Category: Network & Server (網絡及服務器)

Finally A New Breed of Entry SAN from Dell !

By admin, May 8, 2015 1:50 pm

DELLSCv2020

After reading several SCv2000 reviews, I found it’s simply an upgraded version of PowerVault, so I seriously doubt the performance is even near Equallogic PS4100 series.

Quote from StorageReview:

The Dell Storage SCv2000 Series is a customer-inspired series of entry-level storage arrays that offer same common management and several of the core features as higher-end Dell SC Series arrays, such as the SC4020. According to Dell, the SCv2000 Series offers the “best performance and protection in its class with integrated data protection to support specialized projects, database and test environments, or simple storage consolidation.” The new arrays enable customers to standardize on a common platform, saving time and money. The SCv2000 Series comes in three models with expansion options: the SCv2000, SCv2020, and SCv2080

Benefits include:

  • High performance at an affordable price – The arrays deliver best-in-class performance in a single, affordable 2U enclosure. Feature-rich options include proven data protection features, RAID tiering to optimize capacity, thin provisioning, flash support, data-migration services, and multi-protocol connectivity. Granular data protection allows up to 2,000 snapshots and 500 replications.
  • Integrated data protection- Remote Instant Replay, Local Instant Replays and Replay Manager along with RAID tiering that optimizes SAN capacity. Tight integration with common application environments, such as Microsoft and VMware, help to simplify virtualized data centers, allowing local and remote data protection features to take consistent snapshots of virtual machines without sacrificing performance.
  • Future flexibility with Storage Center – Customers can gain more value from existing investments with data migration to higher-tier Dell Storage SC4000 and SC8000 Series arrays. The SCv2000 Series supports today’s needs and future growth as the arrays can be managed with the same, single user interface as other SC Series arrays. Customers can scale capacity when needed with the benefit of no capacity licensing and choose from a broad range of hard drives and SSDs, configured for a variety of applications.

SCv2000 Series specifications:

  • Models: SCv2000 | SCv2020 | SCv2080
  • Internal Storage: 12 x 3.5” drive bays | 24 x 2.5” drive bays | 84 x 2.5” or 3.5” drive bays
  • Supported expansion enclosures:
    • SC100: 12 x 3.5” or 2.5” drive bays | SC120: 24 x 2.5” drive bays | SC180: 84 x 2.5” or 3.5” drive bays
    • SC100: 12 x 3.5” or 2.5” drive bays | SC120: 24 x 2.5” drive bays
  • Maximum drive count:
    • 168 (12 internal, plus 156 external) |168 (24 internal, plus 144 external) | 168 (84 internal, plus 84 external)
  • Total storage capacity: 504TB
  • Supported drive types:
    • HDD: 7.2K RPM | 15K, 10K, 7.2K RPM | 10K, 7.2K RPM
    • SSD: write-intensive, read- intensive (different drive types, transfer rates and rotational speeds can be mixed in the same system)
  • Controllers: Single or dual controllers | Single or dual controllers | Dual controllers
  • Processor: Intel Xeon 4-core
  • Memory: 8GB per controller
  • Network/server connectivity (front-end):
    • 2 x 16Gb FC ports per controller
    • 4 x 8Gb FC ports per controller
    • 4 x 1Gb iSCSI (BASE-T) ports per controller
    • 2 x 10Gb iSCSI (BASE-T) ports per controller
    • 4 x 12Gb SAS ports per controller
  • Internal drive connectivity (back-end): 6Gb SAS ports (2 per controller)
  • Product OS: Storage Center 6.6.2
  • Server OS:
    • Microsoft Windows Server
    • SLES
    • VMware
    • Citrix XenServer
    • Red Hat
  • RAID: Supports RAID 5, 6, RAID 10 and RAID 10 DM (dual mirror)

Third Equallogic Disk Failure

By admin, April 30, 2015 11:38 am

This is the 3rd time a 15K SAS failed, not bad actually giving the long time period of 4 years and 6 months (The first was in December 12, 2012 and 2nd was in Jan 11, 2014), and this time the reconstruction took approximately 4 hours to complete.

