Brocade - ЦОД будущего

Editor's Notes

• #8: Page 7
• #14: We work with customers all over the world to help them solve storage networking problems. They describe an evolving landscape in IP storage. The demand for storage capacity is exploding. The transition to higher-performance, more business-critical IP storage is also well under way. IP storage that was once used for lower tier workloads, file sharing, and home directories is now being used for business-critical workloads like server virtualization, analytics, and other emerging workloads. In parallel IP storage connectivity is transitioning from 1GbE to 10GbE and moving from Terabytes in size to Petabytes. And the fastest growing segment of IP storage is scale-out storage such as scale-out NAS and object storage. We are also hearing that mixing these new and business-critical IP workloads with existing workloads on a legacy shared front-end network create big performance and operational challenges. Are you seeing these trends in your environment?
• #15: There are three key trends driving the deployment of scale-out infrastructure, which in turn drags both scale-out storage and scale-out networking. The first is the rapid pace of growth in Unstructured Data. A hallmark of this data is that it inherently has value to the business, so many organizations are keeping this data in warm storage rather than relegating it to cold storage on a tape stored in a mountain somewhere. There are many sources of this data from machine logs to video cameras to Facebook feeds. The second is the emergence of distributed file systems. These file systems can take advantage of direct-attached or clustered storage and federate multiple nodes to appear to be a single logical name space. This allows for a more affordable and simpler way to store the massive amounts of unstructured data. Examples include Isilon’s OneFS for scale-out NAS, and the Hadoop file system (HDFS). Third is the availability of new analytics engines top run on this unstructured data. There is business value in the data itself which is being mined by tools such as Greenplum and others. The analytics engines are driving scale-out compute, storage and network deployments
• #16: This is a picture of a typical legacy shared front-end IP network. Here’s the challenge. These business critical workloads must be managed under increasingly strict SLAs dictating availability, performance, and agility that are similar to the storage area network that you’re already familiar with. However, many customers that we speak with are running these top-tier workloads on a shared, general purpose network infrastructure. Why is this a problem? Stringent SLAs cannot be guaranteed when the storage traffic is running over a best-effort, shared infrastructure managed by a network team who doesn’t understand the unique requirements associated with storage traffic. And when applications don’t perform as needed you’re held accountable, even when the root cause is a network issue for which your team doesn’t control. Consider this. Storage capacity is growing at a compound annual rate of 40%, doubling almost every two years. The number of applications and workloads that depend on this storage is skyrocketing. When application performance degrades or even worse, there are outages, the “mean time to innocence” clock starts ticking. This is the time it takes to understand and prove that the storage is not the root cause. The problem could be related to a recent update to the network switches or misconfiguration during a recent provisioning of new ports. Regardless, it takes longer to resolve issues and there’s a lot more finger pointing and blamestorming. There are operational limitations associated with running your storage traffic over the shared network as well. The storage team must rely on the network team to schedule change windows and often the network team has competing priorities. This hinders the storage team’s ability to respond to the needs of the business in real-time. <Click through each specific challenge> <Use customer examples of this pain, for example>: This is best illustrated by a recent situation experienced by one of our customers. This customer recently attempted to deploy a new rack of servers that was supposed to add new capacity to their VM farm. The server team racked and stood up the servers quickly after getting them in the data center. The storage team had new NFS mounts and capacity provisioned within minutes. However, the network team needed a couple of weeks to plan for the switch and port reconfigurations that were required to add new network capacity.   Making matter worse, once servers, network, and storage were put into production, they kept running into performance issues on a few of the new VMs. The application owners thought it was a storage problem; the storage team thought it was a network issue; the network team thought it was caused by a rogue application. It turned out that the network had to be taken down due to fix misconfiguration issues, causing an impact to all of the applications connected to the IP storage.   This is a scenario we see repeated at many of our customers that are finding that they can’t keep up with the demands of the business due to limitations and inflexibility of their IT infrastructure, as well the complex operational and organizational model.
• #17: This is a picture of a dedicated IP storage network Imagine being able to control your own destiny by eliminating your dependency on the network team for your network capacity. By managing a dedicated network for your IP storage, a network that’s optimized to handle the availability, performance, and agility for your current and future workloads, you can say goodbye to the finger pointing when something goes wrong; critical applications aren’t impeded by non-essential workloads; and storage and network capacity is added without delay. To achieve these benefits you should start by separating your mission-critical storage traffic onto a dedicated network for IP Storage. But that’s not enough. To achieve the level of agility that your business owners demand, you must be able to add network capacity and make changes at the speed of business. This requires an agile infrastructure where you can provision the network without manual intervention. This not only accelerates the time to deployment, but reduces the risk of human error and downtime. Based on your experience with Fibre Channel SANs you know that managing the network and storage holistically will enable you to deliver storage infrastructure SLAs, maintain separate change control procedures, and most importantly minimize the risks of downtime and disruption.
