack of a unified messaging center. Another feature that I am sure folks would appreciate is the ability to close an app (swipe to close). Other than that it has shaped up to be a pretty solid offering with UX which is miles ahead of Android and in some cases iOS too. These 2 changes should be coming in the next release version. Who could imagine, Mickey of laggy Windoze and IE fame could turn the foundations upside down to produce a buttery smooth OS.

I use wireless charging on my 928 at home and its perfect for a bed stand type setting - wired just feel clunky and encumbered after that. For travel I keep the stock wired charger in backpack permanently. Wireless charging is slower than wired but other than that is pretty seamless.

Sammy has big ambitions and their partnership with Chacha is just a stepping stone along the way. In some ways they are like the old Mickey in terms of their ability to leverage massively brute strength to make up for lack of creativity/innovation but then the marketplace is no longer what it used to be for old Mickey....in fact even Mickey had to change its approach to deal with this shift. So we will have to wait and see whether our whole world will be run by a whole universe of interconnected self-aware Sammy devices or not.

[Mort Walker]

^^^Yes, there are a lot of low end Android devices, but once (if) WP becomes widely accepted, then the same will happen with it when the chi-panda companies produce cheap/buggy/slow WPs similar to Android phones. Only iOS is exclusive to Apple h/w, so there are tradeoffs - no low cost devices, but a guarantee of a "superior" UX.