Gigabit LAN Performance

[wwaag]

Just upgraded to a Gigabit LAN. Hardware included a couple of Linksys Gigabit cards and a Netgear 8 port Gigabit switch. When I run Sandra, I only get around 34 MB/sec, which is roughly half of what others have reported. Each network card is set to 1000 Full Duplex. Any suggestions? Thanks.

wwaag

[Soopafresh]

Make sure your cabling is minimum CAT 5E ---- NOT just CAT 5

Set your cards to half duplex and test again. You should see around 200mps real thruput on Gig Ethernet

[aedipuss]

depends on what your sending and where it's coming from. most consumer hard drives can't sustain more than 45 MBs throughput, so 34 MBs is reasonable unless you are using 10,000 rpm scsi drives.

[ViRaL1]

You also might want to find out if your switch supports jumbo frames.

[wwaag]

Thanks for the replies.

Soopafresh. Discovered that one of my cables was Cat 5--replaced with new Cat 6 (all Cat 6 now). Found that there is no option for half duplex--only full duplex.

ViRaL1. Switch does support jumbo frames which is enabled in both network cards.

Results are the same. Sandra indicates 34 MB/sec on one machine and 32 MB/sec on the other.

Did some "real world" transfer tests--used a 1 GB VOB file. From my main machine writing to a Raid 0 drive, it took 53 sec which works out to be around 19 MB/sec. Writing to a "regular" HD took 67 sec, even slower.

Again, is this performance what is to be expected for a Gigabit LAN? Seems like it should it higher. Any other suggestions?

wwaag

[MOVIEGEEK]

I agree with aedipuss,I think the HD or chipset is the bottleneck.This problem is similar to firewire or USB 2.0 throughput:you will never see 50-60MB/s in the "real world".

[wwaag]

Moviegeek

HD is not the bottleneck in this case. Copied the same file to another HD on my system in 20 sec which works out to be 50MB/sec. For whatever, my network is SLOW!

wwaag

[Cobra]

I made a post about the performance of Gigabit Ethernet on another forum a while back - I was wondering how hard drives would bottleneck performance. Anyway, here it is!

Hi guys,

I just realised tonight that I actually have two machines packing gigabit Ethernet in my house. We have a big eight-port gigabit switch, and a router with 802.11g wireless and four further 10/100Base-T ports on it (one used to connect it to the switch). All our machines connect in to the switch, but they're all running at 100Mbps (except my two, which obviously run at 1000Mbps).

This test was conducted at a time when two 100Mbps machines were on the network using the Internet, alongside my two 1000Mbps machines. The first test was 700MB or so of MP3 files - lots of little files. The second was a whole DVD ISO image file.

The machines were both connected to the switch by 25m CAT6 FastCAT UTP patch cables, making 50m of cabling in all.

In the small file test, it averaged 140Mbps (17.5MB/sec). In the single file test, it averaged 280Mbps (35MB/sec).

Now, this is a strange thing. My new machine uses SATA-II hard drives which operate at 300Mbps. This can support (possibly) these transfer rates. My old machine, however, uses an original SATA drive running at 150Mbps and shouldn't be able to cope. How is this performance possible?

Charts attached. The smaller one is the small files test, and the longer one is the single file test.

[Soopafresh]

Interesting info, Cobra.


There's a useful throughput testing app that's opensource called Iperf. That particular test bypasses disk access issues.

http://dast.nlanr.net/Projects/Iperf/

[edDV]

I'm also getting ~ 200-240Mb/s max but have noticed a fair amount of CPU activity on the slower machines even to the point of a 2.4GHz P4 hitting 100% CPU when transferring large files to/from a Core2Duo over a Gigabit link. This seems to limit throughput performance.

I haven't made detail measurements yet but I don't think the hard drives are maxing out yet. The PATA drives test out around 40-55MB/s (>320Mb/sec). The USB2 drives run around half that sustained rate..

[aedipuss]

don't forget we're talking ms windows networking..... there is a lot of overhead involved. a packet is sent, looked at, inspected, checksum matched, probably packet sniffed by the firewall/virus checked by antivirus, and then an ok to send another packet is sent back. it will never be as fast as hard drive to hard drive on the same computer.