I would watch out when you send stuff in for RMA. It is often typical that they send you back something that someone else returned. Be careful. I have heard of this happening before. Maybe switch retailers.
Scott, I realize you are a NAND guy, but can you pass along something to a colleague who might work with CPUs?
I (and many others, according to a Google search) have been experiencing random short beeps with our i5 (and i3?) CPUs. I found (on another forum) someone who said, in part:
"Intel processor has internal bug with the digital thermal sensor when C3/6/7 enabled, it occasionally generates false value that triggered off the CPU temp warning in ITE chip used on the board."
And, sure enough disabling the CPU temp warning in BIOS stopped the beeps.
If this *is* an Intel CPU bug -- and it may well be a mobo bug (but there are complaints from users of non-Gigabyte boards) -- it sure is annoying a bunch of folks, and I wondered if a workaround (other than disabling mobo alarms) exists or is being worked on.
Thanks!
(FYI, http://forums.anandtech.com/showthread.php?p=30288342 is my thread on anandtech.com about the problem.)
You guys are over-analyzing this. Scott was just trying to say that the missing RAID support in Intel firmware isn't due to negligence i.e. Intel guys weren't idiots that overlooked it. Minimal returns on such an investment might have been the crucial factor.
Even worse than that, he was a Lit major!
I think its more like there are technical challenges passing the required TRIM commands in RAID mode, like with SMART packets.
maybe my english is bad, still i don't get you,
first you got help with your question and then asked something like "there seems to be so many experts let me ask, do you even have an ssd?"
so sorry, this might sound to you like a simple question, yet it doesn't seem to sound so innocent, instead, of saying "thanks", you start this carp,
then we say, o.k, let's leave it behind, and now this?
what's your problem..?
you got your answer, i was working few hours in order to figure out what happens there, and it even isn't written within anand's review, so instead of appreciating it, you just keep on with your line of childish behavior, and you DID ask others before this, and you DID got your question answered, so why asking it all over again?
do you like people wasting your time?
why would you waste others?
that kind of sucks.
Last edited:
I appreciate your help and I know why SSDs slow. Your response was a rehash of fundamental operation that I hoped had been improved with the G2 series.
My statement about read times slowing was part of the background for the question about increasing the drive's spare area......not the question you answered.
No I don't but to me the question of "when are we getting TRIM in RAID" asked by 3 different people is a time waster. I'm sure Intel is aware and it's been asked a zillion times and asking a zillion more isn't going to make it any faster.
Just because you've given your definate answer or you think it's a time waster doesn't make it so.
I'm going to suggest you continue taking your meds and punish me by not helping me.
your read performance has decreased from 499MBps to 493MBps sequential, and another 3MBps per 4KB @ 64 thread..
not in 50 percent psychiatrist guy..
which, by all means, is nothing.
i never have asked this, that was sub.mesa, and your question as well was actually referring TRIM with RAID, that would be your ultimate solution.
my answer to u was legitimate and seems like the only solution to your problem (instead of RAID) currently, and you never even have answered whether or not you'r going to try that sequential writing,
i specified few time, you can create a sequential pattern on the drive using IOMetere and you just skipped it, so either, you underestimate the answer or rather, you didn't understand the solution which made you ask the actual SAME question again.
so it would've been better, if you just have said, "look, i don't understand" or at least show your doubts regarding it rather then just going with the same line of thought, thinking that Intel guy would answer it better for you, which i tell you,
there isn't any other solution,
he simply can't.
and that is the difference between understanding the answer and not understanding the answer,
BUD.
i think you'r just full of your self.
You're correct it was the write performance that dropped 50% and I consider that more than nothing.
You're more than welcome to think what you want and I will do the same.
BTW, I'm sorry for buggering-up your thread SSDelightful. I just asked a simple question that got outta hand with asinine bickering. Please accept my apology.
Last edited:
yeah, that's exactly the issue, when each is thinking whatever he like and nothing comes out of it.
the purpose of arguing, studying, asking and answering IS to find what is true, what is real, and what is not,
to clear out any doubts, to reach a level of new understanding,
so there's nothing like, you think what you like and i think what i like,
there's nothing like thinking what you like, there is only what is real, true, there.
and that is the point which was important here at this 'discussion', this is exactly what takes one to go on without paying attention to what has been said,
i think this is also a typical egoistic behavior where each is at his own planet.
and actually, have never brought this thread into any "silly" quarrels, there are things which are important to be said and notice, other wise you just spin around yourself endlessly, or till you figure your are spinning which is the first step out.
son't make it look petty and unnecessary,
it isn't,
other wise, i wouldn't have spend the time clarifying all of it.
