back to article Symbolic IO reveals tech bound to give server old guard the willies

Symbolic IO has rewritten the server hardware and operating system rule book with its its computationally defined storage (CDS) and says apps like database queries run 60 times or more faster on its system. If these claims are real then server suppliers like Cisco, Dell, HPE and others are in for a mighty market shock as one …

  1. Arthur the cat Silver badge

    WTF?

    For example, the sequence "01100010111101110011100" could be replaced by "0110", meaning four bits, a symbolic bit marker, are used to represent 23 original bits; this is just my example of how the algorithm could work.

    When I wer't lad, this was called compression. In fact it sounds remarkably like Huffman encoding with the table kept by the OS.

    When the newly encoded, and now much smaller, image and its data files are read into IRIS memory, the SymCE operating system computes the original values from the bit markers and executes the application with no or minimal IO, thus accelerating it.

    And that's called decompression.

    1. Pascal Monett Silver badge
      Pint

      Re: WTF?

      Agreed. When I read that I thought "ZIP file" immediately.

      I'm always happy for anything that speeds up rust storage, and it's a good sign that the people involved seem serious, but there's got to be something more than a basic encoding of that kind, otherwise Windows would be a damn sight faster than it is.

      Here's to the future of SIO anyways !

      1. Charles 9

        Re: WTF?

        Thirded. I didn't know the term off the top of my head, but I recognized the technique immediately. I also recall one of the caveats of the system, especially when it comes to modern content like compressed multimedia files. That being they're usually already compressed (SO much in fact that most files are lossy). So how do they expect to get more efficiency out of files that have been squeezed so hard as to bleed quality?

        And as for the anti-theft features, one thing I wonder about GPS-based tech is what if someone exploits it in a DoS-style attack by employing something like a fake GPS transmitter. The IRIS has a self-destruct mechanism? What if some jerk found a way to pull off a premature trigger?

  2. Anonymous Coward
    Anonymous Coward

    "A series of patents covers this technology and we wrote about them here."

    What bullshit is this? You're describing dictionary compression, which was perhaps state of the art 60 years ago.

    "They are therefore effectively encrypted and inherently secure."

    I just threw up a little. No. That's not how it works.

  3. robertcirca

    Huffman Code

    This smells like a multidimensional Huffman tree. Using two bits as a starter you could go 4 different ways in 1 step. Doing the same again in step 2 you can go 16 different ways just using 2 comparison steps.

    Oh, I do not really have an idea how it works. Maybe I had a beer too many.

  4. Anonymous Coward
    Anonymous Coward

    Sorry, ain't some magic formula unless, and that's a yuge unless, they've come up with a new compression method. I can't blame the patent office people letting it through but, as a practitioner of the art, geez. Personally, an ASIC is a nice touch. I prefer an FPGA here so I can recode it as circumstances change. That's me though.

  5. Anonymous Coward
    FAIL

    Are you for real, Chris Mellor?

    Is author "Chris Mellor" as real person, or is this one of those articles written by an AI bot?

    C'mon, man, you work for The Register. You're supposed to know something about this stuff and (main reason we read El Reg) see all vendor pitches with a witty, fair but slightly jaundiced eye.

    You are *not* supposed to breathlessly paste their slides and fluff into an article as if that nice Mr Moses had just given you two tablets and asked you to compile a Top Ten list.

    You didn't think that storing a sequence of bytes with a shorter sequence of bytes might already have been invented? And refined dozens of times in various algorithms?

    You didn't think that a system where the data could only ever be read by the system that stored it screamed lock-in and eventual obsolescence?

    You reported, without comment, the vendor's claim that a data replication technology that did not exist at the time of the demo would "blow peoples minds".

    You went looking for independent verification of their claims, and spoke to...the CEO, CTO and one of their investors?

    That's it? That's your journalism?

    I wish IRIS nothing but the best, but this was feeble work by El Reg's standards. Send Trevor Potts to the next one please. He knows how to test storage. and if you ask nicely, he won't always set it alight.

  6. John Smith 19 Gold badge
    WTF?

    I read somewhere about how hard drives used to do something called RLL compression

    IE what was on the disk was not what was actually in the file. Which seems to be this things deal.

    An ASIC implies faster clock rates than I've seen with FPGA's. They seem to struggle > 1/2 GHz.

    This beast looks like the server for the TLA's

    But I've got some real problems with this story.

    a) It's all proprietary. b)Who controls the destruct button controls the servers. Is that customer or the company?

    c) The big one. If that much of a speed up was possible why has it not been done before now? This is supposedly an x86 box at heart and that architecture has been around for close to 40 years.

    Compressed executable image and compressed media images on disk. OK. Decompressed and executed but still needs small memory footprint? Sounds like executable is being shared between streams. That's time sharing to me. I would add the "thumb" instruction set on an ARM sacrifices some flexibility (smaller register file) for shorter, more densely coded instructions.

