Skip to Content

Well by now you have heard about HANA from SAP.  In-memory technology is a significant revolution in database and real-time analytic technology.  The rest of the industry appears to be playing catch up because the relevance of this technology is significant and can’t be ignored.

The HANA database resides in memory in a columnar architecture.  All read/write activity is occurring in memory.  The system also writes to a traditional hard drive based database.  The real purpose of this traditional database is for system restarts. So it can load the database back into memory.

HANA technology has also taken the ceiling off of a single server resource limitation by offering multi-node scale-out configurations.  Now you can aggregate multiple systems’ memory together to create massively scalable in-memory database.  Having a 100 terabyte+ system is not out of the question.  You will still require a storage system for the traditional database store and of course now it is going to get bigger.  The benefit is that you won’t need a separate data warehouse to do your analytics, since you will run it against the HANA database that’s in-memory.  For a hosting company or even your enterprises datacenter, this means that you will consume half the amount of infrastructure as you used to do in a more traditional model.

This last week a friend of mine who is a veteran and pioneer of IT infrastructure was telling me about a new storage array system he recently deployed.  Michael Glogowski is the AVP of Core Network Systems at North Island Credit Union.  The storage array was from a company called Violin Memory.  It is a storage system that writes not to hard disk, but to NVRAM.  There is no disk or even any cache involved.  All data is written directly to NVRAM.  The performance gains were significant for Michael.  Jobs that ran for 5 hours now completed in 90 minutes.

I was so impressed.  I arranged a conversation with the folks at Violin Memory.  Beyond the throughput and performance potential of the system, the more remarkable piece was the footprint. Remember that 100 terabyte HANA database that would also write data to a traditional storage array.  Well that storage array could consume 5 full racks of hard drive storage equipment.  The same 100 terabytes of storage in a memory based storage system would consume a quarter of a single rack.

Violin Memory was founded by Ex Fusion-io and EMC executives.  This start-up has a very experienced leadership from the data storage space.  In our discussions about the company and its investors, an interesting relationship was revealed.  SAP Ventures is a significant investor of Violin Memory. Then it all started to make sense.

In one single rack of equipment you could have the computing power to run the largest enterprise business systems.  With a footprint this small, cloud providers can now operate greener, and more efficiently scale their infrastructures.  This also means the age of the on-premise SaaS appliance has just become a reality.

Nowhere in this solution would a hard drive exist. This investment and direction in storage technology has tremendous implications for the IT industry. From my view this completes a vision that SAP has for its own future. I guess it’s safe now for me to dust off my box of old electronics in the garage that holds my View Finder, Polaroid camera, Sony Walkman, VHS machine, floppy drive, and now toss in my hard drive.  If you think the next generation of computing has yet to occur, think again.

To report this post you need to login first.

9 Comments

You must be Logged on to comment or reply to a post.

  1. Kevin Grove

    Sina:

    Thanks for sharing this post. I had noticed the SAP investment in Violin also. I do have some questions/concerns. There is a perception that NVRAM has a somewhat limited rewrite limit, perhaps 100,000 times before degradation may begin. That seems to be a low number for storage for a large database system. Perhaps Violin has overcome that limit. Another knock on NVRAM is random-access address performance. Again perhaps Violin has developed new technology to address these issues.

    Of course as the density of memory continues to increase, perhaps we might soon have the option of the pocket 1-TB Hana system — Personal edition, of course 😉

    Regards,

    Kevin Grove

    (0) 
    1. Sina Moatamed Post author

      Hi Kevin,

      I love the idea of a personal edition!  I hadn’t considered that option.

      No question some benchmarking studies need to be done.  Along with analyzing degradation of memory due to rewrite, we also have to see how well a virtual RAID solution will hold up over time as well.  Following the trajectory of Moore’s Law, I see both issues being addressed.  Much like any disruptive technology, other heavy weights like NetApp, HP and EMC will begin their entrance into the market as well.

      My sense was with the HANA architecture the heavy IO activity would reside in system memory and the storage memory would be a write only system for the most part.  What
      inspired me was the performance and footprint. Two challenges for the delivery of on-premise cloud delivered solutions.

      I will pose the question to the Violin folks and see what their thoughts are on the subject.  Great feedback thank you!

      (0) 
  2. Sina Moatamed Post author

    I had a quick exchange with the folks at Violin regarding some of our concerns.  Here are some of their responses.

    In response to NVRAM degradation:

    Our SLC product has a design guarantee that you can write the entire
    capacity of the array every hour for 10 years before you “wear out” the
    media. For an MLC system this guarantee is to write the entire capacity
    once a day for 5 years.  The system is instrumented so you can see how
    much of the endurance has been consumed so you will have years of warning.

    In response to random-access address performance issues with NVRAM:

    Flash inherently has the same performance for random and sequential IO,
    unlike spinning drives.  We quote our performance with random 4KB IOs.


    Flash storage systems seem be the next hot ticket item.  I read an article that EMC may be looking to acquire XtremeIO.  I need to spend some time looking at their architecture as well, but it’s undeniable that this is the future look of the datacenter and cloud infrastructure.   This next year should bring a lot of activity in this space.

