Cloud Computing

flyspirit

Chris Auld from Intergen is talking about cloud computing on Tech-Ed NOW

cynosure

我对云计算的兴趣主要还在于它能提供计算能力的租赁，但对存储的租赁，不是很感兴趣。

长期来看，我觉得云计算的兴起，主要还在于目前本地的计算能力无法提供分布式，尤其是社会化的计算需求，如果有一天本地的计算能力能够强大到满足这种需求（传说中的量子计算机？），云计算也肯定会犹如当年Unix终端一样烟消云散。

starchu

原帖由 flyspirit 于 2010-8-27 11:40 发表
Chris Auld from Intergen is talking about cloud computing on Tech-Ed NOW

Everybody is talking about cloud, but what they are actually doing developing the web apps using PaaS. It is the simplest form of involving in Cloud computing. Everyone who knows .NET or Java can do the same without deeply understand Cloud computing. Don't really think this kind of tech talk really makes sense.

starchu

The biggest challenge for Cloud at the moment is the vendor lock-in problem. It is happening in every layer of Cloud.

SaaS is always offered by different vendor, you are locked in naturally.

The situation in PaaS is the same, every vendor offers a different platform for developing cloud applications and also host them in different data centers.  Could you able to migrate from one platform to another? Not at all at the moment, basically, if you start your cloud apps with Microsoft Azure or SalesForce.com or GoogleAppEngine, you have no other choice of moving one application from on company's data center to another one.  It is not like the traditional way of software, where you choose one language to write, but you are able to migrate from one place to another by installing them. In cloud, you have no way to control that, no installation is ever required for any piece of software. If you stick with one company, you have to stick to it for ever.

IaaS layer is even worse. People talk about virtual appliance as one major application for IaaS, which you can build customized virtual machine template and offer them as a bundle. It all looks perfect, however, it locks you into a particular template format. What does it mean is that, if you are using Amazon EC2, and start building your virtual machine template, and one day you found out GoGrid offers you a better deal for your machine renting. You can't change from Amazon template to GoGrid template easily, as they are different. Even bare metal Xen template and VMWare template are not compatible. What it means is that, monopoly behavior is dominating the IaaS offering, as you can't easily change your vendor, they can charge you as mush as they want. I have a long conversation to Amazon Asia Pacific Chief Scientist three month back in one of the international conference, and asked him about how many VMs I can start freely at one time. The answer had been a big surprise, less than 1000 VMs at most. That is very small in terms of large scale I have to say. The big problem is this, there is no infinite resources in Cloud (as people always say, yes, we have infinite resources from Cloud), that is simply not true. Your vendor has its capacity constraint as well. What has been a real constrain is, because of lack of standards, vendors cannot trade resources each other, so they cann't offer you much more than they have. That is the next step to achieve an even bigger scale for Cloud: standard and more flexible market model for resource trading. As someone may notice, lack of standard so far has made some company very popular. An example is RightScale which offers a way to deploy your VMs on different vendors with their meta-template. They will translate the metadata and build different templates for different vendors.

The good news is Google and a lot of other big players have been working on Open Cloud, which enables Cloud be open, and everyone can easily migrate the Cloud apps from one place to another, from SalesForce to Google, and from Google to maybe Microsoft. How does that sound, must be very good.

飞飞鱼

根据Gens, F. 2009, New IDC IT Cloud Services Survey: Top Benefits and Challenges, paper presented to IDC，15 December, 2009统计调查，云计算的top three challenges是security, availability and performance。这份报告里有一个很清楚的比较图表，可以看下。

泛泛而言，我觉得云计算的Challenges主要有如下几个：
1.Security
Google的事故大家都知道了，就不多说了;
CSA (Cloud Security Alliance)列出了13个Cloud Security Domains。感兴趣的朋友可以参考http://www.cloudsecurityalliance.org/guidance/csaguide.v2.1.pdf

2.Dependability (Availability and Performance)
这个可以从三个方面谈
Network Dependence。呵呵，NBN对云计算还是有好处的，顶姐辣的;
Disaster Recovery。我目前还没有看到拿的上台面的Public Cloud的DR方案；Private Cloud另外算。
SLA。呵呵，哪位有Public Cloud服务商的SLA可以拿出来分享下，基本就没有吧，便宜，但自求多福。

3.Standards and Supplier lock-in
这个是各大厂商惯用的伎俩了。不过还是有几条和咱们澳洲有关系的消息的，Telstra和CBA加入了Enterprise Cloud Buyers Council，参与云计算的标准制定。

4.Legal concerns
这个是很明显的问题，当你的服务要全球化的时候，有很多法律上的纠葛得折腾清楚；
特别是Privacy的问题，米国和欧盟的标准是不一样的。

5.Cost
Smart IT executives might ask,” What if my end users, enabled by the
cloud model’s self-service provisioning capabilities, use more than I have budgeted for?”

