While Auckland Regional Council can see blade servers in its near future, Capital & Coast District Health Board is enthusiastically embracing the technology.
Capital & Coast is buying HP’s Proliant BL20P blade servser. Fourteen arrived in January; the DHB plans to order 12 more before June to accompany its non-blade HP Proliant 1600, 2500, ML350, ML370, DL360 and DL380 servers. The servers are presently used for file and print, email, SQL, Citrix, Oracle and the like, though the DHB plans to have fewer file and print and more Citrix servers.
Systems manager Martin Stephenson says the DHB chose blade servers over traditional servers because they offer higher density, better cable management and value for money.
“They consume less floor space, have fewer moving parts, reduce network and power cabling requirements — all of which makes them physically easier to manage. We have a centralisation model which says that all critical computing will be confined to two locations (New Plymouth, Newtown in Wellington) for disaster recovery purposes,” he says.
On the downside, Stephenson believes blade servers are less portable than traditional servers, needing a specific chassis, and he doubts they are any more scalable than other server technologies. Software remains a major issue with server performance, Stephenson saying he can get an extra 10% of users on a Citrix server by deploying cheap RAM management software.
The blade servers come fitted with Insight Management tools to predict failures — not that there have been any, he says — while in test mode. “The best part of the [server] package was the Altiris rapid deployment software,” though this is not exclusive to blade servers. “Basically Altiris allows you to deploy a fresh server image in seven minutes, which is probably the quickest it can be done. In a healthcare scenario, where time is of the essence, this has been seen as very important,” Stephenson says.
ARC CIO Tony Darby agrees with the physical improvements of blade servers. “I don’t know of many server rooms that don’t look like a dog’s breakfast as far as physical cabling is concerned.”
The regional council describes itself as “an Intel shop” with approximately 40 servers, 32-bit and 64-bit, and disk farms in varying configurations. They are used for file, print and mail functions, as well as specialist applications including intranet/database/telephony, SAP, park bookings, Oracle applications, treasury systems and geographical information systems.
The regional council has a capital replacement programme that includes moving older equipment to less power-hungry applications, and Darby says the council does a lot of work getting this balance right.
Server technology and reliability is continually improving, he says, and agrees software is often the issue. He would like to see servers developing the ability to be managed remotely through the use of PDAs and other mobile devices. Servers should also run at lower operating temperatures to reduce the need for air-conditioning and specialist server rooms, but he thinks the quest for higher processing and disk speeds may render this impossible.
Looking ahead, Darby foresees smaller server farms using blade technology and possessing disk capacities far beyond what we can currently envisage, such as terabyte capacities for single disks, with improvements in rack mounting capabilities.
“The form factors have reduced dramatically over the past five years and will continue to do so. A couple of months ago, one vendor released a four-CPU single blade server. This effectively means we could effectively collapse our five racks on to one, so things are just going to get smaller and more powerful. And what else could be improved? The obvious, of course — price,” he says.
Stephenson believes the biggest change over the coming decade will be ASP-type solutions, even if the concept has been tarnished by the burst of the internet bubble.
“There are still lots of hosting providers in the market and this sector should experience strong long-term growth. If this does occur, then it will generate an even greater requirement to lower the TCO of a server environment. Over time, I would expect that it becomes more cost-effective to scale up rather than out, which is generally in line with Moore’s Law.”