Web-scale to Enterprise
Discussing how web-scale architectures are moving to Enterprise IT. Join the movement.
Stuart Miniman
#Webscale @stu Q: Is #webscaleIT a legitimate infrastructure term or is it a semantic hijacking?
2 hours ago
Quoted By @stu
Jack Clark I think a better term might be "Comp-Sci PHD IT"
Kong L. Yang apparently, a standards body has to legitimize the term for infrastructure... with a sacrifice or an equivalent exchange :-)
Stuart Miniman @mappingbabel enterprise won't have PhDs (and I questions how many coders) - are the PhDs building solutions that can be standardized? Enterprise has different scale (distributed is useful for many reasons).
Aditya Vempaty if something requires a PHD to do..we have failed as companies.
Jack Clark @KongYang Yes. I do think clarifying what this term means for enterprise is worthwhile though - if only to avoid the future FUD.
Aditya Vempaty @stu exactly..most end users aren't PHD people. they care about making apps for their clients..this means infrastructure should cater to app needs without requiring constant configuration. @cohodata #webscale
Jack Clark @avemii Point I'm making is a lot of webscale tech is actively developed by some internal PHDs. Making it consumable can be quite difficult. Companies do have a role here, but it's a tough one.
Suda Srinivasan @stu If #webscale is about scale-out dist., sw-centric, shared-nothing architectures, then v. much legitimate. No point nitpicking w narrow definitions of terms. The key qn is what the problems are and how to get past them.
Steve Kaplan The whole idea of web-scale converged infrastructure is to provide #webscale without the PhDs.
Aditya Vempaty @mappingbabel agree..we have alot of PHDs working on these issues. But we are cognoscenti of making sure its consumable without requiring constant config for the end user. @cohodata #webscale
Stuart Miniman @avemii what does the IT job of the (near) future look like? what goes away and what new things must they learn/do?
Jack Clark Yes! Though stuff like easy multi-DC replication/load balancing/DB management is so tricky I think lots of enterprises are having to hire some sophisticated&expensive people to manage it.
Steve Kaplan Storage administrators need to become more diverse. The days of LUN carving & fibre channel zoning are coming to an end
Jack Clark @stu @avemii Familiarity with cluster management systems & knowledge of IO will always be important, but IT will have to do less fiddly stuff.
Stuart Miniman @mappingbabel are we hiding/abstracting the problems (if things go wrong, we are in trouble) - or are we creating SIMPLE solutions that can be managed easily and fixed/upgraded easily?
Aditya Vempaty I think due to virtualization and making infrastructure app aware. The typical IT person will have to know the entire stack not just their area (networking,storage, apps).
Jack Clark Yes! Storage is on the cusp of a pretty major usability/maintenance/management change. (Networking maybe in five years, I reckon).
Aditya Vempaty agree..key will be understanding what your applications requirements are..to make sure you deliver value to your end users (from an IT persons view) @cohodata #webscale
Aditya Vempaty @mappingbabel This could be based on type of infrastructure you chose for the app, as #webscale isn't entire just virtual can be for non virtualized apps. As those are the most performance and maintenance hungry. @cohodata
Suda Srinivasan @stu If your business is IT, go ahead, hire a team of PhDs and get to work. It makes sense. If you're a retailer or a hospital, then no. But "#webscale inside" is on the rise. Products that look like trad. infra but behave like webscale architecturally.
Suda Srinivasan @stu it's about automating issue detection and resolution, creating self-healing systems that are easy to manage. Automation at all levels is key to simplification.
Jack Clark
Is "webscale IT" an actual design philosophy or a bunch of technologies that have been glued together by startups?
Stuart Miniman 1. It's an operational philosophy; 2. There are technologies that web-scale companies are doing; 3. Enterprise IT will NOT end up looking the same (scale and app mix different). What do others think?
Gabriel Chapman unless there is a true benefit to doing so, Ent IT will not adopt. Re-writing 99% of their apps is not an option. Ent apps today dont take advantage of webscale properties.
Jack Clark @Bacon_Is_King what level of redesign is required for a typical enterprise app to make use of some kind of PAYG IT infrastructure, though?
Gabriel Chapman Twitter uses webscale because they need to, and because they are on the "Web", standard Ent e-Biz suites cant integrate into a webscale design, the apps have no awarness of those properties.
Karthik Ranganathan @Bacon_Is_King - you're spot on. Thats exactly why Nutanix tries to let the enterprise run existing apps, but makes the rest of the architecture webscale (using VMs)
Gabriel Chapman @Bacon_Is_King taking advantage of the webscale properties is what the app needs to do to leverage them, its not a philosophy of operations, its about the ability to merge the app, with the resiliency of webscale design.
