Vertex available at the end of 02/09

[taltamir]

That is what a buffer cache does for you. The bench writes to the cache and presumes it has been written (flushed) to the media despite the fact it isn't (yet, flush in progress albeit slower than the cache was written to).

Reading from the same cache is just as fast as writing to it if you are reading the file you just wrote to the cache or if the file you want to read was preloaded to the cache (prefetching).

mmm, which means as soon as the cache is full you start stuttering.

 

[coolVariable]

Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

 

[boomhower]

Originally posted by: Idontcare

Originally posted by: VaultDweller

Originally posted by: Denithor

That's my hope - $200 X25-M anyone?

I mean, it's still a lot of cash but 80GB is certainly large enough for OS + games and I can use my WD6400AAKS for storage.

It's large enough for OS, yeah, but 80 GB won't even cover my Steam games, let alone all the other games I'd like to have installed.

Not sure about what most people think when they see these small drive capacities but I just think "yummy, I'll take four in a raid-0 config please!".

I don't care for the large capacity races they are getting into. Let them make 256GB drives, good for them, but for me I'd rather buy four 64GB drives for maybe 20% more total price than that single 256GB monolithic drive but my four drives will give me the same net capacity and 3-4x the bandwidth when put in raid-0.

And some people have notebooks that only accept one drive. Different drives for different applications.

 

[Idontcare]

Originally posted by: taltamir

That is what a buffer cache does for you. The bench writes to the cache and presumes it has been written (flushed) to the media despite the fact it isn't (yet, flush in progress albeit slower than the cache was written to).

Reading from the same cache is just as fast as writing to it if you are reading the file you just wrote to the cache or if the file you want to read was preloaded to the cache (prefetching).

mmm, which means as soon as the cache is full you start stuttering.

Precisely, but if the cache is large enough then it isn't likely to be overwhelmed by typical desktop application usage. I'm sure a benchmark could be orchestrated in such a manner to overwhelm the cache though.

Alternatively it could be coupled with a smarter more intelligent write algorithm so that the filled cache doesn't cause stuttering. Vertex drives are supposed to have the new controllers for this.

Poor jmicron though, they really got the shit-end of the credibility deal in all this. Will they ever recover market share even if they produce a controller without the stutter problems? Unlikely.

 

[Idontcare]

Originally posted by: boomhower
And some people have notebooks that only accept one drive. Different drives for different applications.

Oh yeah, hadn't thought about that. Great point.

 

[boomhower]

Originally posted by: coolVariable
Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

Look nice but I bet they will be very pricey, certainly more than these are proposed to be.

 

[tokie]

Originally posted by: Idontcare
Poor jmicron though, they really got the shit-end of the credibility deal in all this. Will they ever recover market share even if they produce a controller without the stutter problems? Unlikely.

is it not possible to add cache to the existing jmicron controller? i am guessing not, since otherwise somebody would likely have done it by now..

 

[taltamir]

Originally posted by: boomhower

Originally posted by: coolVariable
Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

Look nice but I bet they will be very pricey, certainly more than these are proposed to be.

it is slightly slower reads and slightly faster writes than the intel X25-M according to them, and the 60GB will cost 200$ and the 120GB cost 400$... how is that PRICEY?

 

[taltamir]

Originally posted by: tokie

Originally posted by: Idontcare
Poor jmicron though, they really got the shit-end of the credibility deal in all this. Will they ever recover market share even if they produce a controller without the stutter problems? Unlikely.

is it not possible to add cache to the existing jmicron controller? i am guessing not, since otherwise somebody would likely have done it by now..

only if it has been specifically designed to accept cache, and if that was the case it would have had it already...

But the apex takes and puts two jmicrons along with a RAID0 (sorta) controller and gets decent performance, and the raid0 controller they use DOES have cache (err, it does right?).

 

[Idontcare]

Originally posted by: taltamir

Originally posted by: tokie

Originally posted by: Idontcare
Poor jmicron though, they really got the shit-end of the credibility deal in all this. Will they ever recover market share even if they produce a controller without the stutter problems? Unlikely.

is it not possible to add cache to the existing jmicron controller? i am guessing not, since otherwise somebody would likely have done it by now..

only if it has been specifically designed to accept cache, and if that was the case it would have had it already...

But the apex takes and puts two jmicrons along with a RAID0 (sorta) controller and gets decent performance, and the raid0 controller they use DOES have cache (err, it does right?).

Not sure if the raid controller on apex has its own cache, but using two jmicron controllers means the drive overall has 2x the cache to buffer write requests.

As to tokie's question, jmicron controller already has cache...but then the next question would be "why not add more cache then?".

By the time it was understood that adding more cache would be necessary, the effort and time to do so would have been comparable to the time/effort already underway in developing the next gen controller boards anyways. (adding cache to these fixed hardware setups is not as simple as adding ram to your mobo, they weren't designed to be flexible like that)

So even if the answer is "yes they could have added more cache" that's not to say that it would have resulted in a solution being brought to market any faster than they are already scrambling to deliver.

 

[Denithor]

I don't think those dual-JMicron controller drives have cache on the controller. The Titan series from G.Skill still suffer from the same stutter, albeit perhaps not as pronounced as the single-controller SSD drives (see the other thread on this topic).