飞飞鱼

原帖由 flyspirit 于 2010-8-26 10:05 发表
飞飞鱼也做cloud computing的？

  我是混的.....

superblue

各位搞app的我是看不太懂，我是弄cloud computing网络这边的，我感觉cloud没什么稀奇，就看你把server放那里，云里面可以动态分配资源，存储，一切的一切都是建筑一个超级高速的网络上，如果没有10G的网络在里面撑着根本就是不可能的。

这几年比较火是因为网络速度的提升把这些都变得可能了，还有以前SAN存储现在可以FCOE了，价格也下来了。

bulaohu

是没什么稀奇的，心里要坚信这一点，然后才能扛得住忽悠

不过这确实是目前这一波evolution的方向。个人机以及小型内部网络面对cloud的优势基本上丧失殆尽，一切computing转移到cloud中去是大势所趋。但迟早有一天，当存储和CPU power极大地小型化以后，personal computing的方便和隐秘会重新占上风的

superblue

几年以后大家人手一智能手机，到了动作地点把手机dock进去，然后出现全息屏幕和投影键盘，所有的东西都在cloud里面，呵呵。

应该不是太远了。

Bessy

Security，privacy  是比较大的问题。提到把数据放到数据中心(IDC)里，每个人都摇头。
个人感觉。像IBM，微软这样的公司一直推classter, High Performance Computing 和 grid computing. 30多年了也没多少大客户租用IDC。这是正赶上web service等的技术兴起。赶快加几个新概念，再让客户租用IDC. 看看当年他们是怎么夸奖grid computing的强大就知道了。现在搞cloud的基本都是当年做High Performance Computing 和 grid computing那帮人，以前叫做“服务器端”，现在叫做“云端”业务。

[ 本帖最后由 Bessy 于 2010-8-28 22:46 编辑 ]

bulaohu

那时候的网速，datacenter放在公司以外根本就不现实。现在不同了

另外互联网的通讯安全基本上就靠TLS了，这玩意儿也是最近几年才成熟的

qunlbao

就是分布式计算,把大型机的集中计算任务分散到小机器上,可以异步.
别搞得太复杂了,很早就有人提出来了,一些科学探索,天文软件采用让用户下载帮助计算.

Bessy

TLS是1999年的东西。他是处理信息传输安全的。

数据存储在云端如何保证不被别人，添加，修改，删除，浏览，传播。去年google 要离开中国就是因为云端数据（gmail）被浏览了。虽然google 找一堆借口，但是无法补偿客户的损失。信息只要放到云端，就基本没有秘密可言了。
这个与钱存在银行不同，钱少了银行可以赔，钱多了偷着乐就行了，钱被别人看多少次都没问题。

bulaohu

1999年的是TLS 1.0，现在真正在用的是1.2，2008年的版本

Google的Gmail问题不是传输过程中被解密的。整个cloud computing之所以能成为热点，就是因为网络传输的安全问题基本解决了。至于客户端还有cloud端本身的安全问题以及cloud提供商的信誉问题，那是另一码事了

ericlgq

大家还都传说WINDOWS有后门呢，各国政府还不是都在用？

starchu

原帖由 qunlbao 于 2010-8-28 22:59 发表
就是分布式计算,把大型机的集中计算任务分散到小机器上,可以异步.
别搞得太复杂了,很早就有人提出来了,一些科学探索,天文软件采用让用户下载帮助计算. ...

