International law firm Stephenson Harwood recently completed a virtualised datacentre migration and upgrade that took a practical approach to reducing server traffic on its Ethernet local area network (LAN) and cutting the cost of its Fibre Channel storage area network (SAN).
The company installed two Xsigo VP560 I/O Directors to offload the majority of its server to server data traffic from its 10Gbit/s core Ethernet network as part of a move to a new facility earlier this year.
"The Xsigos sit in the middle of everything: they are connected to the SAN via FC and the LAN via Ethernet, and the host servers are connected to them [the Directors] via infiniband," said Stephenson Harwood IT director Chris Petrie.
"In the past, we have needed servers with chassis capable of hosting both Ethernet and FC network interface cards (NICs), and the Xsigos are unbelievably cheap by comparison [official retail prices start at around $20,000]."
Using I/O Directors in this way has increased server I/O performance by roughly 400 per cent, with an estimated 66 per cent reduction in storage and connectivity costs, said Petrie. The savings came largely from cutting down the number of required interface cables down from 300 to 16.
"We had to look at the build quality of the cables and cards, which is often an issue for us with some of the other things foisted upon us. We wanted the cabling layout to be as simple as possible, so we looked at the drawings of the Cat5 patch lead and FC cables before the final [purchasing] decision was made, and that alone might have sold it to us," said Petrie.
"You also get remote direct memory access [RDMA, which lets one computer transfer data into the memory of another independently of the operating system, thereby reducing processing overheads and bandwidth demands], so we might see a performance lift for memory to memory traffic, but we have not had time to quantify that yet.
"The hosts we introduce have smaller chassis and only two small NICs, so there is a powersaving, but we have to be careful because that could be offset by the Directors themselves."
Petrie evaluated a number of alternative approaches to datacentre infrastructure consolidation prior to installation, including unified server/storage platforms based on the Fibre Channel over Ethernet (FCoE) protocol designed to consolidate LAN and SAN infrastructure by forwarding Ethernet traffic over Fibre Channel networks.
"We did not see the point of FCoE: it is not a well-known system or protocol, whereas Infiniband is and the [Infiniband] controllers are well priced," said Petrie.
"FCoE is relatively new and we did not want to be its vanguard. Shortly, Xsigo will qualify quad range Infiniband connectors that will go from 20Gbit/s to 40Gbit/s and that roadmap is not there for Ethernet."
Stephenson Harwood maintains about 180 VMWare virtual servers or machines (VMs) on six physical hosts at any one time, with Compellent SAN equipment providing 155TB of raw disk capacity. However, mirroring and replication demands mean that only about 50TB is available to accommodate the firm's data.
That data is accessed by up to 460 Stephenson Harwood staff in London, and by another 200 located elsewhere. The law firm is looking to port a range of new services onto the LAN in the future, including video conferencing and IP CCTV.
"Without offloading the server traffic, we would have had to upscale the LAN infrastructure to support those, so we are hoping for further cost savings there, and to make the backup window smaller," said Petrie.
Installing the upgraded datacentre network was much less trouble than expected: the old system took only about 16 hours to decommission, move, and bring back online in April this year.
The 23-strong Stephenson Harwood IT department was aided in the project by various consultants from the key suppliers involved: Xsigo, VMWare, Compellent and Extreme Networks.
"It can often be problematic to take old kit out and put new stuff in. There are all sorts of issues that can arise when you move to a new building and have an interior refit, like particular issues with wide area network (WAN) circuit and telephony providers," said Petrie.
"We moved two SAN units and had only one SAS drive fail, which is remarkable. Whenever you turn a SAN on or off, you expect to lose a drive anyway, which is why we try never to turn them off. Overall, the move was a huge success and our end users were extremely complimentary about the quality of our work."
Sometimes, the power of the mainframe is the most cost effective answer. Computing's Peter Gothard puts Computing's readers' questions on the future of the mainframe to IBM's Z13 expert Steven Dickens.
This Dummies white paper will help you better understand business process management (BPM)