 

[boomhower]

Originally posted by: taltamir

Originally posted by: boomhower

Originally posted by: coolVariable
Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

Look nice but I bet they will be very pricey, certainly more than these are proposed to be.

it is slightly slower reads and slightly faster writes than the intel X25-M according to them, and the 60GB will cost 200$ and the 120GB cost 400$... how is that PRICEY?

I must have missed the price estimate in the article. If they are priced in that range then great, I am all for it. Seems to be a bargain against the current competition. Granted with OCZ's release schedule we may have holographic storage by the time they come out.

 

[Idontcare]

Originally posted by: Denithor
I don't think those dual-JMicron controller drives have cache on the controller. The Titan series from G.Skill still suffer from the same stutter, albeit perhaps not as pronounced as the single-controller SSD drives (see the other thread on this topic).

You talking about cache on the jmicron controller or cache on the raid controller?

For sure the jmicron controllers themselves have built-in cache. Just not very much it would seem.

 

[Denithor]

Not enough, anyway.

:laugh:

 

[Idontcare]

Yeah not that spindle-drives need cache for anti-stuttering purposes but when ram was getting cheaper and cheaper and cheaper I never quite understood why the onboard cache on spindle-drives never really took off. We had a few jumps from 2->8->16->32MB but I really thought we'd have seen a jump into the 128MB and higher territory when 1GB got to be less than $10.

At any rate, what a crappy thing to cut the cost-corner on when producing multi-hundred dollar products. Kinda like those ford pinto's needing a $2 gas-tank fix that went unfixed.

 

[boomhower]

Originally posted by: boomhower

Originally posted by: taltamir

Originally posted by: boomhower

Originally posted by: coolVariable
Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

Look nice but I bet they will be very pricey, certainly more than these are proposed to be.

it is slightly slower reads and slightly faster writes than the intel X25-M according to them, and the 60GB will cost 200$ and the 120GB cost 400$... how is that PRICEY?

I must have missed the price estimate in the article. If they are priced in that range then great, I am all for it. Seems to be a bargain against the current competition. Granted with OCZ's release schedule we may have holographic storage by the time they come out.

I hate to quote myself but the article was updated:

UPDATE: I just got some information from OCZ on pricing and availability this morning. It looks like the Summit series of drives will be available near the end of March with capacities of 250GB, 120GB and 60GB. Prices on those drives will be $999, $499 and $249 based on our estimates as well.

Still not bad for the performance if it does in deed come through but $200 sounds a lot better for the 60GB. What confuses me is the prices are right in line with the Vertex series. Something has to give there as well. That just may end up being the ticket for the money.

 

[taltamir]

Originally posted by: boomhower

Originally posted by: taltamir

Originally posted by: boomhower

Originally posted by: coolVariable
Hey just saw this article about an OCZ Summit series SSD.
Based on a Samsung controller supposedly ...


http://www.pcper.com/article.p...=666&type=expert&pid=1

Look nice but I bet they will be very pricey, certainly more than these are proposed to be.

it is slightly slower reads and slightly faster writes than the intel X25-M according to them, and the 60GB will cost 200$ and the 120GB cost 400$... how is that PRICEY?

I must have missed the price estimate in the article. If they are priced in that range then great, I am all for it. Seems to be a bargain against the current competition. Granted with OCZ's release schedule we may have holographic storage by the time they come out.

wow, I didn't realize they are just smoke and mirrors... inPhase has "products" listed and prices but those are EXPECTED products, they aren't actually shipping them...
Turns out they fired a bunch of people when they missed the may 08 deadline (which was pushed from 07, which was pushed from 06)... some of which came forward and said it wouldn't be done in a while, but the CEO refuses to listen to realistic time tables...

 

[taltamir]

Originally posted by: Denithor
I don't think those dual-JMicron controller drives have cache on the controller. The Titan series from G.Skill still suffer from the same stutter, albeit perhaps not as pronounced as the single-controller SSD drives (see the other thread on this topic).

then its not the SAME stutter, its reduced stutter. and that is the whole point.

 

[taltamir]

Originally posted by: Idontcare
Yeah not that spindle-drives need cache for anti-stuttering purposes but when ram was getting cheaper and cheaper and cheaper I never quite understood why the onboard cache on spindle-drives never really took off. We had a few jumps from 2->8->16->32MB but I really thought we'd have seen a jump into the 128MB and higher territory when 1GB got to be less than $10.

At any rate, what a crappy thing to cut the cost-corner on when producing multi-hundred dollar products. Kinda like those ford pinto's needing a $2 gas-tank fix that went unfixed.

that wasn't unfixed, that was broken on purpose, that is why it was the first case of punitive damages. They found internal memos where the passed along from department to department, calculating how many rear ends they will have a year, how much they will save per car, and how much they normally pay per dead person, and figured paying for the dead people would make them more money if they moved the gas tank to the location where it explodes. AFAIK it was 12$ not 2... still not much.