They all belong to distributed computing, however, they aim at different types of applications. The focus of Grid computing is scientific applications which are normally CPU intensive or data intensive. That is so called HPC apps. The focus of Cloud computing is normal business apps.

kawara

每隔几年那些寡头公司都要整出个东西糊弄人

flyspirit

云计算有一个特点就是对CFO们很有吸引力，以前不管是LOCAL的DATA CENTER还是IDC主机托管，都需要大量的Capital investment, 最后都是固定资产。而且投入多少初始固定资产很难预料，可能投入不够，或者太多，或者有时候用得多有时候用的少。

云计算很好地解决这个问题，对CEO/CFO们很有吸引力。技术上没什么新东西，但换个方法包装和吆喝，效果大不相同。

Bessy

原帖由 flyspirit 于 2010-8-29 18:48 发表
云计算有一个特点就是对CFO们很有吸引力，以前不管是LOCAL的DATA CENTER还是IDC主机托管，都需要大量的Capital investment, 最后都是固定资产。而且投入多少初始固定资产很难预料，可能投入不够，或者太多，或者有时候用得多有 ...

IDC很久以前就可以租用带宽，主机和空间了。用的时候签个合同就好了。

Bessy

原帖由 starchu 于 2010-8-29 13:54 发表


They all belong to distributed computing, however, they aim at different types of applications. The focus of Grid computing is scientific applications which are normally CPU intensive or data intens ...

HPC大部分是做科学计算的。 Grid computing (http://en.wikipedia.org/wiki/Grid_computing) 是一些“super virtual computer”上面安装 middleware，  middleware上面跑各种应用。各个层和应用是低耦合， 异构的。
cloud computing 把
Grid computing 的 “super virtual computer”叫"Cloud infrastructure"。
把grid middleware 叫"Cloud platforms".
把grid上的software-as-a-service叫做 Cloud software-as-a-service
把grid client 叫 cloud client

starchu

原帖由 Bessy 于 2010-8-29 21:17 发表

HPC大部分是做科学计算的。 Grid computing (http://en.wikipedia.org/wiki/Grid_computing) 是一些“super virtual computer”上面安装 middleware，  middleware上面跑各种应用。各个层和应用是低耦合， 异构的。
clo ...

wiki does not tell you the truth, I have been working with Grid middleware and applications for five years. All the major benefits you get from Grid is scientific applications, that is where the Grid original comes from. It is built on top of collaboration, different organisations have sort of agreement, and install some sort of Grid middleware to form a Grid. The most difficult part is the over-complexity of the middleware makes it very hard to use and manage, and administration constrains crossing organizations make the work even harder, a lot of applications could not utilise full resources as local organisation policies don't allow you run certain software.

Bessy

原帖由 starchu 于 2010-8-30 09:24 发表


wiki does not tell you the truth, I have been working with Grid middleware and applications for five years. All the major benefits you get from Grid is scientific applications, that is where the Gri ...

Wiki的定义还是基本准确的。 你们应该是做的是Science Grids。 除了Science Grids还有
Compute Grids, Enterprise Grid, Data Grids, Access Grids, Knowledge Grids, Bio Grids,
Sensor Grids, Cluster Grids, Campus Grids, Tera Grids, and Commodity Grids......
下面是HP的 grid, 他们主要面对企业客户的
http://h30423.www3.hp.com/index. ... 18c2dd0a5ec79dc7d6f
AC3是悉尼的一个数据中心,用起来比我的PC快太多了。
http://www.ac3.com.au/home.html
你要不是做这方面研究的不需要了解太深。这东西几年一更新。靠制造这种新概念吃饭的就没办法了。

Bessy

5年后应该提出来“Cloud is a service” 了， 那时候叫做 “sky（or welkin） computing”。

dalaohu

starchu

原帖由 Bessy 于 2010-8-30 11:41 发表
5年后应该提出来“Cloud is a service” 了， 那时候叫做 “sky（or welkin） computing”。

sky computing is there already, there is a keynote in CCGRID 2010 melbourne, talking about sky computing, also known as InterCloud.

starchu

原帖由 Bessy 于 2010-8-30 11:39 发表

Wiki的定义还是基本准确的。 你们应该是做的是Science Grids。 除了Science Grids还有
Compute Grids, Enterprise Grid, Data Grids, Access Grids, Knowledge Grids, Bio Grids,
Sensor Grids, Cluster Grids, Cam ...

we were working with all kinds of Grid because we wrote Grid middleware not applications. The natural of the Grid middleware has determined what kind of applications it can support. Compute and data intensive applications, that's all. I have been working with all kinds of Grid middlewares SGE, Globus, ProActive, gLite, GridSuperscaler, they all share the common features.