Storage-server hybrids coming into vogue

(continued from previous page)

"For business intelligence applications, it is the killer platform. It's far and away the most powerful general-purpose platform out there for this kind of work," Greenplum Chief Executive Scott Yara said in an interview. The system uses Greenplum's modified version of the PostgreSQL open-source database, called BizgresMPP, and Sun's ZFS file system, Yara said.

The system will be sold in multi-Thumper configurations, so large data sets can be scoured for information, he added. "Our software clusters together to build a very large warehouse."

The Greenplum-Sun combination also illustrates one drawback: Storage-server hybrids' unconventional design means that many companies won't know how to put them to use. As a result, many analysts expect hybrids to be popular chiefly as special-purpose appliances, in which a company combines the hardware and software so its customers don't have to figure out how.

Big Blue's hybrid ideas
IBM's foray into hybrid systems is beginning with an appliance approach. Big Blue's TotalStorage DR550 system combines a server, System p, running IBM's AIX version of Unix with a DS4000 storage system, said Clod Barrera, a distinguished engineer and chief technical strategist for IBM's System Storage group. The processors handle tasks such as indexing or deciding which data should be shuttled from fast disks to slower tape.

Eventually, such systems will be very busy handling, managing and retrieving data. "Over time it will become very, very processing intensive," Barrera said.

The company's high-end storage system, the DS8000, is a variation of its System p servers. The Power5+ processors and other features in those systems let administrators carve these "Shark" systems into separate partitions.

Today, that means a DS8000 can appear to be multiple different systems for different business units. Later, it will mean server workloads can run on the storage system

"We have not shipped that kind of function, but we believe that is something that will be done on Sharks," possibly in coming months, Barerra said.

Processing could be offloaded from more expensive mainframes, or some applications such as searching, could be boosted. "It might be cheaper to outboard that search somewhere close to the disks, so I can save myself all the ups and downs across the various software layers," he said.

There are an increasing number of storage tasks that benefit from processing, he said. Among them are encryption and enforcement of access rights to make sure only authorized people can retrieve data.

HP blades: A third way
Hewlett-Packard, which sells more x86 servers than any competitor and has a solid storage business to boot, sees a blades as the way to build hybrid products.

Because some customers need more hard drives in servers, HP moved its servers from prevailing 3.5-inch disk drives to smaller 2.5-inch drives so that more could be tucked into the same server case. But for true hybrids, the company steers customers toward its new C-Class BladeSystem chassis.

That blade server can accommodate either conventional processing blades or storage-only blades with plenty of hard drive spindles, said Dwight Barron, chief technologist for HP blades. HP therefore can offer a flexible ratio of storage to processors to meet a variety of situations--including tasks such as archiving Microsoft Exchange data that requires more processing abilities than stand-alone storage systems can supply, he said.

"We're able to mix and match the ratio of CPU blades to storage blades," Barron said. "We can take that number well above the tight range the industry has seen in the past....We believe that trend has some legs."

[mfischer2]

Brilliant!

Hey, let's put a hard drive or two in our next server.

Brilliant!

[owen.omalley]

intermixed storage and compute servers

We intermix storage and computer servers using standard hardware running Linux, because it leads to much much higher aggregate bandwidth. We are developing an open source software framework named Hadoop to help application writers use their clusters effectively. More information can be found at http://lucene.apache.org/hadoop/