@lshero @Pan940425 又逛了下 TB 原来是有带外壳的 DIY 2k 显示器店铺的，之前没发现。。。
凑齐 2 家了，直接给网址吧，一家广东一家上海，期待小白鼠汇报测试结果~
https://tonylabs.taobao.com/
https://diylcd.taobao.com/

目前用的 tony 家的上一代板，质量 ok ，新的 air 板未测试。不过屏幕外壳没彻底解决， DP 线难固定容易使驱动板变形，而且屏幕如果加边框会漏光。。。

@lshero @Pan940425 TB:"retina 驱动"/"2k 驱动板"

NAT 端口转发
路由器多能桥接尽量设置桥接

才大一，有的是时间，全部试过就知道了。还有楼主的问题百度就好了。嗯，百度。

愿意折腾的话：
iPAD Air 屏幕 + 驱动板 ，成本各 300 以内
rMBP 屏幕 + 驱动板, 成本各 1k 上下
缺点：需 DIY 外壳

@flaminigo Thanks for your detailed explanation, which've taught me more about super computer~

to all ：
搜到 2 份很好的天河 2 文档，有兴趣的 V 友可以看看 ：
官方： https://www.openstack.org/assets/presentation-media/How-to-Deploy-OpenStack-on-Tianhe-2-Supercomputer-2.pdf
非官方： http://www.netlib.org/utk/people/JackDongarra/PAPERS/tianhe-2-dongarra-report.pdf

另外附上 openstack 的官方架构介绍，好复杂。。。
http://docs.openstack.org/juno/install-guide/install/apt/content/ch_overview.html

@ryd994 ssd+机械硬盘混合的机器，我只需要对机械硬盘的某文件夹或分区做加速， os 的 vm 管理无法对特定文件 /分区做加速吧？

@BXIA 查了下应该全部都是 2692v2 。 16,000 台双路 server 组成的 OpenStack 私有云。之前一直以为天河 2 是一台超多路的大型机>.<

参考： https://zh.wikipedia.org/wiki/%E5%A4%A9%E6%B2%B3%E4%BA%8C%E5%8F%B7
天河二號共有 16,000 個運算節點，每節點配備兩顆 Xeon E5 12 核心的中央处理器、三个 Xeon Phi 57 核心的协处理器（运算加速卡）。累计 32,000 颗 Xeon E5 主处理器和 48,000 个 Xeon Phi 协处理器，共 312 万个计算核心。
中央處理器为英特爾提供的，運作時脈為 2.2GHz 的 Xeon E5-2692v2 12 核心處理器，基於英特爾 Ivy Bridge 微架構（ Ivy Bridge-EX 核心），採用 22 奈米製程，峰值效能 0.2112TFLOPS 。

在 UbuntuOpenStack 上運作的雲端運算系統、 Ubuntu Juju 雲端服務排程引擎以及基於 Ubuntu Server 為中國客制的 Kylin Cloud Linux 作業系統。在 2014 年 5 月已分配 256 個節點運行，未來將擴展至超過 6400 個節點。

@BXIA 哇，感谢，居然算了 10b 。。。
不过好奇怪手册里的 CPU 是 4640 ，你贴的是 2692 ，难道同一台机器 CPU 都混着用的？还是其实天河 2 是个多台不同配置服务器组成的[虚拟化?]集群？ 顺便 @flaminigo 大虾~~

搜了下 dmesg 原始信息被重写了。。。

> I wanted to see my original bootup messages again. So I typed
> dmesg, but all I got were 'arplookup' lines.
>
> I'm puzzled, where the bootup messages have gone?

Overwritten by kernel debug messages. There have been a number of
similar questions for a while, the gist is that when the dmesg buffer
gets filled up, it starts to overwrite itself. This affects the
daily periodic output and makes it look munged. I don't think any
solution has ever been offered up - I'd certainly be interested in it :P

ref: http://lists.freebsd.org/pipermail/freebsd-questions/2004-April/042612.html

@BXIA 收到，不过好奇怪的 dmesg 。。。貌似不包含启动信息。。。 有没 linux 高手大虾帮忙解答下~~

前 20 行：

00000000000,600000002f0e0000,000001c7b4402000,0000008013bb4400,
flit[01]= 0x0000000000000000,004500080b4402fc,fcfc0b4400fcfcfc,a08000011c7b4402,
flit[02]= 0x0000000000000000,38262da900000046,0b0c02460b0cb8ee,01400000d8e75400,
flit[03]= 0x0000000000000000,1514131211100000,00000006c72f0000,0000561ca61f0400,
flit[04]= 0x0000000000000000,2d2c2b2a29282726,2524232221201f1e,1d1c1b1a19181716,
flit[05]= 0x0000000000000000,ffff88189c463800,ffffffffa0223736,3534333231302f2e,
flit[06]= 0x0000000000000000,0000248bb6920000,0000000000000000,00030062810152c3,
flit[07]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[08]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[09]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[10]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[11]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[12]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[13]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[14]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
flit[15]= 0x0000000000000000,0000000000000000,0000000000000000,0000000000000000,
nio_dev 0000:04:00.0: _intr, new FPQ packet:
[LLP] - (MISS_ROUTE)
[PKT] - op=0x1(MP_REQ), src=0xb4402, dest=0xb4400
flit[00]= 0x0000000000000000,600000002f0e0000,000001c7b4402000,0000008013bb4400,

