Compellent is adding integrated file-level access to its SAN product, and using Sun's open source ZFS to do so. The zNAS product is a 1U enclosure running the file access software on diskless, dual quad-core Nehalem hardware, which can be clustered in two nodes for high availability. It has 1Gbit Ethernet client access, with …
So Compellents conception of "Unified Storage" is a traditional SAN array + a couple of 3rd party NAS head ducttaped together?
Given enough ducttape is available, why not add a couple of servers, a couple of switches and an administrator (male or female) to "create worlds first selfadministering unified computing thingy"
The unified query (YUK) paradigm, essentially a whopping big space containing lists of objects (i.e. tables,) events (i.e. immutable time sensitive effective sequences,) and generalised pools of objects, has come a long way in a short time. The functional results of unified query, (i.e. the ability to optimally generate efficient, but very complicated resultsets can only really be done, by having super huge data space, and a parallel despatcher engine, which fires off multiple implementations of each of the resultset processes and terminates the lot when the most efficient one has finished.
The derived data component of unified query, also provides the ability to build and throw away enterprise service datasets (i.e. derived data which can be reported, but whose underlying data can't, and whose cost of generation is too expensive to be done multiple times anyway.) An example of the latter would be the generation of average speeds of cars by registration number. The host system can know the locations and times, and compute running averages which can be replicated out individually or as a set, but withold the roads upon which they were clocked.
A vast array of virtual memory has the ability to answer ridiculously complex queries due to the scale of fast space allocation, that conventional computing can't.
It would be ridiculous to ridicule this in the fashion of Eigen, just because the maths hasn't yet got an application.
*** Resultset process - a set of processes which all do the same thing in different ways.
eg. a simple example would be sorting unknown amounts of data. The despatcher contains a list of functions (e.g. shell, quick, bubble) that sort data, and gives them all a go at it, they all work in different ways, and so one of them is likely to be more efficient. When one finishes, the others are aborted, causing their enterprise dataset to be deleted and the space reclaimed, the one which finished has its data kept (or deleted, depending on cost benefit analysis - a result set is just a function* of dry (- don't repeat yourself) data anyway.) This is the exact opposite to current RDBMS which choose a single algorithm, based on clever guessing.
At the enterprise scale, the sort will be one of a hierarchy of functional results, in a Lisp like defun() though using Linq like functionality. The hierarchy of the functions providing the answer. This can simply not be done using conventional architectures.
"Here we come?" Unified storage already here, no?
Article states "Unified storage here we come." Second time in two months I see article about other storage companies doing what Sun already did. Anyone talked to Oracle lately about this? Or is this space for start ups only?
Sun launched unified storage in 2008, out of project Amber Road. Today Oracle sells it as the Sun 7000 unified storage family which incorporates SAS/SATA, flash, ZFS, Dtrace, fibre channel, iSCSI,... Perhaps there are differing definitions of what unified storage means, and when it is available?