Galvin s All Things Enterprise The State of the Clod, Part 2 PETER BAER GALVIN Peter Baer Galvin is the CTO for Corporate Technologies, a premier systems integrator and VAR (www.cptech. com). Before that, Peter was the systems manager for Brown University s Compter Science Department. He has written articles and colmns for many pblications and is co-athor of the Operating Systems Concepts textbooks. As a consltant and trainer, Peter teaches ttorials and gives talks on secrity and system administration worldwide. Peter is also a Lectrer at Boston University and Senior Contribtor to BYTE. Peter blogs at http:// www.galvin.info and tweets as PeterGalvin. pbg@cptech.com In the previos edition of Galvin s All Things Enterprise, clod was the center of attention. In spite of the over-hyping and freqent nder-delivery of clod, it s still an important, new, and evolving area of compting and, therefore, worth some discssion and analysis. The first task was defining clod compting, and the next was exploring why clod what does clod compting bring to the table and why shold yo care? This colmn contines the analysis by giving examples of projects that have sccessflly sed clod compting. Why did they scceed when others have failed, and what did they gain by sing clod technologies? Of corse there are reasons to avoid clod compting, and those are inclded as well. The colmn finishes with a comprehensive list of clod considerations what yo shold consider when determining if a given project shold be based on clod technologies, and whether it wold be best implemented in a pblic clod, a private clod, or a hybrid clod, or bilt sing non-clod technologies. Who Is in the Clods It seems to me that clod compting started in the midst of Web 2.0. What at first blsh was simple co-location rnning an application or an entire bsiness in someone else s datacenter evolved to rnning the same on someone else s gear. That hosting model then frther evolved into one of rnning mltiple companies applications within the same infrastrctre. Sch mlti-se reqired better management tools, monitoring, alerting, and billing. Those became a set of clod compting tools. Along the way, many Web 2.0 companies did very well by not rnning their own datacenters. Take SmgMg as exhibit nmber one [1]. Their se of Amazon s S3 storage clod is a case stdy in how a company can redce overhead, costs, and complexity [2]. This photo sharing and management site stores the photos via the APIs provided by Amazon, while retaining the cstomer information and metadata within their own compters. Certainly that s an example of a company that shold se the clod and an application that was ready-made for clod integration. Necessary clod attribtes (that it s elastic, metered [pay as yo grow], shared, and Internet-based) are all present in this case. Of corse, SmgMg is one of thosands of Web 2.0 companies that base their compting or their storage on the clod. Back dring the dot-com boom, companies needed a lot of ventre capital, as well as a lot of IT knowledge (either internal or for hire), to move their idea from paper ;login: OCTOBER 2011 39
to fll-scale exection. Now, the idea still needs IT knowledge, bt reqires less fnding and less investment. Fortnately for companies looking to be in the clod, there are many providers trying to add to their client list and take their money. The leaders inclde both tried-and-tre companies sch as Amazon, Microsoft, IBM, Google, and VMware, and newer companies sch as Rackspace, Salesforce.com, Joyent, NetSite, 3Tera, Terremark, and GoGrid. These companies vary in their offerings, pricing models, and abilities, bt all provide IT resorces via a pay-asyo-go model. My Clod or Yor Clod? As good as the pblic clod model is for some companies, it leaves many other companies wanting. Qestions abot reliability, secrity, and performance, as well as reglatory reqirements and corporate policies, prevent many companies from tilizing the pblic clod prodcts. Given the recent, very pblic clod failres [3], companies that depend on their data and compting resorces to be available or nder their control are choosing not to se the pblic clod, or at least not to se it for large swaths of their compting needs. Sch a choice does not mean these companies cannot have pblic clod-like featres for their projects. Companies still desire the elasticity, manageability, rapid deployment, and even chargeback (or the no-payment-reqired version known as viewback). What are sch companies to do? The soltion for them is to se clod technologies within their own datacenters private clod, in the parlance of or times. Sometimes companies want to se private clod, as well as sing pblic clod facilities where applicable. This configration is called hybrid clod. Private clods can look a whole lot like what we sed to call infrastrctre, bt there are some implementation choices and technologies that can give them clodlike aspects. Consider one of my clients that had the following problem. Client X had a small, fll datacenter. It was traditional in that there were dedicated servers for each application, a small SAN for all important data, a tape library for backps, and a 1 Gb network for interconnection. X was growing, needed to move to a larger datacenter, needed a DR plan beyond jst shipping tapes off-site, and needed to move qickly to respond to new bsiness-driven IT initiatives. They chose to se a co-location facility to provide ping, power, and pipe for their racks of eqipment. Other improvements inclded moving to VMware ESX to layer applications across a pool of servers, and sing NAS storage to hold their prodction data as well as the virtal machines. The NAS array also provided them with replication of the data to a second NAS array at a second co-location facility for DR. Moving to a 10 Gb networking interconnect gave them better performance and more room to grow withot rnning ot of throghpt. The project also involved deploying tools to enable release management, configration management, capacity management, and change management based on the virtalized environment. Shold this project rightly be called a next-generation infrastrctre or a private clod? Both are correct, bt becase X now has infrastrctre-as-a-service (IAAS) and service management for their application deployment, as well as elasticity, I believe it is a private clod. As another example, consider client Y. They had an existing bsiness continance (BC) plan, bt that plan failed when it was needed the most dring a disaster. They cold not gain access to their normal offices, so declared a disaster and switched over to the disaster recovery (DR) site. Workers started arriving there, 40 ;login: VOL. 36, NO. 5
