Friday, February 12, 2010

Inviting Beta users for our new Silverline service

As I hinted in my previous entry, we are readying a service that will enable compute cloud users to utilize spare cycles in cloud server instances that are running interactive or otherwise important “primary” workloads, for running background workloads. The idea behind this service is similar to the idea behind EC2 “spot instances”: provide a cost effective way to run not-time-critical workloads in the cloud. The main benefit of our Silverline service as compared to spot pricing is that it will be less expensive (since we’re using resources you have already provisioned) and that jobs executed under Silverline will not be terminated without notice. The foundation for this service is our Load Manager / Container Virtualization technology, which is well-proven and has been in use at private data centers since 2008.
The service will be available through a new portal. Background applications will transparently be encapsulated in containers prior to execution, while primary applications are not impacted. The background applications will consume resources not needed for primary workloads, but will not change the behavior of the primary workload –guaranteed! To use the service, background applications need to be able to run on the same Operating System as the primary application.
We plan to launch a (free) Beta version of Silverline that will support Amazon EC2 server instances in March, and if you’re interested in trying it out as a Beta user you can sign up here: http://silverline.librato.com/#beta-signup

(0) Comments  |  (7) Trackbacks  |  Permalink  |  rss   atom 

---

 

Wednesday, January 20, 2010

Background workloads in the cloud – for less

In my entry of January 7 I wrote about the use of “spot instances” for background workloads in infrastructure clouds. Spot pricing is an innovative way for cloud service providers to make temporarily unused capacity available to their customers at an attractive price. Thinking about this further, it seems that in the same way as cloud service providers aim to sell all of the capacity they have available, cloud service consumers should aim to use all of the resources they pay for. For many types of applications the limited choice of server configurations offered by public infrastructure clouds will unavoidably lead to average processor and memory utilization levels that are well below 100%. So why not try to use this spare capacity for background workloads? You’re already paying for it and it cannot be taken away without notice, so that’s two advantages over spot instances. What’s necessary in order to do this is the ability to run multiple application workloads on the same server instance and to monitor and control the amount of server resources used by each workload in such a manner that your “primary” application is not impacted by the background jobs running on the same instance. Layered transparently between the Operating System and the applications, Container Virtualization can be used to do this and provide yet another way to maximize the economic benefit of cloud computing. Expect an announcement later this quarter.

(0) Comments  |  (46) Trackbacks  |  Permalink  |  rss   atom 

---

 

Thursday, January 07, 2010

Spot Pricing in IaaS Clouds

Amazon’s introduction of spot pricing for EC2 servers a few weeks back was a sign that cloud service providers need to find ways to improve the utilization of their infrastructure if they are to make public cloud services attractive for a wider range of applications and for large enterprises that enjoy good economies of scale in their own data centers. Spot instances are clearly an attractive way for the service provider to sell off temporarily unused capacity at the best possible price, but can only be used for running background jobs that do not have to complete by a certain time. Many enterprises no doubt have such jobs, but the challenge with the spot pricing model as introduced by Amazon is that the spot instances are taken away without notice if the spot price exceeds the customer’s maximum price. This limits their use to either short running jobs or jobs that run applications that can gracefully recover from a server failure. I wonder why there isn’t an option to snap a spot instance to permanent storage before it is terminated; after all these are just XEN virtual machines. In any case, until such capability becomes available, cloud users interested in using spot instances for long-running jobs could check out our AvS Checkpoint/Restart product. More in general, if you’re using a public Infrastructure Cloud and have concerns about the reliability of the service for long-running background jobs, a Checkpoint/Restart facility could broaden the range of applications you can run in the cloud.

(0) Comments  |  (81) Trackbacks  |  Permalink  |  rss   atom 

---

 

Wednesday, December 16, 2009

Compute server obsession?

If you step back and try to look at the IT infrastructure landscape unbiased by the commonly held opinions we are indoctrinated with by big players, press and analysts, one of the oddities that jumps out is the almost obsessive focus on compute servers as the resource to be managed. Little or no attention is paid to the fact that other resources that are shared between applications are equally important when it comes to delivering the desired Quality of Service and managing expenses. File Servers, Database Servers, and even software licenses are examples of resources that often are a lot more expensive to own and operate than compute servers and have just as much impact on the end-user experience and the cost of delivering that experience. With server virtualization or container virtualization we can manage the allocation of server resources to application workloads at a fine-grained level. This means that we can consolidate servers, but in some cases just to have fewer of them waste time waiting for other resources. Only if it is possible manage the use of all resources shared between applications, can we be in a position to provision the optimal capacity for each class of resource and balance consumption to deliver the best possible Quality of Service. Is it time for vendors of resource management software to shift some of their attention away from the management of compute servers?