 

[Idontcare]

Originally posted by: taltamir

Originally posted by: Idontcare
Yeah not that spindle-drives need cache for anti-stuttering purposes but when ram was getting cheaper and cheaper and cheaper I never quite understood why the onboard cache on spindle-drives never really took off. We had a few jumps from 2->8->16->32MB but I really thought we'd have seen a jump into the 128MB and higher territory when 1GB got to be less than $10.

At any rate, what a crappy thing to cut the cost-corner on when producing multi-hundred dollar products. Kinda like those ford pinto's needing a $2 gas-tank fix that went unfixed.

that wasn't unfixed, that was broken on purpose, that is why it was the first case of punitive damages. They found internal memos where the passed along from department to department, calculating how many rear ends they will have a year, how much they will save per car, and how much they normally pay per dead person, and figured paying for the dead people would make them more money if they moved the gas tank to the location where it explodes. AFAIK it was 12$ not 2... still not much.

By unfixed I mean after the design was created and implemented (surely not knowingly/intentionally engineered to be broken as you are implying) they discovered the design had a weakness, which to fix would cost them $2 (or $12). They ran the numbers, as you say, and opted not to fix it. It went unfixed.

I don't think they set about creating the problem, it was not broken on purpose. That means something entirely different. Sure it was on-purpose that they allowed it to remain broken once they discovered it was broken (i.e. it was broken by accident), but that is a long phrase that is the same as saying it went unfixed.

I have a lawnmower in my garage that has a broken starter cord. I did not break the starter cord on purpose, it broke quite by accident, but I have intentionally chosen not to fix it for cost/effort reasons. It remains intentionally unfixed (and broken, albeit I did not break it on purpose).

 

[taltamir]

mmm...

Through early production of the model, it became a focus of a major scandal when it was alleged that the car's design allowed its fuel tank to be easily damaged in the event of a rear-end collision which sometimes resulted in deadly fires and explosions. Critics argued that the vehicle's lack of a true rear bumper as well as any reinforcing structure between the rear panel and the tank, meant that in certain collisions, the tank would be thrust forward into the differential, which had a number of protruding bolts that could puncture the tank. This, and the fact that the doors could potentially jam during an accident (due to poor reinforcing) made the car a potential deathtrap.

Ford was aware of this design flaw but allegedly refused to pay what was characterized as the minimal expense of a redesign. Instead, it was argued, Ford decided it would be cheaper to pay off possible lawsuits for resulting deaths. Mother Jones magazine obtained the cost-benefit analysis that it said Ford had used to compare the cost of an $11 ($57 today, allowing for inflation) repair against the cost of paying off potential law suits, in what became known as the Ford Pinto memo.[4][5] The characterization of Ford's design decision as gross disregard for human lives in favor of profits led to significant lawsuits. While Ford was acquitted of criminal charges, it lost several million dollars and gained a reputation for manufacturing "the barbecue that seats four."[6] Nevertheless, as a result of this identified problem, Ford initiated a callback which provided a dealer installable "safety kit" that installed some plastic protective material over the offending sharp objects, negating the risk of tank puncture."[7]

The most famous Ford Pinto product liability case resulted in a judicial opinion that is a staple of remedies courses in American law schools. In 1981 in Grimshaw v. Ford Motor Co.,[8] the California Court of Appeal for the Fourth Appellate District reviewed Ford's conduct, and upheld compensatory damages of $2.5 million ($5.92 million today) and punitive damages of $3.5 million ($8.29 million today) against Ford. It also upheld the judge's reduction of the punitive damages from the jury's original verdict of $125 million ($296 million today). Of the two plaintiffs, one was killed in the collision that caused her Pinto to explode, and her passenger, 13-year old Richard Grimshaw, was badly burned and scarred for life.

However, a 1991 law review paper by Gary Schwartz[9] argued that the case against the Pinto was less clear-cut than commonly supposed. Twenty-seven people died in Pinto fires. Given the Pinto's production figures (over 2 million built), this was no worse than typical for the time. Schwartz argued that the car was no more fire-prone than other cars of the time, that its fatality rates were lower than comparably sized imported automobiles, and that the supposed "smoking gun" document that plaintiffs claimed showed Ford's callousness in designing the Pinto was actually a document based on National Highway Traffic Safety Administration regulations about the value of a human life rather than a document containing an assessment of Ford's potential tort liability.

Due to the alleged engineering, safety, and reliability problems, Time magazine included the Pinto on its list of the fifty worst cars of all time.[5]

Interesting stuff... I am gonna have to go with a maybe for that one.

 

[Idontcare]

gained a reputation for manufacturing "the barbecue that seats four."

Ewww...had not heard that one before.

 

[BlueAcolyte]

No, reviews state that each JMicron controller has 16kb of cache (which is not a lot, the Intel drives have 256k, I think.)

 

[coolVariable]

cache on these controllers, e.g. the intel, is not used for actually caching reads/writes but rather information for the controller to translate physical location into logical location.
Of course intel uses a more advanced algorithm, thus keeping the write amplification low and getting better random writes, but needing more "cache" to store this information.

The vertex is the first SSD, AFAIK, that actually uses cache memory to cache reads and writes.

 

[taltamir]

exactly coolvariable... there is more then one kind of "cache".