Ask Asher

[Asher]

Computer questions!
Life questions!
filosofy questions!
Business questions!

I will ignore questions I deem boring.

[Kuciwalker]

It seems like a huge problem in [concurrency] synchronization is that you have several very fast processors, and memory that is very far away. So why not take a tiny piece of that processor and move it all the way over to the memory (say, a piece of logic that will atomically increment a memory location)? Then you'd never have to lock the bus, or do anything else complex.

Hell, with atomic increment a LOT of cases where you need some sort of mutex primitive would go away.

[Asher]

I'll get us started with some answers I've provided internally at work!

Question: Is there a difference between shrimp, prawns and jumbo shrimp?

Answer: Form most perspectives, the answer is that there isn’t a difference between shrimp and prawns. Both terms apply to the same semi-creepy crawly water based critter. The terms are used interchangeably with different regions having different preferences. Jumbo shrimp, however, are usually restricted to shrimp that are 10-20 count (10 to 20 shrimp per pound), or large shrimp.

How Would I have Known: If you google/live search it, you’ll get distracted by numerous people who claim to have found some subtle difference on which to classify shrimps vs prawns. Sure enough, in biology books, there are many different variety of critters that fall into the shrimp/prawn category and they all have subtle differences and fascinating Latin names. However, from a cooking or taste or practical viewpoint, use shrimp and prawn interchangeably, or when it makes for good humor (e.g. the scene in Muppets From Space where Pepe says "I’m not a shrimp, I’m a King Prawn!")

-------------------------

Question: Why did the chardonnay at dinner tonight have sugar at the bottom?

Answer: It probably wasn’t sugar. During several stages of wine making, you add yeast and sugar. The residue (or lees if you are technical) of dead yeast, grape skin and other stuff falls to the bottom. Most vineyards will filter out the lees before the final bottling. You can also get some crystallization of tartrates which is basically a sugar like crystal formed because the wine was supersaturated with potassium bitartrate when it was made. Again, these sediments are usually removed before the final bottling, but they can also be created when a bottle of wine is suddenly chilled. You can also find sediment from bad corks, or from really poor quality wines that had a lot of sugar syrup added to mask the bitter taste of poor quality grapes.

How Would I have Known: Short of being a connoisseur or wine snob, you can always call up your local vineyard for a tour or talk to someone from wine country. Personally, I hate wine. I looked at http://www.wines.com/magical/tartrates.html and a few other sites found on Google

-------------------------

Question: How do you tie a Windsor knot?

Answer: The Windsor not is the most complicated and bulky of the common “tie knots”. If you want to learn how to tie one, go here: http://www.animatedknots.com/. The site has some nice animations and pretty clear instructions. Most people stick with the four-in-hand or half Windsor though as it’s easier to tie, uses less material and makes a smaller knot.

How Would I have Known: You can take the Julia Roberts approach from Pretty Woman, but I wouldn’t recommend it.

-------------------------

Question: Why does London have so few sky scrapers compared to New York or Toronto?

Answer: London is a very conservative city. Until the 1960’s, the city had laws that all buildings had to be less than 100 feet tall for the reasons of “protecting the views” and “allowing firefighting equipment to reach all floors”. You can find a nice graphical representation of the recent skyscraper boom here: http://www.skyscrapernews.com/britains1.htm. However, if you are really after sky scrapers, ignore London and join the Dubai office.

How Would I have Known: Wikipedia has a good summary (http://en.wikipedia.org/wiki/Tall_buildings_in_London). For Dubai, however, you want to look at one of the many sites including http://further-left-forum.blogspot.c.../01/dubai.html that has pictures of the 15-25% of all construction cranes in the world in action.

[Wiglaf]

Not really a question, but get the penis squirrel with the turban and the windmill out of the avatar.

[Asher]

Originally posted by Kuciwalker
It seems like a huge problem in [concurrency] synchronization is that you have several very fast processors, and memory that is very far away. So why not take a tiny piece of that processor and move it all the way over to the memory (say, a piece of logic that will atomically increment a memory location)? Then you'd never have to lock the bus, or do anything else complex.

Hell, with atomic increment a LOT of cases where you need some sort of mutex primitive would go away.