(0) Comments  |  (117) Trackbacks  |  Permalink  |  rss   atom 

---

 

Friday, December 04, 2009

Cloud resource utilization

I’ve been thinking lately about the impact IaaS clouds are having on the IT industry and some of the more subtle implications of the model IaaS clouds have adopted. They offer companies significant benefits through pooling of resources, rapid provisioning, avoidance of capital investments and reduced operational headaches, but at the same time one has to conclude that virtual resources provisioned by IaaS clouds are only slightly more “elastic” than physical resources. For example: popular IaaS clouds support a limited number of processor and memory configurations for the (virtual) servers they provision. This means that the workload consolidation issues that exist in physical data centers remain. For some workloads Load Balancers or Grid Schedulers can be used for (virtual) server consolidation, but there is no adequate solution for workloads that in a physical data center would be consolidated using server virtualization, specifically when individual workloads are small and only require a fraction of even the smallest (virtual) server available. The costs of cloud resources are relatively high and 100% variable; for example, an EC2 server deployed 24/7/365 is about 3x more expensive than an equivalent server in an efficiently run data center. This means that finding ways to maximize the utilization of virtual resources will be essential in attracting a broader mix of workloads to the cloud. Container Virtualization can be an attractive way of accomplishing this.

(0) Comments  |  (57) Trackbacks  |  Permalink  |  rss   atom 

---

 

Friday, February 20, 2009

BayLISA

Had the opportunity at BayLISA last night to spend some time discussing Librato’s Load Manager and Smart Suspend solutions with a group of savvy sysadmins (thanks guys!). Whenever I talk with individuals who manage mission-critical infrastructure, I’m always reminded that there’s still an elementary set of unsolved challenges facing our industry that cut across almost all deployment configurations. I find particularly interesting the significant gap between infrastructure and applications that must be bridged before we can deliver truly flexible, on-demand computing.

Here at Librato we’re working hard to provide IT administrators and operators the tools they need to cross that gap and optimize their computing resources. Both Load Manager and Smart Suspend install transparently and painlessly into existing software stacks to systematically encapsulate complex applications and provide administrators with a common, simple, but powerful abstraction to monitor and control the diverse set of services running on both their physical and virtual infrastructure.

Don’t just take my word for it, come check out a live demo during our next Webinar on Wednesday, March 18th, 2009! Details should be posted on the main Librato site shortly.

(0) Comments  |  (58) Trackbacks  |  Permalink  |  rss   atom 

---

 

Sunday, December 14, 2008

Welcome to the Librato Blog!

Hello and thanks for checking out the very first post on Librato’s new blog! With our new website design, I am delighted that we now have the opportunity to blog and establish a more informal dialogue with our current and prospective customers.

So for the foreseeable future I’ll be posting in this space on wide range of specific and general topics somehow relevant to the singular goal of improving data center efficiency. Doing more with less should always be the goal of any good IT manager but it’s particularly critical in tough times like these. Financial turbulence aside, it’s an extremely exciting time in IT that’s turning conventional wisdom on its head with new game-changing technologies in workload management, virtualization, cloud computing, etc. I probably won’t be posting every day, but there’s so much going on in the space these days, I’m sure with fairly regular frequency to find something worth opining about. From time to time you’ll also see posts from other members of our team such as our Founder/Chief Scientist Srinidhi Varadarajan or our CEO Fred van den Bosch.

Although this blog’s main purpose in life is to examine emergent IT challenges and trends, I’d be remiss if I didn’t mention that Librato focuses on application-level workload management and provides solutions to many of said challenges. Our broader vision centers around application-level consolidation, SLA guarantees, cross-server capacity planning, analytics intelligence, and automated management. We believe that the oft-heralded “next generation data center” will bear little resemblance to the clumsy, static, inefficient burden found in so many enterprises today and that our solutions fill crucial gaps blocking that transition.

So whether these are the types of issues/topics you face in your everyday job or just the kind of thing that stokes your intellectual curiosity be sure to come back and check us out. Even better, you can just take advantage of our RSS feeds and have updates streamed to the aggregator of your choice. We’ve enabled commenting, so please feel free to chime in and take part in the discussion. Alternatively you can always just shoot me an email at joe@librato.com.

(1) Comments  |  (51) Trackbacks  |  Permalink  |  rss   atom 

---