Los Angeles nonprofit group swaps out Cisco Catalysts for stackable 10G Raptor boxes for greater speed, lower costs
Network technologists at L.A.Care Health Plan had several conflicting ideas for its new LAN backbone: boost core LAN bandwidth to 10Gbps; segment the network into multiple departments; keep the infrastructure a flat Layer 2 topology; and spend as little as possible on new gear.
The nonprofit says it hit all these points when it took a chance on relatively unknown LAN switch maker Raptor Networks. After suffering through some technology growing pains with the 3-year-old start-up, L.A. Care built a resilient 10G Ethernet core on Raptor’s stackable switches and clustering technology. The Raptor deployment also saved the organization more than $250,000 compared with upgrade options offered by incumbent vendor Cisco.
Keeping network costs down was key for L.A. Care, which has around 350 employees and manages healthcare services for 750,000 low-income and elderly Los Angeles residents. But the network was facing a bandwidth crisis; a recent installation of Oracle Real Application Cluster (RAC) servers taxed the network, while the flat LAN setup caused other applications to slow down and interfere with each other, says Rayne Johnson, director of networks and security.
“With all of that, coupled with nightly network backups, we noticed server network slowdowns,” Johnson says. “The main reason for that was because we were on a very flat LAN [a Layer 2 network, where all traffic was mixed]. I needed to segment the network better, according to our business needs.”
Choices Johnson says he had from Cisco included upgrading to Layer 3 at the core, or upgrading to 10G in the core and using virtual LANs to segment traffic. Either move seemed to lead to a hardware upgrade to the Cisco Catalyst 6500 Supervisor 720 switch fabric module for the switches, as well as software and services upgrades.
Instead of moving up to the Catalyst 6509s — which support multiple 10G links on a line card, and 32 10G ports total — the organization installed four stackable 10G Ethernet switches from Raptor, which is based in Santa Ana, Calif.
The Raptor ER 1010 switches are used to form L.A. Care’s LAN core using a multipath Layer 2 mesh with multiple 10Gbps links. Overall, the Raptor network core cost around $180,000 for hardware and installation, while Cisco had quoted a price of around $500,000 for an equivalent network.
The ER 1010 switches each have 24 Gigabit fiber or copper ports, and six 10G Ethernet ports. The single-rack-unit boxes support Layer 2-4 switching and can run a proprietary modification of 10G Ethernet called Raptor Adaptive Switch Technology (RAST), which allows devices to be hooked together in multipath mesh at Layer 2. RAST is used in place of the Spanning Tree Protocol, which does not allow multihoming at Layer 2. (Switches can also be configured to run IEEE-standard Gigabit and 10G Ethernet).
RAST inserts switch routing information into unused header space in standard Ethernet frames, according to Raptor. This injected data provides heart beat and route path information among switches in a multi-10G cluster.
At L.A. Care, RAST binds the four core switches into a 30Gbps core, with three 10G Ethernet links connecting each switch in a multilink mesh. This setup allows multiple paths to exist among the core switches, and provides subsecond failover in case of a cable or switch hardware failure.
The Catalyst 6509s were moved to the LAN edge, where they connect user machines as well as aggregate other stackable switches from D-Link into the core.
Since L.A. Care was an early adopter of the Raptor technology, the 10G rollout hit a few bumps, Johnson says.
“Initially, we had some big problems, to be honest,” Johnson says. An incompatibility between Cisco IOS and Raptor’s RAST protocol caused a large Spanning Tree loop issue between the standard Ethernet ports on the Catalyst switch and the non-RAST ports on the Raptor boxes. After a software revision from Raptor and troubleshooting from the company’s engineers, this was solved in about a week.
“Since then, the switches have been running like a clock,” Johnson adds. “RAST is pretty solid. If you lose functionally from one [switch], you don’t see whole segments of the LAN failing; machines attached to the switch might fall off, but the LAN overall will stay healthy.”
The Raptor boxes were deployed to segment the company’s flat Layer 2 LAN into VLAN subnets, keeping it at Layer 2, with 10G Ethernet in the core. Three 10G Ethernet pipes connect each box in the mesh; the Catalyst switches have been moved to the LAN edge for connecting the organization’s 350 users and other devices.
Servers are also plugged into the Raptor core on non-RAST Gigabit Ethernet ports. These include the Oracle RAC servers, which run several portal-based applications, along with IBM p5 servers hosting some legacy financial applications. Each of the RAC servers are multihomed — connected with two Gigabit Ethernet cables — to separate Raptor switches.
“Each RAC node has a [network interface card] going to a different Raptor,” Johnson says. “If one goes down through the heartbeat [the way Oracle RAC’s cluster nodes identify each other] or though the LAN, the server stays up and operations continue.”
The Raptor switches are also used to segment traffic from various applications among departments, such as finance, customer service/call center, and human resources. Coupled with the new 10G Ethernet core, this setup has boosted application performance. The organization’s internal Web portal, for instance, now responds in subsecond time, instead of taking several seconds to process a request, Johnson says.
Despite suffering through some early interoperability headaches with what at the time was new Raptor hardware, the results are good.
“It’s been smooth and very fast, and for the price I couldn’t beat it,” Johnson says.