Alright. First of all thank you for dropping in these forums.
Let me ask you this: you're probably aware that performance on SSDs can scale very well because it's capable of doing true parallel I/O, as opposed to HDDs which are fundamentally serial based since they can only do I/O at one location at a time. Even with multiple heads the actuator still is locked to one position; thus HDDs are serial operation devices by nature.
Now the real question: what difficulties do you encounter when trying to scale performance? Couldn't you just release a 256-channel controller for example and push speeds into 1GB/s+ range? As long as you can predict the I/O pattern (sequential pattern; backwards or forwards of logical LBA), you should not have issues with insufficient queue depth. Upon two contiguous I/O requests you could assume the pattern is sequential and thus apply internal read-ahead to make sure all channels are busy.
So in my mind, i cannot see the reason Intel isn't pushing speeds against the interface limits. The Intel controller with SRAM cache (i believe 256KiB?) in X25-M does a very impressive job at latency and i still think its the best controller around these days even though SandForce and Micron are serious rivals when looking at both sequential and IOps performance.
Would Intel be looking more at PCI-express as interface instead? Why doesn't Intel build a platform on PCI-express where Intel delivers the PCI-express card with Intel native PCI-e to NAND-controller to the customer, and the customer can insert NAND to the PCIe controller much like SO-DIMM. So you could start with 2x 64GiB NAND and the controller will be able to do I/O faster when you add more NAND modules to the PCIe card.
If someone is capable of building such a platform, it would be Intel. High-performance storage would always be in demand, and Intel already has a head-start as it can build good native controller chips of its own.
Last question: ever considered integrating HPA mapping tables stored in DRAM memory chip to on-die SRAM in the controller-chip itself? Howmuch space would be needed and how big would the impact be on random I/O workloads?
That's all i can think of right now.
Thanks for your time! I highly appreciate and commend your efforts of posting on these forums.
Alright. First of all thank you for dropping in these forums.
Let me ask you this: you're probably aware that performance on SSDs can scale very well because it's capable of doing true parallel I/O, as opposed to HDDs which are fundamentally serial based since they can only do I/O at one location at a time. Even with multiple heads the actuator still is locked to one position; thus HDDs are serial operation devices by nature.
Now the real question: what difficulties do you encounter when trying to scale performance? Couldn't you just release a 256-channel controller for example and push speeds into 1GB/s+ range? As long as you can predict the I/O pattern (sequential pattern; backwards or forwards of logical LBA), you should not have issues with insufficient queue depth. Upon two contiguous I/O requests you could assume the pattern is sequential and thus apply internal read-ahead to make sure all channels are busy.
So in my mind, i cannot see the reason Intel isn't pushing speeds against the interface limits. The Intel controller with SRAM cache (i believe 256KiB?) in X25-M does a very impressive job at latency and i still think its the best controller around these days even though SandForce and Micron are serious rivals when looking at both sequential and IOps performance.
Would Intel be looking more at PCI-express as interface instead? Why doesn't Intel build a platform on PCI-express where Intel delivers the PCI-express card with Intel native PCI-e to NAND-controller to the customer, and the customer can insert NAND to the PCIe controller much like SO-DIMM. So you could start with 2x 64GiB NAND and the controller will be able to do I/O faster when you add more NAND modules to the PCIe card.
If someone is capable of building such a platform, it would be Intel. High-performance storage would always be in demand, and Intel already has a head-start as it can build good native controller chips of its own.
Last question: ever considered integrating HPA mapping tables stored in DRAM memory chip to on-die SRAM in the controller-chip itself? Howmuch space would be needed and how big would the impact be on random I/O workloads?
That's all i can think of right now.
Thanks for your time! I highly appreciate and commend your efforts of posting on these forums.
This is the kind of stuff I came here to answer (not that I can and will answer everything in your post). I'll work on a response to this next week when I'm back at work, although I'll be at the Flash Memory Summit in Santa Clara for most of the week.
By the way, is anyway else attending the summit? I heard there might be a surprise appearance!!
you should get stronger inside,
good luck man.
p.s -
your links have been deleted by the mods, you might want to add one at your signature so people can follow, same as JF-AMD does.
regarding the PCI-e SSD -
Intel seems to have tested some PCI-e SSD last year and there might also be one at this year (or next year) new release.
that's a NVMCHI standard developed by IDT in a cooperation with Intel to enhance support for PCI-e SSD's.
GL.
