Chinaunix首页 | 论坛 | 博客
  • 博客访问: 422774
  • 博文数量: 31
  • 博客积分: 1860
  • 博客等级: 上尉
  • 技术积分: 1622
  • 用 户 组: 普通用户
  • 注册时间: 2008-12-09 10:07









分类: 系统运维

2010-11-25 17:31:25




Posted by  on Nov 18, 2010 7:51:40 PM

We’ve received  several requests to understand the future direction of Zenoss products  and would like to share a bit of what’s coming up. Before we start, I  need to remind you that what we share is directional and does not  represent a commitment from Zenoss to deliver any of the items  discussed.  Further, our plans may change at any time and you will not  receive any notice of our changed plans.


As  you know, we’ve just completed a major revision of the Zenoss Core user  interface with version 3.0, first available in July 2010.  The new  interface has been generally well-received and thanks to input from our  many users we’ve been able to address several performance and use case  concerns with two maintenance releases.  We appreciate the input.   Please continue!


One  thing you may not know is that we’ve substantially expanded our  development team during this year, and I’m not referring to the turkey  fry dinner outcome!  Our development staff count has more than doubled,  and our internal quality assurance process is better than ever.


So  what’s coming up?  Our next major release process is internally known as Avalon,  and we’ll be delivering new versions of both the Zenoss Core and  Enterprise products. In this post I’ll focus on the performance  improvements around Zenoss Core.  These features will support the Zenoss  Enterprise product as well, of course.


We’re  now helping some very large organizations run their businesses  on Zenoss and improved performance is a key focus.  We’ve closely  examined the ZenHub component and the event manager and are implementing  a number of changes.  Key changes are in the areas of model storage and  the event system.


We’ve also upgraded Twisted and rrdtool, and have  implemented rrdcache to improve performance data read/write performance, although I  won’t discuss these in any more detail.


Model storage implementation change to RelStorage

The  changes we made to the storage implementation of ZODB were driven  directly by customer experiences with long modeling times.  We  determined that the performance of the current ZEO & FileStorage  implementation could be substantially improved by replacing it with  RelStorage.


Instead  of storing the model in the file system-based ZEO, we’re now storing it in MySQL.   This  provides additional tuning options, since database caching is  provided  by MySQL, and simplifies security since database connections  are  authenticated by standard MySQL protocols and there is no longer a  need  to access ZEO in the file system.  I do want to point out that we  are  still storing objects, and there is no relational database access  to the  model.  You’ll continue to use dmd to get at model data.


Our  internal tests have shown approximately 50% better read performance and  20% better write performance, and the shared level two memory cache for  all Zope clients leads to greater levels of throughput for anyone  implementing several zenworker processes.  At one pilot site, we were  able to reduce modeling time by a factor of three.



We’ll  be providing an upgrade migration step as part of the release that will  assist in migrating the model from the existing file storage into  MySQL.  This should take substantially less time than the pre-release  catalog population step in the 3.0 release.  Our internal testing  migrated a 500,000 object database in about 15 minutes.


You  may want to add memcached to further improve performance.  It can be  implemented within Zenoss with just a couple of lines in zope.conf.


My next post about the upcoming release will cover the new eventing system.

阅读(2555) | 评论(0) | 转发(0) |