Comments on: “Federated Array of Bricks” – Deja Vu? http://www.pythian.com/news/331/federated-array-of-bricks-deja-vu/ News and views from Pythian DBAs Fri, 10 Feb 2012 13:01:25 +0000 hourly 1 http://wordpress.org/?v=3.0.4 By: Alex Gorbachev http://www.pythian.com/news/331/federated-array-of-bricks-deja-vu/#comment-10828 Alex Gorbachev Sat, 16 Dec 2006 04:49:56 +0000 http://www.pythian.com/blogs/331/federated-array-of-bricks-deja-vu#comment-10828 Just for reference - <a href="http://storagemojo.com/?p=324" rel="nofollow">here</a> is the <a href="http://www.hpl.hp.com/research/ssp/papers/2004-10-ASPLOS-FAB.pdf" rel="nofollow">newer</a> version of the white paper. Just for reference – here is the newer version of the white paper.

]]> By: Alex Gorbachev http://www.pythian.com/news/331/federated-array-of-bricks-deja-vu/#comment-9706 Alex Gorbachev Wed, 06 Dec 2006 17:49:35 +0000 http://www.pythian.com/blogs/331/federated-array-of-bricks-deja-vu#comment-9706 I wonder in Big Iron will be back in Oracle camp or Oracle will continue pushing <a href="http://www.oracle.com/technologies/grid/megagrid.html" rel="nofollow">Project MegaGrid</a>. Let's see how far Sun will go with its <a href="http://sun.com/blackbox/" rel="nofollow">Project Blackbox</a>. I wonder in Big Iron will be back in Oracle camp or Oracle will continue pushing Project MegaGrid. Let’s see how far Sun will go with its Project Blackbox.

]]> By: Kevin Closson http://www.pythian.com/news/331/federated-array-of-bricks-deja-vu/#comment-8919 Kevin Closson Thu, 30 Nov 2006 00:30:23 +0000 http://www.pythian.com/blogs/331/federated-array-of-bricks-deja-vu#comment-8919 oh, I forgot to mention. There is nothing cheap, simple or freely effective about RAC. People can afford RAC today because they are migrating from systems currently cost more than RAC licensing for the HW maintenance contracts alone. So on paper it looks good to go from an outdated legacy system to a cluster of commodity servers even though RAC costs exponentially more than the commodity hardware. All told, there are large numbers of legacy servers that could be decommissioned today and ran entirely on a single node commodity box. After all, the best UE10K TPC-C number was on the order of 155,000 and there is no shortage of current results from sustems with 4 and fewer sockets doing that kind of a number. Don't need RAC to beat an aged legacy system... oh, I forgot to mention. There is nothing cheap, simple or freely effective about RAC. People can afford RAC today because they are migrating from systems currently cost more than RAC licensing for the HW maintenance contracts alone. So on paper it looks good to go from an outdated legacy system to a cluster of commodity servers even though RAC costs exponentially more than the commodity hardware. All told, there are large numbers of legacy servers that could be decommissioned today and ran entirely on a single node commodity box. After all, the best UE10K TPC-C number was on the order of 155,000 and there is no shortage of current results from sustems with 4 and fewer sockets doing that kind of a number. Don’t need RAC to beat an aged legacy system…

]]> By: Kevin Closson http://www.pythian.com/news/331/federated-array-of-bricks-deja-vu/#comment-8918 Kevin Closson Thu, 30 Nov 2006 00:26:25 +0000 http://www.pythian.com/blogs/331/federated-array-of-bricks-deja-vu#comment-8918 This is just a bit of left-hand networks or Pivot3 type thinking (a little ISILONish too). The following slide set shows a more current state of the technology: http://www.ysaito.com/fab2003-stanford.pdf I'm sure it will have uses, but taking something this far without measurements with commercial workloads is a bad sign. All in all, I'm sure ASM will make it all work :-) No, honestly, the main problem with these sorts of things is reconfiguration...misery This is just a bit of left-hand networks or Pivot3 type thinking (a little ISILONish too). The following slide set shows a more current state of the technology:
http://www.ysaito.com/fab2003-stanford.pdf

I’m sure it will have uses, but taking something this far without measurements with commercial workloads is a bad sign.

All in all, I’m sure ASM will make it all work :-)

No, honestly, the main problem with these sorts of things is reconfiguration…misery

]]>