DB2 pureScale (db2 v9.8) being annonced

〇〇

原文中没有9.8

bpmfhu

ibm 的site 已经有9.8的号

希望能实现rac的类似的功能

wolfop

原帖由 bpmfhu 于 2009-10-15 11:29 发表
ibm 的site 已经有9.8的号

希望能实现rac的类似的功能

光从和RAC的share disk cluster比较，个人认为PURESCALE的机制远强于RAC。IBM的研发还是比较严谨，绕过开销大的TCP/IP栈，直接利用IB的RDMA，类似Z的全局锁控制节点。一上来就做从2节点到112节点的scale测试。而且本身就是要针对RAC那种要用的好，其实必须做数据分区，避免cache fusion热点而研发的。可是，毕竟是最新版本，稳定性肯定还需要逐渐成熟，另外限制也蛮多，起码明显看到只能用P的特定机型，purexml 等等特性均不支持等。

[ 本帖最后由 wolfop 于 2009-10-15 23:26 编辑 ]

wolfop

原帖由 diablo2 于 2009-10-15 08:30 发表


详细说说

竟然不支持back 和recover，真是晕倒。

diablo2

原帖由 wolfop 于 2009-10-15 23:13 发表

竟然不支持back 和recover，真是晕倒。

Pythagoras

I haven't seen any tech details on PureScale yet, hope it be the great one. Talking about ORACLE RAC, my colleagues always complain it as a fake, there maybe some values in HA, but nothing related to capability improvements, anyone ever saw a RAC with more than 4 nodes?

wolfop

原帖由 Pythagoras 于 2009-10-16 09:54 发表
I haven't seen any tech details on PureScale yet, hope it be the great one. Talking about ORACLE RAC, my colleagues always complain it as a fake, there maybe some values in HA, but nothing related to capability improvements, anyone ever saw a RAC with more than 4 nodes?

Successful RAC more than 4 nodes in production envirment may be all  DSS system. ie Taobao's data warehouse. For OLTP, I don't think their have enough successful case. But if the data can be carefully partiontioned and the application choose the net link according to the data partion, it performance can be also OK. Howerver, the ocssd will be a very weak ring the total system avilability.

bpmfhu

share nothing的架构当然不是share disk可比的，这也是rac在dw的环境下面有种种局限的原因，我曾经就把oracle rac拆成单个节点做dg来跑。对于单节点的数据库目前只能做成热背的模式，purescale如果能实现z平台的data sharing引入到开放式平台，那将相当激动人心

macrozeng

DB2 pureScale is an integrated solution that provides virtually unlimited scalability for all your transaction processing needs while ensuring continuous database availability. It allows you to scale capacity dynamically to meet peak demands, and pay for only the resources you require.

Pythagoras

Yes, 'Derived from System z Parallel Sysplex & Coupling Facility technology, Software based', while system z parallel sysplex is a hardware and software solution. The principle and concept are totally identical to system z data sharing, as I thought. Thus, DB2 X (or any other name of the Vnext) for LUW and Z will be announced at the same time, by PureScale eliminating the biggest obstacle, I believe more and more.