Slashdot is powered by your submissions, so send in your scoop


Forgot your password?
Compare cell phone plans using Wirefly's innovative plan comparison tool ×

Comment Re:massive parallel processing=limited application (Score 1) 79

On a 8-core machine, a processor will be placed into a wait queue roughly 7 out of 8 times that it needs access. Further, The expected length of time in the queue is (1-(1/8)). This is of course, for an 8-core system. Adding more cores results in the waiting time increasing asymptotically towards infinity.

Sorry, that doesn't sound right. The expected length of time in the queue should be on the order of nt, where n is the number of cores and t is the average time required to process a memory-request. (A better formula would use the average length of the queue instead of n but to first order it still would be roughly linear with n.) So, the time required would increase linearly with the number of cores.

You're right, I worded it incorrectly (it's late, and I've been working 80hrs/week for the last year due to a private project. Forgive me). What I meant to say was "The expected delay when accessing memory is (1-(1/n))", but even that is off by an entire exponent.

The expected delay is (probability of queueing) X ( probable length queue). The probability of queuing is (1-(1/n)):

With 2 processors, you have a 1/2 chance of getting exclusive access, (1-(1/2)) of queuing.

With 3 processors, you have a 1/3 chance of getting exclusive access, (1-(1/3)) of queuing.

With 4 processors, you have a 1/4 chance of getting exclusive access, (1-(1/4)) of queuing.

With n processors, you have a 1/n chance of getting exclusive access, (1-(1/n)) of queuing.

The probable length of the queue is linearly proportional to n, so the expected delay is (1-(1/n) * n). In terms of performance this is O(n^2) - IOW it's piss-poor performance.

Or maybe I'm still doing the numbers wrong - feel free to derive a better statistic for predicting time-in-queue when processors are all using a single address bus. This is the one I got, and some trivial simulation does actually fit this profile.

Comment Re:massive parallel processing=limited application (Score 4, Interesting) 79

With a multiuser, multitasking OS you can have 25 different unrelated processes running on something with 25 cores. Or you could have 25 threads in a dataflow arrangement where each is a consumer of what the last just produced. Or you could go over the members of an array or matrix 25 members at a time with the same transformation. Some things are serial, but there are plenty of ways more cores can actually be used.

Nope. You'll generally hit the wall with around 16-20 cores using shared memory. You need distinct processors with dedicated memory to make multi-processing scale beyond 20 or so processors. Those huge servers with 32-cores apiece have their point of dminishing returns/processor after around 20 cores.

First, the reason you aren't going to be doing multithreading/shared-memory on any known computer architectures, read this.

Secondly, let's say you aren't multithreading so you don't run into the problems in the link I posted above. Let's assume you run 25 separate tasks. You still run into the same problem, but at a lower level. The shared-memory is the throttle, because the memory only has a single bus. So you have 1000 cores. Each time an instruction has to be fetched[1] for one of those processors it needs exclusive access to those address lines that go to the memory. The odds of a core getting access to memory is roughly 1/n (n=number of cores/processors).

On a 8-core machine, a processor will be placed into a wait queue roughly 7 out of 8 times that it needs access. Further, The expected length of time in the queue is (1-(1/8)). This is of course, for an 8-core system. Adding more cores results in the waiting time increasing asymptotically towards infinity.

So, no. More cores sharing the same memory is not the answer. More cores with private memory is the answer but we don't have any operating system that can actually take advantage of that.

A project that I am eyeing for next year is putting together a system that can effectively spread out the operating system over multiple physical memorys. While I do not think that this is feasible, it's actually not a bad hobby to tinker with :-)

[1] Even though they'd be fetched in blocks, they still need to be fetched; a single incorrect speculative path will invalidate the entire cache.

Comment Re:Just what America needs (Score 1) 225

"But at least this new automated-lawsuit system will keep a lot of lawyers employed."

Me, I would rathe see lawyers unemployed. We could use them as compost or for ginning cotton ("...Premium organic hand-ginned cotton...".

My ex-wife is a lawyer. You are significantly overestimating her utility as an unemployed lawyer.

Comment Re:Maps will never be good enough (Score 1) 144

"Not now" may mean hundreds of years in the future too. The current state of AI is very far off to do what an average drive has to do. It's more likely that we change roads than the AI catching up (ie, sends out electronic signals that the auto can read).

Certainly - and we can even include a mechanical interlock device so that the car can cannot physically deviate from a lane even if it wanted to. Intersections could be implemented via signalling so that cars will not t-bone each other.