That's a different way to look at it. Most people think "hey, memory is so far and so slow from the CPU" and conclude that putting memory (cache) on the CPU is the best solution. AMD decided putting the memory controller on the CPU is the best solution.

Ultimately, I can't see this solving the issue. The performance bottleneck is very seldom related to memory bus locking or anything related, it's either latency or bandwidth. Ultimately, I can see the memory controller being placed on the CPU as a solution to latency as well as bandwidth (in multi-core CPUs especially). I do not see any benefit from placing some memory controller functionalities onto the RAM from the memory controller or CPU.

[Asher]

Originally posted by Wiglaf
Not really a question, but get the penis squirrel with the turban and the windmill out of the avatar.

My question is: Why do you feel threatened by my black penis?

[Krill]

Will the 9xxx GeForce series be any good?

[Kuciwalker]

Originally posted by Asher
That's a different way to look at it. Most people think "hey, memory is so far and so slow from the CPU" and conclude that putting memory (cache) on the CPU is the best solution. AMD decided putting the memory controller on the CPU is the best solution.

Cache is flaky as far as concurrency is concerned. Esp. when you get reordered reads and writes and other crazy stuff that I don't have to deal with yet (thank god).

Ultimately, I can't see this solving the issue. The performance bottleneck is very seldom related to memory bus locking or anything related, it's either latency or bandwidth. Ultimately, I can see the memory controller being placed on the CPU as a solution to latency as well as bandwidth (in multi-core CPUs especially). I do not see any benefit from placing some memory controller functionalities onto the RAM from the memory controller or CPU.

I'm talking about dealing with the memory bottleneck solely as it related to multiprocessor thread synchronization, not in the general case. (You would probably add this logic to the cache, too. Something like increment-by-one should be fairly cheap.)

[Asher]

Originally posted by Krill
Will the 9xxx GeForce series be any good?

Yes, it's a minor update to the G80/8000 series. It has some tweaks for better HD video, but is mostly a cooler and smaller 8000-series. Soon there will be high-end versions.

[DirtyMartini]

Originally posted by Asher

My question is: Why do you feel threatened by my black penis?

Personally, I find it distrubing that it's dead and necrotic.

[Lancer]

What will happen to the US economy in the next 5 years and will the dollar be stable now that its dropped a bit vs most currencies?

[Asher]

Originally posted by Kuciwalker
Cache is flaky as far as concurrency is concerned. Esp. when you get reordered reads and writes and other crazy stuff that I don't have to deal with yet (thank god).

I'm talking about dealing with the memory bottleneck solely as it related to multiprocessor thread synchronization, not in the general case. (You would probably add this logic to the cache, too. Something like increment-by-one should be fairly cheap.)

I'm afraid I don't quite understand what you are getting at.

You are referring to systems like, say, the Xbox 360. It has 3 high-speed CPUs and one memory system and one set of cache to share. When you say cache is flakey, I'm assuming you mean cache in the traditional desktop CPU sense, where it is managed automagically.

To get around this, some systems (like the Xbox 360 CPU) have cache locking. CPUs can arbitrarily "own" segments of the cache and treat it like ultra-high speed memory rather than traditional cache. This eliminates some of the headaches of multiple cores sharing the same memory space.

I've never encountered a situation where I'd need to lock the bus or would in any way benefit from putting some of the memory logic onto the memory itself. If you do the logic right away or you need to dispatch it to someone else to compute, it's still getting done. I also would be wary of de-centralizing memory address calculation, because the CPU needs to be aware of that at all times. If the memory chip is updating the address, the CPU needs to eventually update as well?

Perhaps you could provide an example usecase where this would be beneficial.

[Asher]

Originally posted by Lancer
What will happen to the US economy in the next 5 years and will the dollar be stable now that its dropped a bit vs most currencies?

The US economy will continue its borderline-recession until January of next year, and then everything will be fine.

The dollar will be stable in January of next year as well.

[Asher]

Originally posted by DirtyMartini


Personally, I find it distrubing that it's dead and necrotic.

It is very much alive. And erect. This is not rigor mortis, this is the real thing, baby.

[Jaguar]

My intro programming course sucks. What should I do about this?