12321312

First Encounter of HA Failover

By admin, April 29, 2015 5:12 pm

Found one of the hosts performed HA Failover on its own at midnight today and all its associated VMs were restarted on the other hosts automatically. The exact cause of host restart is still under investigation as there is no hardware failure found at all.

12:17 Host is Not responding to vCenter Ping, so it triggered HA Failover. In fact, vCenter took about 50 seconds to realize the actual host failure.

12:24 It took 7 Minutes for the host to re-join vCenter cluster. In fact the host & ESX restart only took about 5 minutes counting from the boot screen.

12:28 Everything went back to green, the whole process took about 10 minutes. For individual VM, there was 5-6 minutes of downtime unfortunately.

Cluster Level Event Log

cluster

Host Level Event Log

host

有意思! 再用Power,你就Low了 (轉文)

By admin, April 28, 2015 10:23 am

要说这位嘉宾的来头可不小,他是VMware大中华区高级业务拓展经理, 新一代关键业务云平台解决方案总监王俊标。来自VMware的资料显示——在加入VMware之前,他曾在中国银行,IBM 中国工作。是IBM灾备红皮书作者之一(1999年,美国奥斯汀)。拥有丰富的IT整体架构,数据中心规划经验,熟悉金融,电信,政府等行业。

话不多说,摘抄了演讲中比较有意思的几个梗,分享给大家。

VMware的由来

在1998年的一天,IBM有个项目经理说,目前Intel的计算性能足够好,很多CPU的时间空着,我们看看能不能把空余的时间用起来。他就创建了个项目,就请了伯克利大学一个教授叫做Mendel Rosenblum带着两个研究生,开始研究在大机上的虚机技术能不能够在Intel这样当时所谓低端的服务器上创建虚机。教授和研究生非常了不起把这个项目做出来了,虚拟花费半个小时启动了,项目经理验收通过了。沃森,很好的项目,付钱,抽屉里一放。然后就没然后了。

这个教授很郁闷,回去跟他太太讲,他太太叫Diane Greene。Diane Greene和一个中国人叫做Edward Wang正在找项目做。“我们就把这个项目做起来吧”,在1998年他们在一个有缝隙的房子里创建了VMware公司。

IBM倒贴15亿美元卖掉芯片业务

没想到有一天将这么厉害的芯片卖给了一家叫做GlobalFoundry的公司,我专门去查了一下这家公司,原来它是AMD分拆出去一家专门做代工的公司。我就挠头,这件事情基本就是个信号告诉客户,P可能接下去是很难,这件事情意味着很多很多。

一语道破12306危机本质

在互联网冲击下,现在用户量增长这么快,碰到了结构僵化,没有办法扩展,灵活性太差,还很贵,还没有未来。原来是Power8、Power9,按照一年半就出来,现在说没有未来。面对互联网的压力,原来跑得好好的应用,跑了这么多年的应用,在主机上撑不住了。很多客户告诉我,突然间走向的公共服务,铁道部在12306面向互联网卖票,这个架构一定是因为本身不是这样设计的,甚至是基于大机主机的应用架构也碰到这样的问题,因为当接入互联网的时候,它的业务增长已经是无法预测,不是5倍、10倍,可能是100倍,甚至于1000倍,铁道部要安12306,原来全国可能算的一个5000个网点排队在那里,不会有超过5000个并发,因为就是那几个人在那么卖票,最多是买票的人辛苦一点,可是基于互联网以后第一次有20亿的点击,所以这个系统会撑不住”。

跟POWER比,我都不好意思

说虚拟化和性能下降太厉害,这件事情反复讲,再三讲,这件事情是怎么理解,性能会损耗,但是物理x86和虚拟x86会损耗6%左右。现在拿先进的、快的、便宜的至强的x86芯片跟老的贵的POWER5、POWER6去比,每次比都不好意思,每次随便你跑Oracle DB2,都比你块5倍、10倍、100倍,你老说性能差,差在哪里,怎么差?这是一个误区。

