Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

FNN vs GigabitEther & Myrinet

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Patrick Geoffray patrick at myri.com
Tue Oct 23 19:50:09 PDT 2001


Hi,

Velocet wrote:

> (These questions apply to myrinet as well, obviously noting that its
> highly optimized and better performance than GbE but also more expensive.)

I am always surprised by this statement. It's natural because the GigE NICs 
are much (MUCH) less expensive than Myrinet's ones. However, the total 
price per node is another story for medium/large configuration: GigE switches 
are still (too) expensive for 32+ ports. I don't know if there is a technical 
reason or if the market is large enough to keep a high price tag.

> In the ArsTechnica article, the bandwidth per node of having 3 or 4
> fast Ether NICs in each node was said to be 'just as fast' as a single
> GbE NIC onboard. Obviously the GbE nic will not operate at 100% efficiency,
> which is related to the previous thread I just posted on.

I have never seen a GigE NIC pushing ~1 Gb/s. It's hard to push so much 
with IP, as the IP stack becomes quickly the real bottleneck. To reach the 
full hardware performance, you need an OS-bypass protocol, or a zero-copy 
IP stack. Folks from the project project at Duke used a such IP stack in 
FreeBSD to reach 2 Gb/s TCP over Myrinet. Without this special kernel and 
driver, the same hardware provides 1.2 Gb/s TCP with the Linux 2.4 IP stack.
It was way less efficient with 2.2 kernels.

> I was also wondering if there arent problems with having 4 or 5 NICs on
> a PCI bus - dont you run into shared IRQs which may result in delays on
> each node handling incoming traffic?

5 devices on the same PCI starts to be a problem: high number of interrupts, 
long arbitration loop on the PCI, short DMA transactions. It's OK with Fast 
Ethernet NICs as they are slow anyway.

> Obviously any bandwidth problems with the PCI bus itself will affect a GbE
> NIC too. 32bit 66MHz is technically 2Gb/s+ but you dont get that to
> the cards direclty Im sure. Can regular PCI even fill a GbE network?

I don't know any 32 bits/66 MHz PCI buses. It's usually 32/33, 
64/33 and 64/66. 32/33 has a peak of 1 Gb/s, so it's enough 
to fill up a GigE link, on the paper.

Regards

Patrick 

----------------------------------------------------------
|   Patrick Geoffray, Ph.D.      patrick at myri.com 
|   Myricom, Inc.                http://www.myri.com
|   Cell:  865-389-8852          685 Emory Valley Rd (B)
|   Phone: 865-425-0978          Oak Ridge, TN 37830
----------------------------------------------------------




More information about the Beowulf mailing list