IT Week: As Hitachi Data Systems (HDS) chief technology officer, do you think the current problems in the financial markets will affect the storage sector?
Hubert Yoshida: Last quarter we saw good demand, but in the coming quarters, the financial problems will be coming to a head and this is forcing all the industry analysts to revise their forecasts. What it means for IT users is that they’ll need to consolidate more and increasingly utilise what they have more efficiently.
How should IT managers deal with this?
During the dot-com boom, we had storage growth figures of 100 per cent, but come the dot-com bust, when firms’ IT budgets fell, they were able to use storage area networking (SAN) technology to consolidate their storage. They had to rip and replace, but the savings they made through consolidation more than paid for the SAN technology. You can’t rip and replace today, you’ll need storage virtualisation to cause minimal disruption. Firms will need to combine this virtualisation with so-called thin provisioning, which allows space to be easily allocated to servers, on a just-enough and just-in-time basis.
What other techniques can increase storage utilisation?
To eliminate unnecessary and redundant files, firms will need de-duplication using tape libraries and much better archiving. Between each archive cycle only around 10 to 20 per cent of the data changes anyway, so other techniques, such as single instance store, copy on write and compression can be used to reduce the working set of data. Once that working data set is in the system, it’s much easier to control, search and back up.
Is the kind of data companies need to back up changing?
We’re seeing a big increase in the volume of unstructured data. For example, every flight an Airbus takes generates terabytes of data, which is downloaded, stored and used to trend how the planes’ subsystems are performing. That means checking fuel usage and engine temperatures, and this information helps Airbus to build safer aircraft. Some of the new radio frequency identity tags are also being deployed as sensors. People with medical problems have these tags embedded in their bodies, generating data that shows how their bodies are performing. Another example is US Homeland Security, which needs to store and archive digital photos of everyone who enters the country.
What are your views on Flash memory storage, which EMC recently announced it would be putting onto its DMX systems?
There are a number of problems with Flash memory. First, the price, which can be up to 20 times as much as spinning disks. Secondly, there’s supplier viability – there’s only a few vendors of this technology at the minute. You’ve also got to remember that Flash disks have a rewrite limitation and we need to have some diagnostics to predict when that limit could occur. For hard disks, we have self-monitoring analysis and reporting technology (Smart), but there’s nothing like that for Flash disks. What happens if you’re in the middle of a financial transaction and a write failure occurs, after a write commit? As for performance, just because it’s solid state doesn’t mean it would be any faster than a spinning disk. Remember all those interfaces between your application and the disk are slowing the data transfer rates, so it would be a good idea to benchmark your set-up.
Where do you feel is the current bottleneck for data transfer?
It will be at the storage end of the pipeline, more specifically at the storage controller, the part that funnels the data to the disk, and it’s one area where HDS has an advantage. That’s because we redesigned ours so that the processors involved have functions similar to symmetric multi-processing, which is why our performance is three to four times better than the competition.
What are the current projections for how fast data volumes are growing?
It’ll still be at an exponential rate – and with respect to personal data, there’s more than we can store. It’s the unstructured data I mentioned earlier that will grow the most. We have large customers at the minute who have around 100 petabytes [one million gigabytes] of data, but within the next years some customers will have an exabyte [1,000 petabytes] of data. The industry is now moving towards Serial Attached SCSI (SAS) drives, with Fibre Channel over Ethernet on the front end.
Career history
-Hubert Yoshida is vice president and CTO for Hitachi Data Systems (HDS), which he joined in 1997.
-Yoshida defines HDS’s technical direction, leading the company’s effort to help customers with data lifecycle requirements.
-Before joining HDS, he worked for IBM’s storage division, holding senior roles in hardware performance, software development and product management.n