我们是文明的拆迁办

说迁移太影响我现有的系统要花很多努力,这也是不对的。因为我们是文明的拆迁办,我经常是这样说,是文明的给你房子,但是不拆迁。认为我的新平台搭好之后,搭好的那一天就把老系统给拆掉了,没有这回事。老系统继续跑三年、五年是可以的,但是新平台搭好,你跑。有一个四川电信2009年开始跑到2011年,终于得出来一个结论,虚拟化是信得过的,x86是可靠的,甚至比Power更可靠。接下来两年之类把所有的贵的POWER全部搬走,全部跑x86。没有说搭那一天,老房子就不给你住了,不存在这种问题,所以根本不存在搬迁不搬迁的工作,全部东西都是新的,Oracle都是Linux的,所以没有这回事。

回應: VM家经理别闹了,人家POWER就是要low到底

据说前天有个VMware家的经理(就是食人族最喜欢吃的那种经理)在一个发布会上呐喊:再用POWER你就low了!这位经理童鞋的例证有三个,都很有趣,但不是那种正确的、有内涵的有趣,而是破绽百出、明眼人一看就受不了的那种有趣,这真不是我说话刻薄,接着往下看就知道我这么说着实还挺厚道的。

关于IBM倒贴15亿美元卖掉芯片业务

经理说,“没想到有一天将这么厉害的芯片卖给了一家叫GlobalFoundry的公司,我专门去查了一下,原来它是AMD分拆出去一家专门做代工的公司。”

经理就挠头了,得出结论是POWER就没有未来了。

别闹了,你都知道POWER是“这么厉害的芯片”,一个代工厂还去接这个芯片业务,人家GlobalFoundry活腻啦?明明出售的是“芯片制造业务”,经理非要把这个“制造”二字给忽略,我必须严肃地说这个错误犯得太low了!人家耐克没有制造工厂,按您这个说法耐克许多年前就没有未来,应该死翘翘了呀?

人家IBM说投资30亿做7nm工艺的芯片、神经元芯片、碳纳米管芯片等,这个计划可从来没有变化过,切记!

关于POWER很贵、没未来

经理说:“(POWER)结构僵化,没有办法扩展,灵活性太差,还很贵,还没有未来,原来是POWER8、POWER9一年半就出来,现在说没有未来。”

别闹了,跳票可从来不是IBM的范儿,从2004年的POWER5到2013年的POWER8,这一路从未有过啥意外,不知道这“一年半”的说法,经理是从哪里听来的,真的属于很low的谎言。
而且,恐怕经理还得打自己的脸两下,人家POWER9还是会准时出现在大家面前——在11月18日的第44届全球超级计算机TOP 500公布时候,透露了一些关于POWER9的技术细节,说进一步强化加速器,也将是第一个可通过NVLink总线互联的处理器架构,带宽高达80-200GB/s,而且延迟和功耗都将进一步降低等等——这话可不像经理一样乱说的,人家POWER的未来已经到来,您就别瞎放炮了。

不过,经理说的有一点是歪打正着——POWER真的low了,而且一门心思想要low到底,例如有个OpenPOWER基金会您造吗?就是去年8月IBM联合Google、NVIDA等5家公司成立的专门基于POWER平台进行开放创新的组织,到10月底已经有65家成员了,就是为了让POWER能够low到完全覆盖x86服务器市场。

你看这一Open,入门的门槛确实low了,用户选择也多了,以后POWER服务器可不再是IBM一家的,会有很多供应商出货Power服务器,比如泰安的Power服务器上个月已经开始订货了。另外,IBM还跟北京市经信委和华胜天成成立合资公司,主要业务就是搞POWER服务器,很快还会有中兴、浪潮等推出POWER服务器。就连POWER芯片,也不是IBM一家有,远的不说,中晟宏芯的POWRE芯片可是快来了,具体产品路线图白纸黑字宣布了:第一代产品将于2015年12月推出,12核心,对标英特尔E5-2690;第二代产品将于2017年12月推出,集成自主的浮点运算单元,对标英特尔E5-2620/2650。

门槛low了的同时,POWER的价格也必然会low下去。对嘛,就是要Low到x86市场去,让用户以亲民价格继续享受小型机的RAS性能。再有拿贵说事儿的,先去询一下泰安的POWER服务器价格咱们再讨论论吧。

还有,如果经理硬要说x86比POWER可靠,那这就超出技术和业务讨论范围,只关乎情绪,就不好弄了。

关于性能:5倍10倍100倍

经理提到一个事实,说拿新的x86芯片跟老的POWER6比,“每次都不好意思,随便你跑Oracle、DB2,都快5倍、10倍、100倍,你老说性能差,差在哪里?”

