> The marriage between NuoDB and HDFS shows great promise. We've
> benchmarked NuoDB running at 27,000 transaction per second with 8 or 9
> transaction nodes. At that point, the only thing keeping the archive
> manager alive is a disk bandwidth filter. The application is a model
> airline reservations systems where we can control the ratio of flight
> queries to bookings. The 27K figure was with a 50% update ratio.

I guess we are talking about the currently over-stressed term "commodity
hardware/disks" here?

> The key to this all, of course, is MVCC and network-centric storage.
> The network bandwidth available with even cheap GB Ethernet and switches
> dwarfs the pathetic ability of even raid to absorb the data.
>
> If or when the database community recognizes that serialability is a
> sufficient but not necessary condition for ACID transactions, this is
> going to be an interesting industry. Until then, I think NuoDB is the
> only game in town unless you want to design your application and
> business around VoltDB.

ACID transactions in a distributed environment is a new game. Well,
probably not that new, but NoSQL can't deliver and e.g. Oracle Real
Application Cluster still strugles with relational databases not being
very good for a scale out scenario.

> My rash prediction du jour is that in five years noSQL is going to be
> listed among IMS, IDMS, ADABASE, and DBMS-11 as technologies that seemed
> bright be disappeared without a trace or legacy.

I'm not good when it comes to prediction. I can't even predict if my
twins smash their heads together tomorrow. NoSQL, ah well I know HBase a
bit, isn't suitable for OLTP systems now, possibly won't ever be, but
for (long-term) storage of mass of data, which is querable, although not
ad-hoc (Hive is another story ...) and processable with e.g. the
MapReduce framework, I would say, for such a scenario it possibly lives
longer. The future will show.

Regards,
Thomas

> On 1/21/2012 5:30 PM, Thomas Steinmaurer wrote:
>>
>> Hello Jim,
>>
>> [snip]
>>
>>> HDFS isn't perfect, but it isn't unique, either. There are a variety of
>>> highly distributed, replicated, high performance storage systems out
>>> there, and most likely more will be showing up. It's the way to go.
>>> Definitely.
>>
>> Being currently involved in a Hadoop/HBase project for storing a mass of
>> sensor data, HBase is doing pretty fine on top of HDFS. HDFS is widely
>> used in huge installations (Google et al.) and AFAIK the problem with
>> the name node currently being a single-point-of failure, is going to be
>> sorted out.
>> http://www.slideshare.net/cloudera/hadoop-world-2011-hdfs-name-node-high-availablity-aaron-myers-cloudera-sanjay-radia-hortonworks
>>
>> Needless to say, that a lot of money is currently going into the "Big
>> Data Management" stuff. Will be interesting to see how NuoDB will merge
>> the two worlds: SQL and Scale-Out or in NuoDB terms "Elastically
>> Scalable", as HBase ensures row-level consistency only and no real
>> transaction support, no SQL language etc., but it scales very nicely
>> when you have sorted out how to design your row-key. And I like the
>> flexibel model to add qualifiers, aka fields in the relational world,
>> dynamically without the need to change the underlaying model.
>>
>> Regards,
>> Thomas
>>
>>> On 1/21/2012 7:53 AM, mariuz wrote:
>>>>
>>>> Something to read for the weekend
>>>>
>> http://www.simple-talk.com/sql/database-administration/relational-databases-and-solid-state-memory-an-opportunity-squandered/
>>>>
>>>> And by the way Fusion-io Breaks One Billion IOPS Barrier
>>>>
>>>>
>> http://www.fusionio.com/press-releases/fusion-io-breaks-one-billion-iops-barrier/
>>>>
>>>>
>>>
>>>
>>>
>>> [Non-text portions of this message have been removed]
>>>
>>>
>>>
>>> ------------------------------------
>>>
>>> Yahoo! Groups Links
>>>
>>>
>>>
>>
>>