Read articles from past Farsighted News editions.
OCTOBER 2006 | AUGUST 2006 | JUNE 2006 | APRIL 2006 | FEBRUARY 2006
QUALCOMM Inc. is a leader in developing and delivering digital wireless communications products and services based on CDMA and other technologies. To support its global network of development and design centers, Qualcomm is enhancing its file replication system to be able to fully replicate critical design data to all of its design centers every 12 hours. Currently, Qualcomm replicates over 8TB of data, stored in over 50 million files, but expects that number to climb to over 12TB in the next year, with additional growth in ensuing years as new remote development sites come on-line. Rob Mallory, IT engineer for Qualcomm, Inc., is responsible for enabling Qualcomm’s business model of 24-hour development at remote sites located around the globe. Using innovative technologies, he has slashed the time it takes to replicate over 8TB of data from 42 hours to less than 12 hours, allowing Qualcomm to replicate design and development data at its remote centers every 12 hours. With more design centers planned, he continues to investigate leading-edge technologies to further streamline this massive replication effort. Why is data replication so critical to your business model? We have eight separate design centers and we need to have all of those centers working with the same data as much as possible. There is no technology that will allow us to achieve instant replication.However, with our current approach we can replicate the data in all of our design centers twice a day.So when our centers come online in India, they have the latest data from San Diego replicated at their center and vice versa – supporting our 24-hour development efforts.Also, if there is some type of disaster and our WAN goes down, our remote design centers can continue to operate using the replicated data. Afterwards, we can quickly recover and reconcile the data. What architecture approach has allowed you to replicate so much data so quickly? We had to process the data from our file servers so that we only hit each local server once during each replication cycle. That processing dramatically increased the time needed to replicate the data, since we had to hit the data and then distribute to each center sequentially. This was a problem with the number of sites that we already had – and it would only get worse as we added sites. To address that, we designed and implemented a staging file server that allows us to replicate an entire superset of all remote sites into a large SAN running a SUN QFS shared file system. Once the data is in the SAN, we can distribute the data to the remote sites in parallel. With that design, increasing the number of remote sites receiving the replicated data does not significantly increase the time to perform the total replication. What technology change had the most significant impact on your approach? In the Sun QFS architecture, the data and metadata can be stored separately. We took even greater advantage of this feature by storing the metadata on solid-state disks, providing remarkable performance for disk-intensive tasks, particularly on query-only requests. The bulk of the time required to synchronize files across remote sites is the "stat" process – that is, comparing the current state of two copies of file to see if a newer version of the file exists. This "tree-walk" of the metadata process alone required nearly 36 hours, or 75% of the total time to synchronize the remote systems. Once we fully convert to the staging fileserver SAN, we expect to reduce this to 8 hours or less. At the last moment, we were able to switch from 2Gb to 4Gb technology for our HBAs. When coupled with the Sun QFS file system, this is really what makes our architecture work. When Sun introduced their 4Gb HBAs, we were really excited because it gave us a 4Gb technology that was qualified by Sun. The biggest benefit of 4Gb technology is the reduced latency to reduce time to examine the entire data set for changes. Another benefit is the bandwidth capacity, since the Sun T2000 servers have 2 PCI-E slots and 1 usable PCI-X slot. Cheryl D. Krivda is a freelance writer specializing in IT and business topics. Contact her at cheryl@cmkcom.com If you are an IT Professional with a Storage Implementation Story and would like to be considered for a feature interview in FarSighted please contact us at christina.wong@snia.org. |