这个事情,的确是心有花则见花、心有沙则见沙的事情,我想用这几天很热的一篇文章的标题来说“国产大飞机,你敢坐吗?”不用多说吧?还有,咱们的“天河二号”在最新的全球HPC TOP 100中又夺得冠军了,貌似三连冠了,你能说中国的HPC系统比美国的HPC系统先进吗?

再多说一句,别拿2014年比2010年啦,比2013年的POWER8吧——22nm,12内核,4GHz频率,内存带宽230GB/s,利用率高达65%,数据吞吐量是x86服务器的两倍,内置加密引擎技术,使关键应用和数据可以放心地放到云上。。。。。。嗯,POWER的确不全low,这些特性只能用“High”来描述。

想起一个前几天的段子——听说某省电信运营商把核心系统迁移到x86系统后,曾经吸引了很多人眼球,但现在常常因故障率过高被扣奖金。这不就是疤瘌眼照镜子嘛,你说low不low?冷静想想,世界上哪里有最好?只有合适不合适,客户了解自己和供应商,知己知彼,然后选择适合自己的技术才是最明智的,才会让行家说low。

关于两个破词儿:Scale up和Scale out

经理说,POWER没办法扩展。

别闹了,这都是陈年老调了,经理作为大V家的一员,不应该停留在历史中。POWER服务器早就既能纵向扩展Scale up也可能横向扩展Scale out了,既有scale up的大系统,也有scale out的小系统。随着POWER8的推出,在性能、RAS和性价比方面都大幅提升,例如4月份推出的首批5个基于POWER8的全新Power服务器,就是专门为大规模scale out计算环境设计的,一两个插槽这种事早就不是个事儿了。

今天的POWER,应用门槛儿的确是low了,例如支持Little Endian,所有的Linux应用可以直接跑在Power上;另外还支持OpenStack方案中最为广泛使用的操作系统Ubuntu,够屌丝够low。

另外,虚拟化成本也low了,因为支持KVM虚拟化了。这个可以说是点中了VMware的罩门,怪不得V家经理急了眼胡言乱语呢。

关于12306危机是主机造成的

经理说,“12306危机的本质是面对互联网的压力,原来跑的好好的应用,在主机上撑不住了”。

别闹了,您能稍微严肃点吗?我的意思是说您能举个稍微靠谱的、不让圈内人说你low的例子吗?照您这说法,12306换一堆x86服务器问题就解决了?业界讨论已经很多了,12306这是scale out和scale up都很难解决的问题,Alibaba技术团队不是雄赳赳气昂昂进驻后灰溜溜铩羽而归了吗?再说了,12306后端用的明明是RISC小型机,您好歹是大VMware家的经理,肿么跟那谁谁一样把小型机叫主机啊?唉,压根儿就不是这么个角度的问题,建议经理百度一下,网上有大量深度技术剖析文章说这个事情,我懒得说了。

关于可怕的分布式架构隐患,八卦一下那位马师爷的双十一,某业内人士评价说:“当然马云关掉退款功能不是为了冲业绩,500多亿打个折扣也是惊人的,但是分布式架构的CAP难以三全的问题是个技术框架的问题,没有万能的架构,分布式当然也不是。”好奇的童鞋可以自己去查原文。

跨時代的科技產物

By admin, February 1, 2015 7:33 pm

今天偶然找到的,很多年輕一輩根本不知道這是什麼來的 (就如Walkman卡帶機)。其實把機密資料保存在裡面然後再等100年,保證會成為全世界最安全的設備,因為到時根本沒有可讀取相關媒介的器材。 這令想起了美劇White Collar最後一季裡那間頂級安保庫也是這樣保存機密資料的。

IMG_8210

How to ByPass YouKu’s Mainland China Restriction

By admin, December 10, 2014 2:11 pm

The solution is quite classic, use a proxy!