    As for only 21% of server being used, that would be the window with a msg saying "__%" and the number 21 in it? Like those handy countdown timers all movie timebombs have.

    My instinct is the reporter has been deluged with a hosepipe of data and left with a bit of information overload.

    Yes it sounds stunning, astonishing and a massive game changer. Kind of like that VR helmet from a few months ago.

    Let's see if it's any more real.

    1. SirWired 1

      RLL was just encoding, not compression

      RLL (Run-length Limited) is just a data encoding scheme. HIgh-speed serial data storage and transmission don't work if you just dump your bits on your media... you run into a couple problems:

      - Clock signals are never 100% accurate or synchronized. To get around this, all data receivers use something called a Phase-Locked Loop to "lock on" to the proper timing. But for this to work, it needs some volt transitions to detect. Encoding usually forces a certain minimum number of bit transitions onto the wire; without it, a string of all 0 or 1 would cause the loss of clock sync.

      - The threshold for the receiver can drift. This is called "DC Offset". If there's a continuous stream of, say, 1's, the receiver will start to have problems telling the difference between 0 and 1. Encoding tries to nudge the number of 0's and 1's to be roughly even.

      What RLL isn't is compression. Not even a bit. Under any circumstances. They merely allow data to be transmitted faster/stored more densely than might ordinarily be the case. (In this context, "storage density" means "can cram more physical bits on a given area of magnetic media"; it doesn't refer to compression.)

      1. Charles 9

        Re: RLL was just encoding, not compression

        I guess 8b/10b and the like (like 128b/130b) are all variants of the technique. You're right that RLL isn't a compression technique but an encoding technique meant to maintain clarity when transmitting data through a physical medium.

  7. kedmison

    Something doesn't add up...

    I'm trying to make sense of the StorModule I/O numbers, and I think I'm missing something.

    The StorModules are apparently DDR4-based, and connecting to a X86 chip. 68GBps is quoted as the throughput for a StorModule.

    68GB/s is a magic number for total memory bandwidth of Intel chips like the E5-2680 v3 (Haswell) chip http://ark.intel.com/products/81908/Intel-Xeon-Processor-E5-2680-v3-30M-Cache-2_50-GHz

    That Haswell's memory IO of 68Gbps is the sum of the bandwidth for the 4 on-chip memory controllers at a DDR4-2133 rate. As I understand, X86 memory is also half-duplex, so if the StorModules are based on DDR-4, then the quoted 68GBps is the sum of read bandwidth and write bandwidth.

    The PCIe Gen3 entry is a bit suspect as PCIe is bidirectional, and the quoted speed doesn't actually match any lane count that PCIe typically uses. For instance, PCIe Gen3 x8 theoretical bandwidth is about 7.8GBps. They may be trying to quote realistic vs. theoretical, but in my limited experience the realistic PCIe throughput is a bit higher than that.

    PCIe Gen3, of course, supports lane counts up to x16 (15.6 GBps). PCIe Gen4 is on the horizon which is supposed to roughly double the speeds of PCIe Gen3. If forecasts are correct, it'll perform at about 31GBps bidirectional.

    That's not far from the 68GBps quoted for a StorModule (if I am correct about the full-duplex/half-duplex thing).

    I'm still quite curious about the CDS, how the CPU plays a part on the StorModule DIMMs, and how all this relates to what is coming down the pipes with Gen-Z.

  8. Electron Shepherd

    Sqeaky Serpent

    The problem I have is that my serpent is making a horrible sqeaking sound. It probably needs a bit of lubrication.

    But, having read the article, I think that Symbolic IO have something suitable...

    1. Arthur the cat Silver badge

      Re: Sqeaky Serpent

      Henceforth I shall attempt to make "squeaky serpent lubricant" part of my vocabulary. Well played sir.

  9. JCD56843
    Big Brother

    Chris is a puppet? I think not...

    Great article Chris.

    The commentary is pretty humorous to say the least. Narrow minded personalities always make me laugh. Most of the comments make me think of a group of cavemen arguing about the outcast who has the dream of fire. FIRE? This is impossible.

    I am a huge fan of Mellor's and his articles. The article above states he was in the room and he saw two demonstrations with this beast present. So I guess Mr. Mellor saw fire and is now a casted out with the other caveman outcast.

    Mr. Mellor has written about this company several times, I am sure if it was nonsense as some of you claim he wouldn't have written story number two never mind this article.

    Based on narrow minds illustrated above can we conclude the following:

    1. Mankind will never find a cure for cancer

    2. Mankind will never cure AIDS

    3. If I didn't think of it then it mustn't be true

    4. The computers and storage we have today are the best we will ever have and new technology is hogwash

    5. Everyone commenting here are the smartest people in the planet.

    Bloody cavemen

    In my datacenter we have a vast number of platforms we have purchased. Many of these platforms are just not great. Not much of what we have at our scale makes sense anymore nor is our administrative life easy. So I'll assume this IRIS thing is real and thank you Chris for showing us the future!