@BXIA 所以“> boot.log ”导出到文件~ 文件大么？看了下我的 400 行也才 28k 。
可以传个网盘？或者邮件（我的 V2 用户名）@gmail 我帮忙上传也行哈~~ 好奇这么牛的机器的 dmesg 啥样~~

@BXIA

发个 dmesg > boot.log 如何？

phoronix 确实略复杂。。。要不算个 pi 吧~~ 虽然 22 楼指出跑分没啥意义

http://www.numberworld.org/y-cruncher/#Download

<10M, 不用编译，单线程和多线程范例：（相同位数重算前次结果会被覆盖需保存）

y-cruncher bench 25m -PF:none -TD:1
y-cruncher bench 1b -TD:64

{program} bench {size} [-PF:{framework}] [-TD:{tasks}]

The valid values for {size} are:
25m, 50m, 100m, 250m, 500m,
1b, 2.5b, 5b, 10b, 25b, 50b, 100b, 250b, 500b,
1t, 2.5t

Common Parameters:

These parameters are used by multiple components in y-cruncher.

Parallel Framework:

-PF:{framework}
(Examples: "-PF:none", "-PF:cppasync")

Use the specified parallel computing framework. The valid values are:
-PF:none Disable multi-threading and sequentialize all tasks.
-PF:spawn Spawn a thread for every task.
-PF:cppasync Use C++11 Async.
-PF:winpool Use the Windows Thread Pool. (only available on Windows)
-PF:cilk Use Cilk Plus work-stealing. (only available on Windows for AVX and AVX2)


Task Decomposition:

-TD:{tasks}
(Examples: "-TD:8", "-TD:12")

Run the operation using the desired level of task decomposition.
A larger number enables more parallelism, but at the cost of more
computational and memory overhead. For most systems, setting this equal to
the number of logical cores is ideal. But forcing it higher may reduce
load-imbalance.

Note that setting this value to 1 does not disable multi-threading since
some subroutines ignore the parameter and will decompose anyway. To get a
correct single-threaded benchmark, you need to also set the parallel
framework to "none".

Disable Multi-threading: -PF:none -TD:1

AWS T2.micro 上的测试结果：
http://i.imgur.com/9lGInOK.png

下面 2 个测试能帮忙挑个跑下么~~

http://www.numberworld.org/y-cruncher/#Download
http://www.phoronix-test-suite.com/

@cxbig
@manhere

表示很常见的需求，特别很多 dz 论坛：
http://i.imgur.com/PLXkTa7.png

上面链接貌似贴反了。。 发现 FT 多次访问后要注册。。。

@loading 也许吧。。。

@xfspace 补短肯定的，不过好贵啊。。。

@Yvette 你的 link 是 14 年的，最新的如下，确切地说应该是 DELL 买了 VMWARE 80%的股票：

http://www.ft.com/fastft/405961/what-know-about-dell-emc-pact

EMC "owns an 80 per cent stake in VMware that is worth some $27bn"

http://www.ft.com/cms/s/0/142f5190-6ddb-11e5-8171-ba1968cf791a.html#ixzz3oMubnmrh

"The buyers will pay $24.05 a share in cash, and 0.111 tracking shares for each share. The tracking shares will move in-line with VMWare's stock price."

"VMWare will remain public after the transaction."

@manhere 开始有这需求因为常逛的某图片量很大的论坛限制了上传图片大小，就想找这样的软件（也可以用外链，不过也麻烦）

@jukka 普通的 jpg/png 这些

@xenme 貌似真的要研究就是存数学问题了，压缩算法，矩阵。。。无爱。。。

@lekai63 嗯，今天办个证发现某 ZF 网站也限制图片大小，只能传 80k 。。。
http://218.85.72.204:1234/fjphoto/index.jsp

@shiny 主要是我用 PHP 一般都是调用或改现成的>.<

借用一段 Q&A 回复这个问题吧，来源： http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4172.htm

Objection #1: This feature caters to having functions with many parameters - a programming style that should be discouraged.

We agree that having functions with many parameters should be discouraged. However, the reality is that many legacy APIs that programmers have to work with and will have to work with for a long time, have functions with many arguments, and making it easier to deal with such functions would solve a real problem and be materially useful.

More importantly, however, this feature has a lot of value for functions with few arguments, too. Even for functions with few arguments, when reading a call site one has relatively little information about the roles of the arguments. Having names of arguments present would make call sites more readable, regardless of the number of arguments.