    (0) 
    1. Kevin Grove

      Sina:

      Thanks for following up with Violin. If the performance bears up as they describe, then that seems to exceed the performance specs for regular disk media.

      As you say, this is definitely something to watch for the future.

      (0) 
  3. David Hull

    Please see the detailed testing white paper we just released here: SAP HANA Performance Whitepaper

    Just a few notes – HANA only writes updates to disk for persistence. It is not a “database” on disk that is simply buffered for performance reasons.

    And that disk does not have to be spinning disks. Hardware vendors have all sorts of solutions for storage subsystems for HANA databases, including spinning disk arrays, SSD, and others. Violin or any similar solution would work if a hardware vendor decides to certify it. It is possible they have already, I’m just not familiar with all possible hardware configurations.

    Cheers,

    David.

    (0) 
    1. John Appleby

      If I’m being honest mate this white paper is pretty disingenuous: it implies 16x8TB = 128TB RAM, when actually it is 16x512GB = 8TB RAM.

      The initial dataset might be 100TB but it compresses to <4TB (as expected). Of course your mileage may vary on real datasets and the average compression with HANA is 5:1.

      The results from HANA are still deeply impressive as always but I think the paper is ambiguous and therefore misleading.

      (0) 
      1. David Hull

        The paper says “The test system configuration is a 16-node cluster of IBM X5 servers with 8TB of RAM…” – the 8TB of RAM refers to the cluster, not the individual servers. Immediately after that, it says “Each server has … 512GB of RAM”

        You think that’s misleading? Would it be better if we added the word ‘total,’ as in “with 8TB of total RAM”?

        Yes, we do get very high compression, ~27x I think. That is a side effect of using only generated SD data. Had we used a combination of several types of data across several cubes, we likely would not have gotten the high compression rate, but then we would also not have been able to run single queries across 100 Billion records. 🙂

        (0) 
  4. John Appleby

    Do you have a stock holding in Violin? 🙂 Curious timing here between SAPPHIRE, some leaked press announcements and this SCN post makes me wonder if you have a vested interest here.

    To put this in context, a 100TB HANA appliance (note that 8TB is the largest certified appliance though I hear there are larger systems running in a lab environment) would currently require 1200U or 30 racks of IBM hardware. 20 of those racks would be dedicated to servers and 10 to disk storage. Currently the only people providing blade HANA systems are Hitachi and Cisco, both of which have 2TB in 10U = 500U of servers (12 racks) for 10TB. It is the servers that are the problem, not the storage!

    Of course HANA really comes of age with the next generation servers which can do 8TB in 10U and as you point out the next-generation storage which can look like Violin or one of many competitors and are compact, high performance and green. SSD is in the process of commoditising and I’m not sure why Violin has a special value proposition compared to its competitors, many of which like IBM, HP, Dell and Hitachi provide a combined HANA server and storage single package.

    (0) 
    1. Sina Moatamed Post author

      Hey John,

      Hahaha, no I don’t have stock with Violin.  I’m pretty sure they are just a start up.  Only executives of those companies are the real winners once they are acquired or go public.  Unless of course you are Instagram, then every employee won that lottery.  No, no, I’m not bitter about that one at all. 🙂   Honestly, I never heard of Violin until a couple of weeks ago when my friend Michael told me about his deployment.  It wasn’t really something I would even write about, until I saw SAP co-invested 50 million in Violin.

      You are absolutely right about drawing the distinction between a 100 terabytes of RAM in a system, and 100 terabytes of database in a system.  Given the 5:1 ratio of compression, I was envisioning 20 – 1U systems with 1 terabyte of RAM for instance.  This would be about half of a rack.  The NVRAM storage array would be about 9U.  This really wasn’t, I guess the point of why I wanted to write about this.

      The real question is why would SAP who has historically been fiercely agnostic when it came to the hardware stack, be concerned with investing in a storage company?  This is Oracle’s game to buy the whole stack, not SAPs.  Even for their Cloud hosting environment, they would just buy infrastructure as they needed from any vendor that would serve them best.  Sure its SAP Ventures and not SAP themselves, but I don’t think their investment was altruistic.

      For a long time, I have contended that getting large enterprises to invest beyond CRM in the public cloud will be a challenge.  My feeling is that SAP has recognized the same thing and that the best way to bring cloud services to large enterprises is in the form of an on-premise appliance.  Despite the hardware consolidation that HANA presents, there still was no viable and supportable appliance model, until I saw this technology.  The possibility for any SaaS or PaaS company wanting to make its way to the large enterprise will be a result of them offering a hybrid of cloud deliveries.

      To date all of SAP Cloud offerings for large enterprises have been LoB apps for the most part.  For a company that’s built its reputation on the suite concept, it’s interesting to see the cloud space not develop with the same model.  ByD is currently marketed to medium sized companies and subsidiaries.It certainly could scale, but does not have all the vertical capabilities that the Business Suite does.  If SAP wants to deploy a cloud business suite for the large enterprise, then they must be working on an appliance delivery model. Obviously, getting the Business Suite to fit in that box is a different subject.

      (0) 

Leave a Reply