Expected Throughput For PB 500 Series Ethernet Transceiver?

[Paralel]

I have a 45 Mbps connection, and I'm hardwired to the modem with my 540c on System 7.1.1 & OT 1.2, but I seem to be maxing out at ~540 Kbps downstream. Even for a 10Mbps Ethernet connection, that's pokey.

Are there any known bottlenecks that could be throttling me with regard to the 540c, System 7.1.1, Open Transport 1.2, or the Farallon Ethernet transceiver I'm using?

Any insight would be appreciated.

 

[cb88]

Are you sure you didn't misread KBps as Kbps? ... If so that would probably be in the right range although still only half theoretical 10Mbit ethernet performance. If your connection really is that slow something funny is going on... perhaps you could try sticking a 10Mbit hub between your network and your transceiver or I would imagine a 10Mbit switch would be even better.

 

[Anonymous Freak]

How were you measuring it? I've gotten just shy of 1 MB/s when downloading a file to a flash memory card in the PCMCIA slot. But the spinning hard drive can't keep up with full 10 Mbit Ethernet speed.

 

[Paralel]

I am sure my units are correct.

The system does have an original SCSI drive in it.

I'm measuring it by the download rate as reported for my browser for large files. The rate is accurate given the time it takes to download.

 

[Anonymous Freak]

It's your hard drive, then. The hard drive can't keep up with it. How much RAM do you have? If you have plenty of RAM, you can turn the disk cache way up.

 

[Paralel]

I never realized those period SCSI drives were so slow.

It has maxed out RAM, 36 MB total.

Good thought on the disk cache. I think I'll give that a shot and see if it increases throughput.

I'll really have to try and nab one of those PCMCIA cages.

 

[Gorgonops]

Somehow I can't believe that the SCSI drives in those Powerbooks can't do better than ~540 Kbps. The 10MB hard drive shipped in the original IBM PC XT was capable of around 80KB/s (ie, 640Kbps) when paired with the original controller and formatted at a 6:1 interleave. (It improved to around 100KB/s if you reformatted as 5:1. Replace the Xebec controller with a good Western Digital that could do 3:1 and you could push 200KB/sec.)

If they really are that slow the SCSI subsystem/driver software must really, really, seriously suck.

 

[Paralel]

Is there any software I can use to check the read and write rate on the drive?

 

[cb88]

I doubt the SCSI controller has DMA... if it is anything like the other macintoshes.

That would put a massive hurting on performance. Also don't forget that SCSI-1 is only 5Mbps and SCSI-2 is only 10Mbps theoretical max bandwidth... I woudln't be supprised if the drive maxes out at 500Kbps perhaps you could try a drive benchmark.

http://www.ikasu.net/lombard/hdd/scsi/20-500m/

500Kbps is roughly 60KBps... so perhaps a few random IOs are slowing down the drive access as well? Otherwise thats quite slow given those benchmarks which help put things in perspective.

 

[NJRoadfan]

The RamFAST SCSI card in my Apple IIgs can push over 500kb/sec with DMA enabled and an old Apple 40SC connected to it (the card also has 256k cache on it). Theoretical max transfer speed on that machine is 1MB/sec (8bit bus running at 1Mhz). Somehow I think a machine powered by a 32bit CPU/bus could do better even if it was an older SCSI chip. It could also be the Ethernet controller requiring a bit of CPU time to handle packets combined with other delays.

 

[Gorgonops]

Idon't forget that SCSI-1 is only 5Mbps and SCSI-2 is only 10Mbps theoretical max bandwidth

You're confusing little b with big B. (Note that as a general rule parallel busses are usually measured in bytes, not bits.) SCSI-1 is 5 MBYTES per second. Technically most older Macintosh SCSI implementations are limited to a maximum theoretical speed of around 3MB/sec because they use the slower byte-at-a-time asynchronous mode, because of said lack of DMA. In some of the really cruddy implementations like in the Mac Plus, other limitations slow it down to only a few hundred K per second.

Nonetheless that's faster that 500kb/s. Heck, by coincidence 500kps is what the Macintosh *floppy drive interface* operates at. Machines much older than a Powerbook 500, like an SE/30, are able to push more than 1MB/sec with a fast enough drive.

(Of course, it should be easy enough for the OP to benchmark his machine.)

Theoretical max transfer speed on that machine is 1MB/sec (8bit bus running at 1Mhz).

Nitpicky I know but a 6502 bus at 1 Mhz can only transfer about 500K a second. (That would be the CPU sucking NOPs or DMA. PIO would of course be slower.) The bus has a 50% duty cycle.

(EDIT, again: Well, actually, apparently the IIgs *can* run the bus at a full duty cycle. From the Apple docs:

Features and benefits include:

*  DMA Data Transfer -- on the Apple IIGS computer, data can be transferred at
a rate of up to 1MB per second.  On the Apple IIe computer, data can be
transferred at a rate of up to .5MB per second. (...)

So I was right and I was wrong.)

 

[NJRoadfan]

Yup, the RamFAST specifically states it is DMA capable. There is even an option to switch DMA off as it is incompatible with some revisions of the TranswarpGS accelerator. Whats interesting is the card readily outperforms the CFFA3000 despite the later having ultra fast solid state storage.

 

[cb88]

Heh... I would go and do just what i suspected someone else of doing wouldn't it ;-)

 

[Paralel]

I tested the HD, it's rate is significantly higher than what I am getting on Ethernet downstream, so that isn't the bottleneck.

I also noticed that when it hits its peak of ~540kbps downstream I start getting numerous collisions according to the transceiver. It's connected directly to my modem, which has an integrated 4-port switch in it. Being on a switch, connected directly to the modem, where would the collisions be coming from?

 

[techknight]

the collision is probably because the switch is delivering the data faster than your Ethernet processor can handle the datarate is my guess, if the buffer overflows or the overhead is too large for the current transport, collisions happen.

Yea, the link is 10Mbps, but that doesnt mean the rest of the hardware can buffer, process, and handle the data stream at that exact rate. A good example of this is the arduino, and the wiznet ethernet controller. its stable at 100Mbps, but that doesnt mean the processor can handle 100mbps data rate.

 

[Paralel]

The buffer on this old Ethernet hardware is hardware based, right? Probably nothing I can do about it then.

I wonder how Anonymous Freak sustained such a high downstream rate on his machine?

 

[Gorgonops]

So... just out of curiosity, is your AAUI transceiver one of those ones that has two RJ-45 ports on it? Those essentially have a two port hub built into them. (They were designed so "Mac People" who were used to LocalTalk-style daisy chaining could sub Ethernet in the same geometry. It's actually a pretty stupid idea and violates a couple rules of Ethernet network design but it did work after a fashion. The drawback of them is they by definition mean your onboard Ethernet chip behind the AAUI port can't operate in full duplex mode even if it's capable.)

 

[Paralel]

Nope. Single RJ-45 jack.