On Sat, Mar 21, 2009 at 8:50 PM, fooler mail <[email protected]> wrote: .. > ibm's db2 and ncr's teradata are the > leaders when it comes to shared-nothing database..
Teradata isn't used for transactional at all.. And DB2 hasn't won on any SAP ERP benchmarks recently. So shared-nothing is not the end-all solution. I did point out earlier that for business intelligence workloads (which is what TPC-H is) Teradata is very dominant in this space. >> Very true. Expertise doesn't come free with any technology. > > is this coming from a closed source mentality? how about the open > source mentality? No, it works both ways. I don't think MySQL Cluster can be tuned effectively by someone like... me. It takes a lot of effort to tune any technology the best way, and that costs time and/or money. If expertise was easy to come by, we would not be on this mailing list. .. > but your question is who uses mysql cluster and alcatel and vodafone > are examples the ones using it... panterra networks for their on > demand unified communications.. bredbandsbolaget a broadband internet > provider in sweden for their subscriber profiles and authentication.. > viasuisse AG for their real-time traffic information and others... amusingly, all of these are from MySQL Cluster's wikipedia entry.... but I digress. .. > the industry standard for complex business intelligence processing is > the TPC-H benchmark... large-scale and multi-terabyte procesing shows > that databases using shared-nothing architecture are the ones heavily > used in this arena... furthermore.. shared-nothing architecture have > been shown to produce tremendous scaling results for OLTP application > (aside from database) when partition.. Yes but TPC-H is a data warehousing workload. Which is not transactional. None of the major OLTP vendors releases TPC-C workload results anymore, but IBM and Oracle do release SAP ERP benchmarks. >> As N gets large, RAC scalability for writes gets to the point of >> diminishing returns. Not a limitation unique to RAC, even SMP boxes >> suffer from this. > > simply because shared-everything is a SMP while shared-nothing is a MPP... And yes... for the commodity market, everybody is still going with SMP. My point is that MPP is either exotic or expensive. Expertise is harder to come by. >> Shared-nothing is a known scalable solution, almost all the Top500 >> supercomputers are MPP. But shared-nothing has been traditionally very >> hard use get working properly for traditional workloads. > > TPC-H will show you the reality... see my last paragraph below... TPC-H is not a transactional workload, which is what most people do. >> Greenplum and Netezza (and DATAllegro) have tiny market share. The >> much-derided RAC has over 40% of the worldwide data warehousing >> market. So if you went for these products, you'd have a much smaller >> community to turn to for help. And these products (well even Exadata) >> aren't designed for OLTP workloads which I believe are most >> interesting to PLUG readers. > > naahh... you are talking as salesman as well as a marketing guy here.. > we should focus the technical merits of a shared-everything versus > shared nothing... See, this is where things get to religion. Shared-nothing is a well-known scalable technology, but for the vast majority of the market, it's still exotic. Top500 doesn't count. Teradata users don't count (they have lots of money). Core banking systems, even our beloved web forums, mail servers, most of them run on shared-everything. Shared-nothing is still not within the reach of the "common technical user." Which is my entire point -- shared-nothing requires more rocket science skills than shared-everything. Of course we all know the limitations and performance plateau of shared-everything: but that plateau is still high enough for most use cases. .. >> Besides it's built for a query workload where it's permissible to lose >> data. Which is an important use case but not one that most of us can >> use. > > thats your perception as you simply show that you dont know the inner > workings of a shared-nothing architecture... shared-nothing > architecture is reliable too.. I was talking about Google's use case being query-heavy, and tolerant of data loss. Not shared-nothing in general. Off topic but one of the products I've been working with --- Oracle Coherence --- is a distributed shared-nothing architecture. So I'm not exactly ignorant of "the workings of a shared-nothing architecture." .. > ok then.. assuming you have the million bucks money.. you are not > working from oracle and looking for a scalable database... ill refer > to you the TPC-H benchmark website: > > http://www.tpc.org/tpch/ > > which database clustering solution you would choose as a technical guy > as usually we technical guys are always looking for price over > performance? I would use Exasol for TPC-H. No brainer. 1/25th the price of Oracle solution and stratospheric performance. But TPC-H is not transactional and is hardly something that is relevant to PLUG readers. Most BI/DW workloads simply suck at transactional loads. But they don't have to be good at transactional loads because complex (read-only) query performance is paramount. But again, that's not something we on PLUG generally do as part of our daily lives. -- Orlando Andico +63.2.976.8659 | +63.920.903.0335 _________________________________________________ Philippine Linux Users' Group (PLUG) Mailing List http://lists.linux.org.ph/mailman/listinfo/plug Searchable Archives: http://archives.free.net.ph
