IP over FireWire Performance
IP over FireWire Performance
- Subject: IP over FireWire Performance
- From: Chris Irvine <email@hidden>
- Date: Thu, 05 Jun 2003 16:47:50 -0700
I didn't see much traffic on the FireWire list so I'll post this here.
Hopefully a couple of you would like to comment....
I realize that this module is still a preview release... but we don't let
that slow us down much. The idea of using IP over FireWire to setup SAN
style backup networks or high speed interconnects between load sharing
servers is quite interesting. The possibility of pulling down 800Mbps on a
dedicated link separate from the normal GigE link could be useful and
inexpensive. To purse these goals we setup some networking benchmarks to
evaluate IP over FireWire. Specifically, we measured throughput and related
CPU load. I'm hoping to get a few questions answered and provide
information about our experience.
__Test configuration__
two identical machines:
Dual 1.4Ghz G4 w/ FW800
Mac OS X 10.2.6 w/ IPoverFireWireB2
512MB RAM
6ft 9pin-9pin 1394b cable
18in 9pin-9pin 1394b cable
3ft 6pin-6pin 1394 cable
Direct Cat6 5ft patch cable (negotiated 1000TX, full-duplex, off-network)
Tests performed using both perftool and ttcp. All tests were TCP/IP.
__Summary__
IP over 1000T 790 bits 10^6/second
IP over FireWire400 277 bits 10^6/second
IP over FireWire800 358 bits 10^6/second
CPU utilization is also interesting:
For the sender, CPU utilization is proportional to throughput. However, for
the receiver, CPU utilization is over over twice as high when packets are
received via FireWire. Specifically, kernel_task and the system load
average climbs.
There were also strange problems when pushing massive UDP through FireWire.
The ttcp utility would error out. The same experiments worked fine on 1000T.
I'd be happy to forward my spreadsheet with details for anyone who wants it.
__Questions__
Any ideas as to why the FireWire throughput isn't higher? I've read that it
"can reserve up to 80% ... for one or more isochronous channels." Does this
mean I should never expect better than 640Mbps? Any ideas why the FireWire
receiver would consume so much more cpu in the kernel_task as compared to
Ethernet? May it be caused by differences in interrupt handling or DMA
support? I would expect Ethernet and FireWire to perform similarly in
hardware considering both PHYs connect to the U2 controller.
--------------------------------------------------------------
Chris Irvine On-line store->
http://www.tfaw.com/
Information Systems Manager phone: 503-652-8815
Dark Horse Comics, Inc.
http://www.darkhorse.com/
mailto:email@hidden spam
mailto:email@hidden
PGP Key ID: 0x0263648A
PGP F.P. 8CEF 1BC8 F763 DF79 6F38 3156 EA30 50DF 0263 648A
_______________________________________________
macnetworkprog mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/macnetworkprog
Do not post admin requests to the list. They will be ignored.