  10. Ketansheth

    Haters Gonna Hate

    Unbelievable how people down play technology they cannot comprehend or different than what they are used to. Take a moment and do some investigation before you form your conclusions.

    Mr. Ignomirello,

    Congratulations on your ground breaking technology and keep on pushing the envelope.

    "Be so good they can't ignore you."

    Steve Martin

    1. Charles 9

      Re: Haters Gonna Hate

      "Unbelievable how people down play technology they cannot comprehend or different than what they are used to."

      As an engineer once said, "You can't fight physics." Tell me how you can get massive compression out of a multimedia file (and note that one of the demonstrations was to stream 80 such files at a time, with only 21% CPU usage) that's already been compressed so much as to lose quality. And how many times have been warned about something that sounds too good to be true? How many times have we read articles about some extraordinary new technology (Remember 3D holographic crystal storage?) only for it to never hit the market (That was over 20 years ago IIRC. Where is it?)?

      So we may be jaded, but not without reason. For that matter, where's the independent test proof complete with published results?

  11. Anonymous Coward
    Anonymous Coward

    WTF? Why can't something new actually come along that actually works!!!

    Whoa, Hold on a Minute.

    All you old school mathematicians and even possible enemies (probably from some tree-hugging company) of something new should read all this a little further…

    Pardon me if it’s a bit early out here on the left coast of CA, buried in the Red Ocean of SDS, COTS, Private Cloud, etc, but I just read all of this along with Drunken Data, let alone the not so nice comments from the peanut gallery that even dumps on Chris Mellor.

    It’s high time something new came around the corner. And I hope for damn sure it is. Good hardware and Great Software/Firmware should come along every so often so we just don’t polish the turd and kick it down the road.

    When you read closely and look at the descriptions and pictures here on the Register AND from Drunken Data (Jon Toigo must have seen this personally from what I read), THERE IS SOMETHING DIFFERENT HERE Gents. NIH is the worst thing people can do..let alone be Human Multi-Henry Inductors…

    Compression, Decompression takes serious computation time. It also EATS memory. Look at every damn SDS play out there from bottom up Pure, to Modified Open Source Storage Arrays to Converged Open source plays like Nutanix, Pivot3, etc. and you SEE HUGE computational engines with large memories and tons of SSD…Most have such horrific performance as they do the compression, decompression thing, let alone those that do de-dupe.

    And btw, looking this morning at Brian and the other guys, Peglar, and Sicola’s background, those guys are not stupid and would not stoop to PT Barnum style marketing,IP, let alone products.

    This thing does NOT eat CPU, It does not Eat memory…it eats data and spits it out like frickin DNA! And without eating the CPU, so it actually CAN do more work!

    WHO says there can’t be new math maybe mixed with old math? See the movie ‘Hidden Figures’?

    One has to know something is up when it was actually SHOWN in public in front of some of what looked like some serious press, let alone letting Jon Toigo look at it…He eats smoke and mirrors for lunch.

    Given that this thing is going GA, they must have wowed some customers to take a new server, let alone something that does things to and with the initial input date….

    Open your damn mind guys. Math wins, Architecture wins…let’s see what IS….finally….hopefully

    1. find users who cut cat tail

      Re: WTF? Why can't something new actually come along that actually works!!!

      You are correct that the prescription does not explicitly prohibit washing down the pills with kool-aid. But, sir, please do not forget to actually take the pills next time.

  12. John Smith 19 Gold badge
    Meh

    Welcome new members JCD56843, Ketansheth and AC

    Well you certainly sound very excited, and in a first ever post for at least two of you.

    Software compression/decompression is quite processor intensive and should be farmed out to an ASIC for maximum benefit, although I guess a GPU could help as well (which is also a sort of ASIC given they are usually single source).

    Yet quite a lot of embedded Linux hardware does this in software to save on flash storage for the software.

    What we're having trouble with is just how "novel" is this IP? Most of the parts seem to have been done before and never achieved what seems to be a massive speedup on an architecture that's existed for decades.

    As for things like real time 3d rendering this was a demos for a single and 4 transputer systems back in the 80's.

    However I've been in situations where I'm looking at a product and I can't see what the USP is until I dug below the surface to understand what really made it different.

    Right now I'm not seeing how what they say they are doing is giving the benefits they say they are getting. Perhaps the company should try to improve it's communication shills a bit more.

  13. Anonymous Coward
    Anonymous Coward

    How can you get a GPS signal in a data center? Most times, I can't even get a decent cell signal in one.

  14. SirWired 1

    And again, how is this not compression?

    Every time these guys get asked "how is this not just compression", all that comes back is a bunch of ridiculous arm-waving. 95% reduction in RAM usage? Seriously? This is at least the second time you have written about these guys, and STILL not prodded them for answers to these basic questions. I'm rather concerned that every single skeptical B.S. detector you have isn't on full-alert with these folks.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon