Critical Path - Veritas

"With Solid Data's solid-state disks we doubled the speed of our VERITAS Cluster Server switchover and resumption. We're confident that our customers are receiving the benefits of the fastest, most robust server clustering available."

Chris Torlinsky,
Advanced System Architect
Critical Path

Executive Summary
Critical Path^SM was already using Solid Data's solid-state disks to enhance email server scalability. As a leader in internet messaging, Critical Path is continuously looking for ways to further improve service levels. By extending use of solid-state disks to server clustering, Critical Path was able to reduce switchover and resumption times and improve overall cluster performance.

The Problem Description
Once Critical Path had a proven solution to meet its explosive e-transaction growth, attention turned to further improving the performance of their robust, fault-tolerant architecture. Because many companies completely outsource their important messaging and collaboration services to Critical Path, having the fastest possible message processing capability is very important.

To design its high-performance server cluster, Critical Path gathered a team that included experts from VERITAS^® Software and Solid Data Systems. VERITAS provides a highly integrated suite of software, including VERITAS Cluster Server™ (VCS), VERITAS File System™ (VxFS) and VERITAS Volume Manager™ (VxVM). Solid Data provides solid-state disk hardware modules that are shared among the Sun™ Enterprise™ 4500 servers in each server cluster.

Critical Path's advanced system architect, Chris Torlinsky, headed up the team. "At Critical Path we're committed to providing our customers the highest level of performance," Torlinsky said. "Outages and long delays are unacceptable in our world. We've had such great success with the Solid Data solid-state disk on our messaging systems, I decided to investigate how they could speed up our clusters."

An Architectural Solution
At Critical Path, cluster performance depends on several elements of the messaging infrastructure. The architectural solution included integrated software (file system and clustering) and hardware (solid-state storage).

Software: Integrated File System & Clustering
VxFS is a journaling file system: it records each file write before it is committed to disk. Journaling allows file recovery in the event a disk, application or server stops working.

In VxFS, journals are called intent logs because there is an "intent" to write the data. Since they record every file write, the intent logs are highly I/O intensive.

Normally, intent logs are located on the same device as the file system. However, VxFS includes an important feature - QuickLog™ - which allows the intent logs to be moved to a separate device, thus eliminating contention for the same I/O connection. According to VERITAS, the use of QuickLog with VxFS on rotating magnetic disk can boost performance over the standard Unix file system by as much as 10 times. Using QuickLog to move the intent logs to a solid-state disk can further accelerate performance.

VCS and VxFS work together to accomplish a rapid resumption when a server or application stops working. VCS directs VxFS (running on a switchover server) to bring the file system up-to-date. In response, VxFS reads through the intent logs from the point of switchover forward, and applies all uncommitted writes to the file system. Then VCS restarts stopped applications, and processing resumes where it left off. The tight integration of VCS and VxFS permits very rapid cluster resumption.

Hardware: Solid-State Disks
VERITAS intent logs are highly I/O intensive: they are accessed continuously during normal system operation, and in bursts during switchover and resumption. The cluster architecture team decided that multiplying I/O speed on the intent logs was the most promising architectural route to higher performance and faster resumption. The team decided to place the VERITAS intent logs on the Solid Data solid-state disk and to conduct before-and-after performance measurements.

Critical Path had been using Solid Data's UltraSCSI models with its email servers; however the new cluster architecture required a shared solid-state disk. To meet this requirement the team chose the Solid Data 800 solid-state disk, which supports dual-ported Fibre Channel connections. With this shared-storage architecture, each server in the cluster would gain additional performance, and costs would be amortized over the number of servers.

Torlinsky was under pressure to implement the solution under very tight time constraints. "The Solid Data solid-state disk was really simple to set up," he said. "All I had to do was plug it in, format and mount it and I was ready to go." Adding a QuickLog device to a new or existing VERITAS File System was equally easy. "I remounted the device, which took only a couple of minutes," according to Torlinsky.

Results
Having extended the benefits of solid-state disk technology to a second I/O-intensive application, Chris Torlinsky summed up the results: "With Solid Data's solid-state disk, we doubled the speed of our VERITAS Cluster Server switchover and resumption. We're confident that our customers are receiving the benefits of the fastest, most robust server clustering available."

Critical Path - Veritas Case Study PDF: 25 KB (cpveritas.pdf)