Suda Srinivasan Technologies embody the philosophies and architectures (e.g., scale-out, resilient, flat arch, etc.).
Kong L. Yang webscale is a services delivery & consumption philosophy.
Suda Srinivasan @Bacon_Is_King Waiting for apps to become scale-out, distributed, etc. is neither an option nor required. Ent. apps should be able to use web-scale infra without needing rewriting or arch.
Gabriel Chapman @sudasrinivasan only if your view of webscale is limited simply to the underlying infrastructure and not the ability of the apps to have awareness and leverage that infrastructure for resiliency, scale, etc.
Jack Clark @Bacon_Is_King @sudasrinivasan how does an app get awareness in the first place? Isn't the key instrumentation & perf telemetry - how do you accomplish this for the enterprise?
Aditya Vempaty that actually happens by leveraging app context (assigned by the app admin). As entire infrastructure should cater to app
Aditya Vempaty once the infrastructure knows the profile of the app you it can cater to it. This allows #webscale on demand of performance/bottlenecks etc. @cohodata
Stuart Miniman
The Dell Nutanix deal http://siliconangle....
What Qs or comments do people have?
Dell takes another shot at convergence market with Nutanix alliance | SiliconANGLE
Hot on the heels of arch-rival Hewlett-Packard’s unveiling of a slew of new enterprise and supercomputing systems at its annual Discover conference earlier this month, Dell has upped the ante with its own fresh batch of data center products.
Petri Rantanen As Nutanix partner this puzzling. Not sure how Dell work with partners.
Stuart Miniman @petrirantanen the channel is critical to this being successful - Nutanix is making sure not to damage relationships or give an OEM an unfair advantage and Dell wants to give partners what they need to succeed
Stuart Miniman @petrirantanen also see http://www.nutanix.c... from @roidude
Implications of the Dell OEM Relationship for Nutanix Channel Partners | Nutanix
In the technology industry, OEM means that a manufacturer purchases a product at a deep discount with carte blanche to package and sell it as desired. Nutanix, consistent with our corporate DNA for breaking the rules, is also changing the OEM game. W...
Jack Clark @petrirantanen @stu agreed - infra hardware companies must work with the channel unless they are vending basic commodity stuff.
John Furrier @mappingbabel Sam is cool exec Dell is lucky to have him
Nikita Maheshwari @petrirantanen Did that post help clarify? If not - Email us directly and we can have a live convo with you. Will DM you my email.
Jack Clark
So @avemii says webscale is an actual design philosophy - you make storage deform around apps. Fairly sensible.
Stuart Miniman Does this make things like VMware (abstracting the app from the infra) less relevant?
Alex Newman More context please
Gabriel Chapman which webscale companies are even using a hypervisor today?
Jack Clark @posix4e the idea seems to be that you automatically tier and place storage according to needs of app without having to know much about it.
Jack Clark @Bacon_Is_King Most use containers internally, but serve up hyporvisors for untrusted users. Amazon: Xen. Google: Linux Containers with KVM on top (Compute Engine).
Gabriel Chapman whats old is new again.
Petri Rantanen @mappingbabe Hypervisors still needed in enterprises, over 1k win apps + linux apps.
Karthik Ranganathan @stu - actually still relevant. VMWare abstracts the app from CPU and memory. Storage virtualization (storage density of machines, storage tiers based on performance - flash, disk, etc) is another piece.
Aditya Vempaty exactly, focus = giving resources to applications without having to lift a finger. Do this by giving context to applications via app level.
Alex Newman "the idea seems to be that you automatically tier and place storage" - That will only work for applications with trivial storage requirements. It's easier to move the code than the data.
Jack Clark @petrirantanen @Bacon_Is_King agreed - press has got carried away and sees containerization as a solution to all woes. It's not.
Karthik Ranganathan @posix4e - depends on whose code it is. If the infrastructure supports moving the data, there is no need to move code. In any case, I feel this works more for application with complex requirements rather than simple ones.
Jack Clark @karthikr @posix4e I have a feeling that application complexity is only going to grow over time. Is there evidence to the contrary?
Karthik Ranganathan @posix4e - simple principles like storing cold data (archival data) in cheaper tiers and hot data (cache hits, etc) in high performance tiers is universal. The idea is to expand on this.
Karthik Ranganathan - it will become more complex, no doubt. The more stable infra you provide, the more the applications will evolve and push it.
Alex Newman Hmm i feel like for large amounts of data, moving it is going to only grow in expense. SATA controllers are getting faster than network controllers are.
Alex Newman @karthikr For instance SATA revision 3.2, is something like 16 gbps. Expect even desktops to have 2 of these. Servers more like 6. When do we expect 100gbps in the datacenter?