Last edited:
Scott,
Its unfortunate that, in the past, SSD vendors, including Intel, delivered a few half-baked products to market. But that's in the past. Maybe Intel could offer some of these aggrieved customers a coupon or something, or outright replacement if the problems were particularly bad (ie: like the P5 FDIV bug back in the 1990s!).
I personally could, in the future, see a hybrid SLC/MLC drive as being useful. Because let's face it ninety-five percent of the accesses to a typical hard drive (or SSD) are to 10 percent of the data. This would meet the dual mandate of cost reduction, as well as greater system longevity.
Obviously, from a marketing point of view, the challenge for Intel is that, by selling SSDs as a 'miracle system upgrade' (which they are!), this implicitly cannibalizes sales of new and faster CPU's. For instance, my 3-year-old laptop got a SSD upgrade, and wasn't replaced.
On the other hand, for cash-strapped consumers, a SSD may very well be a way for Intel to sell a product, where otherwise, the consumer would simply not buy anything (unable to afford a new computer!).
(Samsung 128gb PM800 user here....but if you want to send me an Intel product, maybe I can write you a piece on how the Intel drive is so much better (or not)).
Its unfortunate that, in the past, SSD vendors, including Intel, delivered a few half-baked products to market. But that's in the past. Maybe Intel could offer some of these aggrieved customers a coupon or something, or outright replacement if the problems were particularly bad (ie: like the P5 FDIV bug back in the 1990s!).
I personally could, in the future, see a hybrid SLC/MLC drive as being useful. Because let's face it ninety-five percent of the accesses to a typical hard drive (or SSD) are to 10 percent of the data. This would meet the dual mandate of cost reduction, as well as greater system longevity.
Obviously, from a marketing point of view, the challenge for Intel is that, by selling SSDs as a 'miracle system upgrade' (which they are!), this implicitly cannibalizes sales of new and faster CPU's. For instance, my 3-year-old laptop got a SSD upgrade, and wasn't replaced.
On the other hand, for cash-strapped consumers, a SSD may very well be a way for Intel to sell a product, where otherwise, the consumer would simply not buy anything (unable to afford a new computer!).
(Samsung 128gb PM800 user here....but if you want to send me an Intel product, maybe I can write you a piece on how the Intel drive is so much better (or not)).
Last edited:
Intel doesn't have control over very many of the factors involved with delivering RAID/TRIM support.
Intel can only make their products standards-compliant.
Obviously the chipset/firmware/driver software group is different from the group in Intel that develops SSDs.
Developing TRIM, which required contributions in terms of low-level SSD firmware, hard drive driver software, and the operating system itself, is quite an accomplishment in the relatively short time frame.
I'm sure that, over time, TRIM commands will be passed properly through RAID. Although the problem there is reasonably complex; for instance, what happens if a SSD is RAID-1'ed with a HDD? The software would have to pass TRIM commands to the SSD, but not to the HDD.
(plenty of people are looking to, or are in the process of RAID-1 combining SSDs and HDDs, especially in applications where reads are extremely common, but writes aren't!)
LokutusofBorg
Golden Member
- Mar 20, 2001
- 1,065
- 0
- 76
Scott, I'd like to know why Intel is not focusing on PCIe solutions. These are set to be game changers in the server space.
How well does contemporary software support PCIe hotplug?
I thought the prevailing strategy in 'servers' was to move away from embedded storage altogether?
The miniPCI-E solutions with 40gb or 80gb should be very interesting as that's enough to get a server running with VMware or another virtualization solution, to boot VMs off a SAN.
Personally I'm surprised miniPCI-E is basically only seen in laptops right now. There's tons of applications, ranging from set-top boxes, home theatre PCs, and even desktops, that could make good use, instead of buying a clunky adapter card.
LokutusofBorg
Golden Member
- Mar 20, 2001
- 1,065
- 0
- 76
FusionIO cards deliver high-end SAN performance. When you're using a SAN for performance reasons, or resorting to large spindle counts in your DAS RAID arrays, these PCIe drives are far and away the better choice. I'm pushing hard at work to adopt this as our strategy instead of trying to build up a SAN infrastructure (which we don't have currently). We have multiple 100+GB databases running in multiple environments. Getting the company to agree to very large DAS arrays or a SAN has been a hard sell. So our DB servers continue to be the bottleneck. These PCIe drives will revolutionize the performance of our applications. I have just about gained approval for a server with 2x Z-Drive R2 drives to vet the tech and provide hard numbers to make the larger decision of whether to adopt this strategy across the board.