and all was well ntil the nmber of workers increased. The plan had been tested, bt not at the scale of the entire company. The DR infrastrctre fell over and work cold not proceed. After sorting throgh the varios options, Y decided to pgrade their BC plan and facilities. Rather than have workers go to the BC site, the workers wold work remotely, across encrypted tnnels, sing a virtal desktop infrastrctre (VDI) facility. The applications rn within their BC site, bt the workers get remote views of their virtal desktops from anywhere that Internet is available. Becase of nderlying virtalization, prodction applications are replicated to the BC site, so all apps are kept p-to-date. Internet technologies allow remote access, and by adding more CPU and memory to the BC farm, they can easily scale the facility as needed. Again, this cold be labeled with varios names, bt private clod is certainly one of them. Clod Candidates Are there certain application and IT initiative aspects that predispose them to be best deployed in a pblic clod, private clod, or left as is on traditional infrastrctre? There certainly are trends and sccess (and failre) stories that show that some projects are better matches for clod than others. While there are not any absolte rles, a project involving these aspects is probably a good fit for a pblic clod: Software as a service Adio/video/Web conferencing Sales/CRM atomation BC/DR Training/demonstration services Collaboration Email Development/test facilities Other aspects show a tendency to be best left in a private clod: Large data movement Sensitive data Reglated data Complex processes/complex transactions Low latency reqirements Non-x86 applications Yet other aspects may reveal projects that shold be left on existing infrastrctre: Legacy applications Indstry-specific applications Real-time applications Very large (CPU, memory, data) applications As with cars, yor (project s) mileage will vary. Every site is complex, with many decision points, criteria, and experiences. All that will provide gidance on what to place in clod infrastrctre and what to leave as is. ;login: OCTOBER 2011 Galvin s All Things Enterprise 41
Clod Considerations In my experience, it is possible to codify at least some of those what to rn where decision criteria. The following set of giding factors can be sefl in applying logic to the task of determining how best to rn a given application or given facility. For each of the following technology areas, yo shold decide whether the area is a factor or not. If an area is a factor, then docment why. For example, the operating system might not be a factor becase yor application can rn on any OS, bt networking might be a factor becase it s 1 Gb and yo need to move to 10 Gb for the throghpt yor application needs. The list incldes: operating systems, applications, servers, storage, networking, Internet technologies, virtalization, logging/ reporting/analytics, mobile access, seasonal resorce se, elasticity/scalability, and any other technology criteria that might be important to yor site. Next is a set of design reqirements that cold steer the project toward one type of infrastrctre or another. This list incldes large data movement, non-virtalizable software, low latency, and high cstomization reqirements. On the financial front, the following areas cold be rated in terms of importance, from not important throgh very important: redcing OpEx, redcing CapEx, licensing cost redction, ROI reqirements, and chargeback/viewback reqirements. Another area to consider is the line of bsiness that the application or facility is destined to spport. The LOB again might have importance ratings in areas sch as keeping the infrastrctre separate from others, reqired SLA strength, capacity or performance garantees, the need to control recovery from problems, atomation of workflows, and self-service abilities. In the risk and reglations area, some factors yo shold consider are the inclsion of validated systems, reglated data, sensitive/proprietary data, reglated systems, HIPAA/SOX or other reglation compliance, corporate secrity policy reqirements, and whether there are strong secrity needs. In the final area of project exection, yo shold think abot whether staff members have the skills to design and implement the project within the facility selected, whether they can do so within any time constraints, and whether the team has the knowledge and tools for ongoing monitoring, maintenance, and management of the facility. Beyond these considerations, don t forget any site-specific, project-specific, or staff-specific reqirements or limits on the broad isse of where to rn the facility. Experienced IT managers know that beyond those broad decisions, a project scceeds or fails based on the myriad of details it encompasses. Clod is not a panacea that removes the need for planning and exection. In fact, clod compting can place more emphasis on management, teamwork, decision-making, and debgging than more standard projects do. One final note: clod compting is important and is changing how infrastrctre is bilt and sed and how mch it costs. That does not mean that clod compting can solve all problems or is right for all environments or all projects. Sometimes the internal strctres of a company or the ways in which roles and responsibilities are divvied p can mean the difference between sccess and failre of a clodcentric project. Many companies are finding that between politics and those old 42 ;login: VOL. 36, NO. 5
strctres, mch internal change is needed in order for the company to embrace clod compting. Tidbits If yo are interested in performance analysis and debgging, especially based on DTrace, yo shold have a look at the new DTrace: Dynamic Tracing in Oracle Solaris, Mac OS X, and FreeBSD by Brendan Gregg and Jim Maro. It s everything yo cold want in a DTrace book. See my fll review in the Book Reviews section of this isse. If yor interests lie more in the direction of ZFS, then yo might want to check ot my first video pblication. This one is based on the Solaris ttorials I ve taght many times for USENIX and elsewhere. The official name is Solaris 10 Administration Workshop LiveLessons (Video Training): File Systems, bt it s 90% ZFS, inclding both theory and hands-on examples of configring and sing it [4]. On another front, I m pleased to be part of the relanch of BYTE. As a yong lad, I spent many an hor poring over the pages of the venerable magazine, delving deeply into technology details of many aspects of compting. BYTE is back, and I m one of the Senior Contribtors there. Have a look at http://byte.com and let me know what yo think. References [1] SmgMg: http://www.smgmg.com. [2] SmgMg case stdy: http://aws.amazon.com/soltions/case-stdies/ smgmg/. [3] http://www.crn.com/news/clod/index/clod-otages-clod-services -downtime.htm. [4] Available at http://www.informit.com/store/prodct. aspx?isbn=0321753003#lessons and http://my.safaribooksonline.com/video/ -/9780321718372. ;login: OCTOBER 2011 Galvin s All Things Enterprise 43