Available now priced at £43 + VAT, Intel's PRO/1000 MT desktop copper gigabit adapters are among the many gigabit desktop network interface cards (NICs) on the market. But users may find an adapter does not improve performance much unless they have a gigabit backbone, gigahertz-speed chips and fast storage devices such as Raid arrays - despite Intel aiming these adapters at high-end workstations.
Gigabit cards can attach to a LAN via standard Category 5 cabling. The MT desktop adapter is a 32bit card supporting PCI speeds of 33MHz or 66MHz. The MT can run under Windows, Linux, OS/2, various Unixes, NetWare 4.11, and Solaris 7 and 8.
Advertisement
We installed the adapter in a Dell Power-Edge 600SC server housing a 2.4GHz Intel P4 chip, with 30 clients and a variety of file sizes from 512 bytes to 64kB. The adapter was set to autosense by default to obtain gigabit performance, and file transfers were routed via an HP Procurve 5308xl switch.
This gave a throughput of 410Mbit/s under NetBench 7.02 tests, with a CPU utilisation of between 50 and 60 percent. We also checked the embedded MT adapter on the 600SC server, and achieved 425Mbit/s of throughput with a slightly higher CPU utilisation. We re-ran the test at 100Mbit/s full duplex and achieved 90Mbit/s.
Real-world
For a real-world test, we installed the MT adapter on a dual 2.2GHz Intel Xeon workstation with an 18.2GB U/160 SCSI hard disk running Windows XP. An older T series adapter was installed on a standard desktop system incorporating a single 1.33MHz AMD chip and a 30GB Ultra ATA 100 hard disk.
We carried out WinBench 99 tests to check how fast the hard disk systems were, and achieved average rates of 37MB/s for the standard hard disk and 52MB/s for the SCSI disk on the workstation. File transfers were routed through a 3Com 4250T switch. Using this setup we recorded maximum transfer rates of 190Mbit/s, with a CPU utilisation for the 1.33MHz AMD processor averaging 70 percent, while that for a single 2.2GHz Intel Xeon averaged 20 percent.
Enabling a second processor on the workstation resulted in a lower CPU utilisation but no increase in data transfer rate. Again, for comparison, we dropped the speed at which the gigabit adapters could transfer data, to 100Mbit/s, and obtained transfer rates of 85Mbit/s.
Data transfer can be improved in gigabit networks by using TCP/IP offload engines (TOEs), which free CPUs for other processing tasks and basically implement the TCP/IP stack in hardware. But firms would have to calculate whether the performance improvement justifies the costs of upgrading to gigabit technology.
Do you agree?
Have your say on this article