As the result of an IT failure last year, small business Internet telephony service provider Gradwell has deployed a VMware environment on HP ProLiant blade servers in a bid to maximise uptime and avoid future system downtime.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
Gradwell, known for its "Office in a yellow box" service, serves 18,000 customers and handles six million minutes of incoming and outgoing calls per month.
In January 2009, the Bath-based company experienced five hours of system downtime during peak business hours. Gradwell realised that its IT systems were struggling to cope with the business' expansion and that a new strategy needed to be put in place for the company's present and future needs.
Gradwell now relies heavily on server virtualisation -- VMware ESX with virtualised storage on dedicated HP networked storage servers.
I can put in an HP server blade in 10 minutes. Racking and cabling a server took an hour and a half before. Now we only have 14 cables.
Alex Howells, senior system administrator at Gradwell,
At the beginning of 2009, however, the company began experiencing issues with two of its networked storage servers. This was due to faulty hardware and pushing the storage gear harder than it could manage, thus resulting in several instances of system downtime.
During each system downtime, Gradwell had to recover services by resetting all of the ESX servers and storage systems. Rebooting these took several hours, as they needed to be recovered in the correct sequence. The system downtime left many of the company's customers unable to use Gradwell's services for long periods of time.
Gradwell turned to HP for help. The vendor was chosen to build a hosting platform that could be easily scaled. The new platform included HP BladeSystem technology with HP ProLiant BL490c G6 servers and HP Virtual Connect Flex-10. In addition to replacing its three HP DL380 servers, Gradwell's third-party storage, which consisted of three Infortrend iSCSI SAN arrays, was also replaced with HP LeftHand P4000 SANs. Instead of using iSCSI on the new storage servers, Gradwell switched to NFS, as the company deemed it more reliable. Gradwell also upgraded from VMware Infrastructure 3 to of vSphere 4.0.
Peter Gradwell, founder and managing director of Gradwell, said customer confidence dipped after the system failure.
"With VoIP, if you lose any audio or there's a delay in the service, people get annoyed. VoIP call revenue dropped from £86,000 a month to £73,000 in January, when the system failure occurred -- a loss of £13,000. After the HP implementation, by June 2009, call revenue had managed to climb to £94,000 a month," said Gradwell.
Payback in less than one year
Gradwell claims to have experienced a £20,000 monthly revenue increase since the HP blade system deployment, in addition to saving £50,000 a year in freeing up staff time to work on more valuable IT projects. One of those projects is the planned launch of a new VoIP platform from Gradwell in the near future.
Power cost reductions were also achieved -- £3,300 a month -- due to the HP ProLiant BL490c G6 server blade. It costs Gradwell £95 a month to power and cool. One HP ProLiant BL490 G6 server blade is now doing the work of the three HP ProLiant DL380s that the company had previously been using.
Gradwell now has 36 GB of RAM on the G6 server blade, enabling it to run 80 to 100 virtual machines (VMs), compared to the previous 25 to 30 VMs on a DL380 with 16 to 20 GB of RAM.
Christian Keller, vice president EMEA of HP, said HP blade servers may be considered expensive to shell out for upfront; quick ROI from a decrease in power consumption, however, makes the purchase of blades well worth it.
VoIP call revenue dropped from £86,000 a month to £73,000 in January when the system failure occurred -- a loss of £13,000. After the HP implementation, call revenue climbed to £94,000 a month.
Peter Gradwell, founder and managing director of Gradwell,
"One feature, for example, is the use of fans on HP blade servers. They have sensors for airflow,so can be turned on or off in addition to finding hotspots. If you undergo a big project, at a certain time of every year, the sensors will know that more heat is being generated and therefore will turn the fans on near the hotspot," he said.
He also advised that blade servers aren't for everyone, explaining that the need for three or four servers will make blade a more efficient choice. If you only need one blade server, though, blades are not for you.
HP Virtual Connect Flex-10
Gradwell also chose to use HP Virtual Connect with Flex-10 10Gb Ethernet Module technology, to avoid anymore system downtime. The company said it saw a reduction in cabling costs due to this technology.
According to Alex Howells, senior system administrator at Gradwell, the chosen technology has provided faster time to value: "I can put in an HP server blade in 10 minutes. Racking and cabling a server took an hour and a half before. Now we only have 14 cables."
He said using the HP BladeSystem c7000 Enclosure enables him to roll out LAN and SAN profiles to blade slots in advance, which makes switching, deploying a server or setting up a vLAN faster and more consistent.
Gradwell had previously been using HP ProLiant servers when it upgraded from white-box servers two years before.
When the company decided to replaced its third party Infortrend storage with HP LeftHand P4000 SANs, Gradwell claims data performance was improved. Previously, Gradwell said its SAN had throughput of four to five megabits per second, at peak times. With the HP LeftHand P4000, SAN throughput has been increased more than 10-fold to 70 megabits per second.
According to Howells, this has made it easier for him to manage future workloads. The HP LeftHand P4000 SAN shows him utilisation throughput and other statistics, making it simpler for him to carry out capacity planning.
"I can get more disk space before capacity becomes a problem. With our old system, we didn't know there was a problem until it literally started to slow down," he added.
Howells said, due to the company previously having iSCSI SAN and SATA disk drives, there were several single points of system failure that needed to be addressed, so the company also chose to switch to SAS. The company purchased five additional HP storage servers to take the faulty iSCSI server out of service and reduce demand on the others.
The IT team at Gradwell tested the SANs in case of failure, as a direct result of their experience at the beginning of 2009. They have the two SAN systems in a cluster, with replication and failover capabilities included within them.
Howells said he set up each system so it has its own rack and own power supply. According to Howells, he pulled out one or both power supplies from the SANs, pulled out disks to simulate RAID crashes, yanked out network cables, and anything possible to test anything what could go wrong. He still didn't experience any system downtime.
Gradwell said the company considered several other vendors for the job, including NetApp and Dell/EqualLogic, in addition to HP and LeftHand products. It ended up, however, finding NetApp more expensive when factoring in all the features the company required.
Gradwell said due to Dell/EqualLogic servers being sold in isolation in the UK, instead of as a solution, the company was concerned about how the technology would work with others in its data centre and whether this would cause more unwanted system downtime.
Kayleigh Bateman is the Site Editor of SearchVirtualDataCentre.co.uk.