In fact, to make thing efficient, the cars can even narrow the following distance to something lower than the reaction time of humans - say... a metre or so? Actually, just do away with the following distance and have the cars attached to each other. If the linked-cars leave at pre-determined times and arrive at pre-determined times at central locations, that will enable more people to use these self-driving cars.

Gotta go now - late for my train.


Comment Re:Nah (Score 1) 172

You keep saying that as if you know better, but they keep calling it a *limited* production car.

A limited production car is a subset of all production cars. As far as classification goes, the term "production car" comes from the rules for various group autoracing - see this for overview - and each group has different rules for allowing cars. Some need production cars with at least 2500 produced per year while others allowed a mere 200/year production run[1].

The classification "production car" for all groups means "a car produced for sale to the public". No more, no less. Even a handcrafted Bentley is considered a production car.

[1] Thus resulting in such rare beauties like the Ferrari 288GTO.

Comment Re:User friendly (Score 1) 306

That can't be true - Win95 had a timer bug that locked up the machine if you ran it more than X days

No you remember wrong (or this was fixed by the Release B version in Europe). Win98 however could not survive many days without rebooting.

My memory isn't *all* bad :-) I was referring to this bug, which was only fixed for both Win95 and Win98 in 1999: Microsoft patch from 1999, thus Win95, in 1999, could not run for more than 49.7 days.

Comment Re:User friendly (Score 1) 306

Reminds me off a guy that our support staff had to help over the phone on my first workplace. It took them a while to realise that his definition of stopping and starting the computer was powering the screen on and off :)

Or that Windows 95 user who had never turned his computer off since the day he got it, he was utterly impressed that all kinds of stuff began to work after our support staff rebooted it for him...

That can't be true - Win95 had a timer bug that locked up the machine if you ran it more than X days (where X is anything from 16-30 - I don't remember the exact number). Your Win95 user used the computer for less than X days? How new was he? :-)

Comment Re:Patriarchy! Patriarchy! Patriarchy! (Score 5, Insightful) 144

Even marijuana is sexist!!!

Don't be silly! Everyone knows that men and women are exactly the same, it's only the environment and society that thinks there's a difference. This study is obviously flawed and should be repeated until the correct result is achieved.


Comment Re:Nope (Score 1) 124

No one is late to the IoT party yet. It's not clear there's even going to be one.

By the time these kinds of trends have a dopey name, the party has already started. If you didn't already have an IoT product well under development by now, its already too late because all the players that will be successful, in what people are calling the IoT, already have a product at or near market release.

IoT is almost a decade old at this point. If there was a party nobody noticed.

Comment Re:Nope (Score 1) 124

Depends on what you want to do. If you have an IoT product to develop, this could be exactly what you want at its heart.

If you have an IoT product in mind, this is the last thing you want. I do this for a living. The problem with this thing is that it takes too much development effort to shave a few bucks off the unit cost. That extra effort translates into longer time to market. While you are loosing precious months trying to get your OS running with everything you need in a tiny memory footprint, another up-and-coming startup built a product around one of the other SBCs out there that isn't vastly under-powered. They beat you to market by 6 months with a product that only costs $5 more than yours, and you suddenly find your brilliant IoT product just got flushed down the loo because you couldn't execute a viable business plan.

Time to market is huge in this day and age. Just ask Intel and Microsoft how their IoT plans are rolling out, and you'll find out real quick that even a superior product at a lower price point will have a hard time competing when you're late to the party.

This is exactly backwards (which probably explains why IoT is almost a decade old and has yet to find any legs).

In the embedded space a saving of a few cents per unit can mean the difference between failure and success. If your goal is to chase venture capital/funding, then certainly you want quick time to market, but if your goal is to sell units and make a profit on each one then an extra two or three months getting your product cost-competitive is better.

For things like IoT you may *think* that adding $5 on top of the price of a unit is fine as long as you are first to market, but in reality when the consumer wants to buy 20 of them to outfit a building they're going to go with your competitor who spent the extra two-three months.

Comment Re:Goto (Score 1) 671

I'm currently dealing with about 2M lines of crappy code written in an ide that did/does soft wrapping of lines. Looks really pretty in that particular Windows ONLY IDE, but in anything else it looks like shit when you 1k, 2k and some times even 3k characters on a single line.

Just run it through indent and be done with it.

Unless it's Python, in which case you're out of luck.

Slashdot Top Deals

% "Every morning, I get up and look through the 'Forbes' list of the richest people in America. If I'm not there, I go to work" -- Robert Orben