Weathering the data storm
Weathering the data storm
How the Bureau of Meteorology manage their mountains of information and their costs at the same time.
Observing meteorological conditions in Australia is a pursuit as historical as the modern society we are part of today, with the commencement of meteorological observations a consequence of the arrival of the First Fleet back in 1788.
In those days, the gathering and storing of scientific data was rather more rudimentary than it is today, with the Bureau's in-house supercomputer now able to support global operational numerical weather prediction and climate studies, helping it attain a position of pre-eminence in the southern hemisphere, and recognition as one of the most technologically advanced in the world.
Recognising that maintaining and furthering its technological capabilities required a level of funding that was beyond its reach, the Bureau decided some years ago to pool resources with the Commonwealth Scientific and Industrial Research Organisation (CSIRO) to create a better equipped supercomputing centre than either organisation could develop independently.
Despite the fact that the two organisations pool resources, supporting a facility capable of such huge processing power requires a lot of very high end equipment, and that doesn't come cheaply.
"The HPCCC (High Performance Computing and Communications Centre) at the Bureau of Meteorology is a mission-critical facility. We have a lot of very high end equipment in the facility which is quite expensive, even today [with lower hardware costs]. Because of budget pressures, we continuously have to find newer, better or smarter ways to accomplish our goals," explains Phil Tannenbaum, Manager of the HPCCC at the Bureau.
The advent of ATA-based disk subsystems presented the Bureau with the opportunity to acquire something that was almost as good as a high end fibre-channel device, but crucially, at a fraction of the price.
"We started looking at products about a year ago," says Tannenbaum, "and came up with [StorageTek's] BladeStore as the device we decided to use for pilot evaluation. The goal was that once we played with the device and found out what it was good for, we would look at deploying it and adding more of its type.
"What we have done with BladeStore is buy it as a test bed to see what it is good for. This includes things it was not specifically designed for. We expect there will be things it does well and things it doesn't do so well."
Thus far, the Bureau has found that for any given capacity, the device solves the performance requirements for a lot of applications at a much lower cost than higher-end products, with comparative high reliability, another important factor in the selection process, according to Tannenbaum.
"We were looking at the reliability of the technology so we chose a vendor known for good reliable products and who has good onsite support. Failover support was key. You need to be able to replace a failed disk without losing any data. With BladeStore there is RAID protection and a hot spare blade so you can restore data while you keep working."
To date, the Bureau has purchased an 8-terabyte, 10-blade BladeStore disk subsystem from StorageTek. This complements 10 terabytes of existing high-performance disk on its storage area network (SAN) and 2.5 terabytes of disk on an existing supercomputer.
It is currently evaluating which applications can be migrated to StorageTek's ATA-based disk subsystems. At this stage the only application that the Bureau has ruled ATA disk unsuitable for is high speed scratch disk for the supercomputer. Other applications to be evaluated include online repositories, data caches between disk and tape, scratch space for server-based applications, and mass storage for publicly accessible Web servers.
On completion of a successful ATA-based disk pilot, the Bureau plans to dramatically expand the volume of data immediately accessible on disk - improving the quantity and quality of climate research - without increasing storage budgets.
"StorageTek's BladeStore offers multiples of online data compared with what was the norm," says Tannenbaum. "Four or five terabytes of capacity for the same cost is more valuable to us than one terabyte, even with slower performance.
"By finding a more cost-effective solution we could keep data online longer and derive more value from it. The more data our researchers can use, the better the results of their models, and the faster they can get their projects finished."
The Bureau also maintains an extensive tape library, with 120 terabytes of nearline storage available via StorageTek automated tape libraries, and 120 terabytes kept as off-site copies. With the data-intensive nature of meteorological forecasting and research, data volumes are currently growing at around 50 percent per year.
"It would be valuable to have more data online that is currently on tape," says Tannenbaum. "If they were the same price we would use the BladeStore, although you will never replace tape for off-site storage."
The Bureau marked the turn of the year with a move to a brand new facility, central to which is a new supercomputer, containing around ten times more computing power than was previously available to the organisation.
"Our new facility will allow us present our daily weather forecasting models in high resolution, assimilating more data into the models, meaning the forecasts should improve. Additionally, there is ongoing climate, atmospheric and marine research going on," adds Tannenbaum.
Initially, the Bureau will make the BladeStore available to its researchers, enabling them to keep more data online than previously possible, as well as utilising it as an intermediate fast backup device for the supercomputer.
"The supercomputer is so much faster than a tape drive, and the BladeStore is sufficiently faster than a tape drive to make it a very useful backup device," says Tannenbaum. "We also have a fairly large line up of people at the Bureau who would love to have a piece of it-just about everyone who knows it exists has asked for a piece of it."