Here is the article that started me down this track:
http://www.brentozar.com/archive/2010/03/fusion-io-iodrive-review-fusionio/
And this thread from the Fusion IO forums has a link to the Wine.com case study, as well as a nice explanation from a guy who outlines his server strategy with these new drives (money saved from not buying tons of hard drives is used to just buy identical servers with PCIe SSDs -- higher performance, complete redundancy, and still less money).
http://community.fusionio.com/forums/p/56/564.aspx
I would like to see Anand provide some focus in his SSD articles to this branch of the technology, as I personally feel strongly that this is going to be game changing in the server market, as I said. That is, of course, just my opinion, but I am really excited about it.
i can't seem to over look the arrogance here,
why people are saying "thanks" to this guy for joining into the forum?
isn't it part of public relations?
isn't it is part of his job? which he's doing for himself at the end of things, to be appreciated, to be cherished etc? is he doing it for the people or is he doing it for himself and the company he represent?
i feel it is, they do all this crappy stuff with the G1's, with processor pricings, they were bribing other companies to buy they're products and scott seems very happy and proud to be working there,
you take this behavior and say thanks for them to come out and open a blog which is something which should be done out of commitment and not something you say "thanks" too.., something which is part of natural responsible and human behavior.
and ironically i kind of feel like the "bad" guy for saying things that you polite guys are keeping inside, and feel being treated a bit like crazy for even saying them.
twisted hypocrite logic
why people are saying "thanks" to this guy for joining into the forum?
isn't it part of public relations?
isn't it is part of his job? which he's doing for himself at the end of things, to be appreciated, to be cherished etc? is he doing it for the people or is he doing it for himself and the company he represent?
i feel it is, they do all this crappy stuff with the G1's, with processor pricings, they were bribing other companies to buy they're products and scott seems very happy and proud to be working there,
you take this behavior and say thanks for them to come out and open a blog which is something which should be done out of commitment and not something you say "thanks" too.., something which is part of natural responsible and human behavior.
and ironically i kind of feel like the "bad" guy for saying things that you polite guys are keeping inside, and feel being treated a bit like crazy for even saying them.
twisted hypocrite logic
Because he has chosen to spend his free time to educate other users or at least answer their questions, when he could have just spent that time on his family. Instead, he has chosen to help 'us' by answering any questions we may have.
It is beyond my comprehension that you call this arrogance. If you think he's here to 'convert souls', then wouldn't you say he's done a poor job so far? And i do not believe Intel has become so desperate that they would send out engineers to forums to get a few more products sold. First i think that theory is a myth; second if you want to convert souls you don't let engineers do that job.
Interesting to add: Economics and Psychology have shown people don't buy stuff because of technical arguments, but because of emotional values that the product represents. We only think that it's our rationality that makes the decisions to buy something, while in fact it is our emotions that make the decisions. Companies like Intel know this very well, but discussing this here would be inappropriate. Instead, I would like to seize the opportunity of talking to an Intel engineer; just to enrich my own understanding about technology and to feed my curiosity.
Competitors like OCZ are still selling SSDs without support for NCQ/AHCI, using the JMicron controller. The G1 has long been succeeded by the G2, which will get succeeded soon by G3. The G1 also never was advertised to have TRIM support either now or in the future. So i don't understand your hostility towards Intel for releasing the first generation of SSDs, which did not yet support TRIM. I also believe Intel was the first to release TRIM-capable firmware and adjust their drivers to support TRIM (not in RAID; no).
Also, please understand that he is no manager and can never represent Intel to you as customer. He's here to discuss technical stuff to those interested, since this is his work and we may learn from him.
I'm no moderator, but can i just ask politely to stop complaining to him about stuff Intel did wrong in your eyes? Instead, let us focus on things he can help us with, which is answering technical questions about very cool solid state technology - not questions related to Intel's way of doing business.
hmm,
he's not an engineer really, not underestimating him anyway though that would be kind of a different deal...
this is true customer producer relationship though sadly it doesn't seem he's the right address to talk it with.
yet hopefully, they will open it up this year...
he's not an engineer really, not underestimating him anyway though that would be kind of a different deal...
i think Intel did/is doing some inappropriate things, stuff they should openly speak about,
this is true customer producer relationship though sadly it doesn't seem he's the right address to talk it with.
yet hopefully, they will open it up this year...
Last edited:
Actually... background GC is really good. I can watch my Indilinx drives (have both background GC and TRIM) stay basically flat in terms of performance over time in Raid 0. Current array was built about two months ago and it is running 2-3% under new. It really changes your perspective on the need for TRIM in RAID.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter