Talk:Virtual server: Difference between revisions
imported>Howard C. Berkowitz (→Cloud computing: new section) |
imported>Pat Palmer (→references: more information on the issue) |
||
(13 intermediate revisions by 2 users not shown) | |||
Line 3: | Line 3: | ||
== Cloud computing == | == Cloud computing == | ||
While all cloud computing uses virtualization, not all virtual machines run remotely. Nevertheless, you may find [[cloud computing]], [[Software as a Service]], [[Platform as a Service]], and [[Infrastructure as a Service]] useful. [[User:Howard C. Berkowitz|Howard C. Berkowitz]] 21:32, 7 August 2010 (UTC) | While all cloud computing uses [[virtualization]], not all virtual machines run remotely. Nevertheless, you may find [[cloud computing]], [[Software as a Service]], [[Platform as a Service]], and [[Infrastructure as a Service]] useful. [[User:Howard C. Berkowitz|Howard C. Berkowitz]] 21:32, 7 August 2010 (UTC) | ||
== Economics == | |||
Some of the points I raise are rules of thumb from experience, while others, with a bit of effort, probably can be found in public presentations. | |||
The industry rule of thumb is that in general enterprise computing, servers dedicated to applications, or a carefully selected, set of applications run about 20% CPU utilization. There are some queueing theory (Little's Law, etc.) observations that suggest it can be unwise to load them more heavily than 50-70%, but that 30% wastage is a major economic target. Remember that the virtualization management (i.e., provisioning and such, not the hypervisor) will take some, and that some applications are memory, not CPU bound. I myself wonder if we know if these rules will stay the same with multicore processors. | |||
There might be a cite for it in cloud computing -- it was at USENIX or LISA -- where a Google person said they were fast approaching a point at which power and cooling would cost more than the hardware. Intel was told, at one developer conference, that power consumption was, to many enterprises, more important than raw speed. | |||
When looking at cloud computing, and, for that matter, footprint, do look at [[blade server]] and related techniques. With machines designed for efficient cooling, especially chilled water, cooling becomes much more efficient. Indeed, there are cities (e.g., Cambridge MA, and I think Miami FL) that offer chilled water as a municipal utility -- there are enormous economies of scale, but you need quite a bit of cooling requirement to justify your own chiller plant There was a talk at NANOG on the Miami [[internet exchange point]] dealing with the economic effects of water cooling. Google is reputed to have moved a number of its data centers out of California to put them in cooler areas. [[User:Howard C. Berkowitz|Howard C. Berkowitz]] 18:10, 13 August 2010 (UTC) | |||
== Pat's review of this article == | |||
I learned a lot from this detailed and thoughtful beginning, and I love the graphics. In the following subsections, I am recording details notes about things I noticed in the version available at the end of the course (to be added shortly):[[User:Pat Palmer|Pat Palmer]] 00:20, 20 August 2010 (UTC) | |||
=== references === | |||
Despite the decent reference list, the article suffers throughout by citing surveys and mentioned recent statements and then not providing the reader with a reference so we can know ''what survey'' or ''who claimed a thing''. This article uses references, but inappropriately, in particular by failing to offer a reference when it is most needed and expected. The beginning paragraph of the "Virtual Server Market Overview" section has a couple of glaring examples (the phrases to note are "In a recent survey" and "Financial estimates"--whose?). '''Unless the references can be found, this information will likely have to be removed from the article.''' There are many other examples in other sections; many statements in the market part of the article are not supported by references.[[User:Pat Palmer|Pat Palmer]] 00:22, 20 August 2010 (UTC) | |||
=== introduction === | |||
The intro would be even better if it provided a time frame for when this technology "took off". The introduction also might as well admit that this article is about ''operating system'' virtual machines. Also, the section called "Future" could probably be merged into the introduction.[[User:Pat Palmer|Pat Palmer]] 00:25, 20 August 2010 (UTC) | |||
=== please define terms when first used === | |||
"Hypervisor" (and also "CMS") are used a few times in the article before its meaning is clarified. IMO, it's better to substitute some words that are clear to an intelligent layman, and avoid using "new" terminology until the article is prepared to introduce its definition. The term "server" (as used here) is a potentially ambiguous term; as used here, it might more specifically be called TCP applications (software servers for FTP, HTTP, SMTP and the like).[[User:Pat Palmer|Pat Palmer]] 00:30, 20 August 2010 (UTC) | |||
:Or end-to-end layer applications? Consider clustered DNS, DHCP, TFTP, etc. infrastructure servers that run over UDP, not TCP. [[User:Howard C. Berkowitz|Howard C. Berkowitz]] 00:37, 20 August 2010 (UTC) |
Latest revision as of 18:39, 19 August 2010
Cloud computing
While all cloud computing uses virtualization, not all virtual machines run remotely. Nevertheless, you may find cloud computing, Software as a Service, Platform as a Service, and Infrastructure as a Service useful. Howard C. Berkowitz 21:32, 7 August 2010 (UTC)
Economics
Some of the points I raise are rules of thumb from experience, while others, with a bit of effort, probably can be found in public presentations.
The industry rule of thumb is that in general enterprise computing, servers dedicated to applications, or a carefully selected, set of applications run about 20% CPU utilization. There are some queueing theory (Little's Law, etc.) observations that suggest it can be unwise to load them more heavily than 50-70%, but that 30% wastage is a major economic target. Remember that the virtualization management (i.e., provisioning and such, not the hypervisor) will take some, and that some applications are memory, not CPU bound. I myself wonder if we know if these rules will stay the same with multicore processors.
There might be a cite for it in cloud computing -- it was at USENIX or LISA -- where a Google person said they were fast approaching a point at which power and cooling would cost more than the hardware. Intel was told, at one developer conference, that power consumption was, to many enterprises, more important than raw speed.
When looking at cloud computing, and, for that matter, footprint, do look at blade server and related techniques. With machines designed for efficient cooling, especially chilled water, cooling becomes much more efficient. Indeed, there are cities (e.g., Cambridge MA, and I think Miami FL) that offer chilled water as a municipal utility -- there are enormous economies of scale, but you need quite a bit of cooling requirement to justify your own chiller plant There was a talk at NANOG on the Miami internet exchange point dealing with the economic effects of water cooling. Google is reputed to have moved a number of its data centers out of California to put them in cooler areas. Howard C. Berkowitz 18:10, 13 August 2010 (UTC)
Pat's review of this article
I learned a lot from this detailed and thoughtful beginning, and I love the graphics. In the following subsections, I am recording details notes about things I noticed in the version available at the end of the course (to be added shortly):Pat Palmer 00:20, 20 August 2010 (UTC)
references
Despite the decent reference list, the article suffers throughout by citing surveys and mentioned recent statements and then not providing the reader with a reference so we can know what survey or who claimed a thing. This article uses references, but inappropriately, in particular by failing to offer a reference when it is most needed and expected. The beginning paragraph of the "Virtual Server Market Overview" section has a couple of glaring examples (the phrases to note are "In a recent survey" and "Financial estimates"--whose?). Unless the references can be found, this information will likely have to be removed from the article. There are many other examples in other sections; many statements in the market part of the article are not supported by references.Pat Palmer 00:22, 20 August 2010 (UTC)
introduction
The intro would be even better if it provided a time frame for when this technology "took off". The introduction also might as well admit that this article is about operating system virtual machines. Also, the section called "Future" could probably be merged into the introduction.Pat Palmer 00:25, 20 August 2010 (UTC)
please define terms when first used
"Hypervisor" (and also "CMS") are used a few times in the article before its meaning is clarified. IMO, it's better to substitute some words that are clear to an intelligent layman, and avoid using "new" terminology until the article is prepared to introduce its definition. The term "server" (as used here) is a potentially ambiguous term; as used here, it might more specifically be called TCP applications (software servers for FTP, HTTP, SMTP and the like).Pat Palmer 00:30, 20 August 2010 (UTC)
- Or end-to-end layer applications? Consider clustered DNS, DHCP, TFTP, etc. infrastructure servers that run over UDP, not TCP. Howard C. Berkowitz 00:37, 20 August 2010 (UTC)