• #18: The idea to deploy dedicates networks for IP storage isn’t new. You can see from the companies listed here that the industry has been advocating this approach for many uses cases, from server virtualization to distributed file systems and storage replication and backup. This is a best practice we expect to grow in popularity and no other vendor is better positioned that us - the storage networking experts - to lead in this use case.
• #19: Here’s a comparison of a legacy network to see why you may want to reconsider the structure of those legacy networks and how Ethernet fabrics can support your data center network needs. What you see on the left hand side of this slide is a conventional data center LAN. It’s characterized by a classic hierarchy consisting of access, aggregation and core tiers. If you think about modern data centers, this legacy architecture has some problems…. First, the network was clearly designed for the north-south traffic patterns that were prominent with client/server applications. This architecture isn’t well-suited for the fundamental changes occurring in data center applications and traffic patterns where server-to-server, or east-west, traffic will be dominant. Gartner predicts that as much as 70% of data center traffic will be east-west by 2016. The network often uses spanning tree for loop prevention, and that means as much as 50% of the links are not utilized, but instead are in standby waiting for a link failure. That’s a pretty arcane concept considering how much time and energy we spend maximizing the utilization of other data center assets, such as with server virtualization, for example. And even with a network management tool, the network is managed a device at-a-time, making it labor intensive and prone to misconfiguration or policy inconsistencies between devices. The network has no awareness of server virtualization and, therefore, cannot actively facilitate operations such as VM mobility or properly deal with the MAC address explosion we see with increasing VM densities. Finally, the network doesn’t have any concept of native network virtualization, forcing the use of overlay/tunnel technologies wherein the physical network doesn’t add any value to the virtualization solution. On the right side of this slide is an Ethernet Fabric architecture, like the one Brocade delivers with our VDX switch portfolio running VCS Fabric technology. This architecture is very different, and has been specifically designed for the east/west traffic patterns and server virtualization that now dominate data centers. You will notice that I have not drawn a specific topology here. That’s because the fabric doesn’t dictate a specific physical topology. You can use a leaf/spine design, a full mesh, a flat fabric at the access or aggregation layer, or any other topology you deem appropriate for your applications and traffic patterns. Unlike the legacy LAN architecture, all links in the VCS fabric are active all the time, with multi-pathing at L1, L2 and L3 to deliver greater utilization, performance and resiliency. The entire fabric can be managed as a single logical switch, minimizing operational overhead for the IT organization. And, finally, the fabric supports both native network virtualization as well as overlay network virtualization models, giving the customer maximum choice in technologies to virtualize their network. An Ethernet fabric is simply a more modern network technology for highly virtualized and cloud-based data centers.
• #20: This a visual representation of the strength of the Ethernet Fabric capabilities. The Fabric layer enables faster deployment, simplifies installations and ongoing management, so that storage administrators can focus on other business-critical needs.
• #21: This shows a comparison between Brocade and Cisco of the number of steps taken to interconnect 10 switches in an Ethernet Fabric. The example is just a demonstration of how much the Fabric offers in terms of faster deployments and easier management.
• #23: At the end of the day, you have to be able to forward the packets. Ethernet Fabrics are the most often-recommended architecture, specified by companies like VMWare, Cisco and Brocade to name a few, as the best foundation for SDN and other New IP-related investments. Why use a fabric as your New IP foundation? To increase agility you need to eliminate complexity and increase automation. Fabrics are agile, natively automated, easily scale up and out, and adapts to handle instantaneous changes in traffic flows, flow sizes, packet sizes and protocols. The result: Quickly deploy new services, easily add capacity, and manage centrally Begin to recover IT “Innovation Capacity” Evolving legacy DC networking to Ethernet Fabrics is a low-risk, evolutionary way to begin your New IP journey. If you’re not yet ready for Ethernet Fabrics, you can deploy VDX switches and use them as a reliable, high-performing L2 switch. VDX switches are compatible with existing legacy infrastructure. When ready, simply enable Ethernet Fabrics and you’re on your way. ============ Revolutionary Results Reduced Operational costs by 50% Sunpower: Deploy apps 92% faster Reduced Capital costs by 50% Singing River Healthcare Cut EHR login 50% Improved EHR performance Move VMs in minutes vs. days, w/out downtime
• #24: In today’s enterprise, it’s very common for applications to be accessed remotely –whether from Cloud or a private cloud hosted in the Enterprise DC. Additionally, applications and server today leverage services throughout the datacenter –creating high volume East-West traffic Ethernet Fabrics collapse the DC architecture and leverage all links to reduce East-West traffic congestion but that’s only part of the solution. Add in the SteelApp virtual application delivery controller to further optimize East-West traffic and improve application delivery over the WAN Pain Points Poor application performance Decreased employee productivity and satisfaction High server-to-server latency High BW / WAN costs needed to meet app SLAs =========== Enable SteelApp Features: Traffic Manager Web Accelerator Web Application FW Services Controller