Simply set “http://yo.uku.im/proxy.pac” in the proxy configuration URL in your browser and alibaba…everything works.

You can double check if your IP is within Mainland China afterwards by visiting

http://ipservice.163.com/isFromMainland

If it returns “True”, then it works.

Predictive Failure is Working

By admin, August 12, 2014 12:15 pm

Finger crossed as sometimes the raid rebuilt won’t happen automatically, mind you it’s 11th generation of PERC, can’t believe it’s still having such stupid behavior.

pre

How to fix SQL Server Orphaned User after Database Restore

By admin, July 30, 2014 10:05 pm

A simply way to link back the original user and fix all the permission problem after restoring MSSQL database to a different server using the .bak file.

Step 1:
exec sp_change_users_login ‘Report’
Orphaned users listed in the current database

Step 2:
exec sp_change_users_login ‘AUTO_FIX’, ‘username’
Can automatically log in to add a user name corresponding to the same name in syslogins

Equallogic Firmware Upgrade Checkup List

By admin, July 5, 2014 4:53 pm

I’ve received an alert this morning saying “Time-of-day clock battery charge is low” from my Equallogic array.

clock

Googled a bit and turned out to be a false alarm (this was also double verified with Dell Pro-Support with EQL Diag log), it only happens with older firmware prior to v5.2.10 (and currently I’m still using v5.2.2). The only way to dis-alarm such warning is to upgrade to the latest firmware release (ie, >v5.2.10).

However I do remember there are certain EQL components depend on specific firmware version. Luckily, I’ve found out this chart, it’s really a time-saver and it’s advised to check out this list every time you plan to do a major firmware upgrade.

For me, my MEM is v1.1.0, SANHQ is 2.2.0, EQL Virtual Storage Manager for VMware is v3.1.1, so my best and most safe play is upgrade to EQL firmware v5.2.11 which is the latest. If I upgrade to v6 or v7, then many related components may not work.

eql_chart
Of course this chart also shows the possible upgrade path:

upgrade_path

Extra Notes:

* Something special for latest MEM v1.2 only supports ESXi v5.0 or above, so for those of you still running ESX/ESXi 4.x, please stay with v.1.1.2.

* Also found out all Release Notes contains tips regarding upgrade path or different versions compatibilities, so better check out this document first.

Finally, saw this Unfortunate Reminder from the latest EQL v7.x firmware Release Notes…so older Equallogic arrays can loaded with the latest firmware IS NO LONGER TRUE, sigh…

ops

In additional, there is a good stuff that finally we can use 4k or 512-byte sectors, so this should save a lot of wasted space in Snapshots.

4k

Update: July 6, 2014

Problem still exists after upgrading to v5.2.11, what the hell? Local Pro-Support at XiaMen remains unprofessional probably even worst than 4 years ago as he asked me “Can you accept Downtime” when restarting the controller? $#%R$#@!@!!!!!

The other thing I found out is the Restart Button in EQL Group Manager GUI is actually a way for active controller to failover, so your volume will still remain on-line just like firmware upgrade controller restarting  process.

Update: July 7, 2014

Dell L2 support team head called me this morning, he mentioned US EQL support also noticed this specific error remains even with latest firmware v6.x applied, and it will be fixed in a later firmware hopefully.


美圖秀秀

By admin, June 4, 2014 5:18 pm

I watched a TVB Pearl documentary two days ago, it mentioned a startup called 美圖, the 3rd largest Internet company in China behind Alibaba and Tencent (ie, QQ)

Then I downloaded their flagship software MeiTuXiuXiu today…of course it’s free and you can get it from http://xiuxiu.meitu.com and installed on your iPhone and Android.

The attached is the modified photo using MeiTuXiuXiu, it only took me 5 mins to do all the effects, it’s quite amazing! This is bad for all the Photoshop designer though.

Before:

717z

After:

chichester

Pages: Prev 1 2 3 4 5 6 7 8 9 10 ...26 27 28 Next