This Gigabit switch proved a challenge to test. One of the limitations facing PCI is one of bandwidth. That being, it is limited to a 33MHz bus, which is 132 MB/s of available throughput. And that bus is shared. Now, not all of us are graced with wallets that can provide multiple 64-bit systems, which would allow a Gigabit network to stretch its wings. So we had to make do with the supplied 32-bit PCI 100/1,000 network cards.
The other major bottleneck which could impact performance is the hard drive. Unless running RAID 5, SCSI hard drives in each system, it is unlikely that your drives will keep up with the speed the bits of data fly across a small Gigabit LAN. So, when testing, we set up a RAM drive on each system, and ran AIDA32 with the network benchmark plug in from those RAM drives.
The results were disappointing due to the PCI bus choking the bandwidth. CPU load was high during the transfer, as it tried to cope. We managed an average rate of 38MB/s, which doesn't equate to expected performance.
Our next attempt was a real world type simulation. Let's call it a 'leech', for want of a better word. We grabbed a folder containing 14GB of video files, zipped files and compressed audio, a total of 24,971 files over 803 folders, and pumped it across a network share. DU Meter was used to time the transfer. Total time taken was 21 minutes and 9 seconds, with an average transfer rate of 27.15MB/s.
By comparison our 10/100 network offered 9.09MB/s. In other words, a Gigabit network will not give you 10 times greater performance.
Is this enough to convince the home LAN users to upgrade to Gigabit? Perhaps small LAN parties, where leeching is par for the course, will see some improvements.
Gigabit has its place, but not without the hardware to justify it. When you all have Athlon 64s in your home networks, come talk to us about Gigabit.
Issue: 133 | February, 2012