50% ping on dual 450

theed

Registered
I feel like I mentioned this long ago, but I never got an answer. I'm running 10.1 on a dual 450, and if I flood ping anything else on the network, my machine states that I'm getting back 50% of the pings. All other machines report ~100%. tcpdump shows what looks like 100%.

I also occasoinally (still) get a 1 second error on ping times. I'm on a 100Mbit network switched per port, my actual ping times are ~3ms. ping will report 1003ms. Anyone else out there still getting issues with stuff like this? It makes me think that the multiprocessing code isn't quite straight.
 

~vert

It's short for vertigo...
I don't have an answer to your problem..., sorry. I just wanted to state that I have a G4 DP450 and I have had no problems with packet loss on my network or through the DSL line. I don't think the problem resides in the double processors - at least not by themselves.
 

theed

Registered
I'm not actually having any problems, and I'm not actually losing packets, it just seems to be ping that is getting all confused. It's disturbing though when a test utility misbehaves.
 

~vert

It's short for vertigo...
ah..., sorry for the misunderstanding. I just tried pinging my network and a few internet sites and I am getting 100% return, so I don't have the symptom of ping-weirdness (to coin a phrase) of good connection but bad numbers that you are seeing.

Mighty vexing, though.
 

soellman

nice face!
Originally posted by theed
I feel like I mentioned this long ago, but I never got an answer. I'm running 10.1 on a dual 450, and if I flood ping anything else on the network, my machine states that I'm getting back 50% of the pings. All other machines report ~100%. tcpdump shows what looks like 100%.

I also occasoinally (still) get a 1 second error on ping times. I'm on a 100Mbit network switched per port, my actual ping times are ~3ms. ping will report 1003ms. Anyone else out there still getting issues with stuff like this? It makes me think that the multiprocessing code isn't quite straight.
maybe it's your networking hardware, I just did a flood ping to another box on my switch and I get 100% from my dual 450.
cheers
 

theed

Registered
I hereby withdraw my complaint of the multiprocessing code. But I defend my honor on the grounds that I was given statistically unlikely data.

warning, this post is worthless and irrelevant to virtually anyone whe might want to read it. You have been warned.

I am using a 100MHz PPC, Mac OS 9.1, and IPNetRouter as my router in here, connected at 100Mbit to my 100Mbit switch to my Gbit capable dual 450 G4. On pinging stuff through that router, or the router itself, I get precisely 50% packet loss (flood ping) the farthest deviation I have gotten is 48% loss. From other machines I was recieving 0% packet loss. I guess this was the only machine running fast enough to challenge the router.

Now with my laptop on 10.1, it gets wierdness, but the loss fluctuatues between 20% and 40% packet loss. It seems that 10.1 is fast enough on a G3 233 to challenge my feeble router as well.

So the issue is not multiprocessing confusion so much as 10.1 outperforming Mac 9 on networking to the degree that things formerly working now seem broken. However, I still occasionally get the ping miscount error of 1 second interval pings telling me times that are precisely 1 secand longer than their transmission. this is not multiprocessing specific.
 

soellman

nice face!
Originally posted by theed
I hereby withdraw my complaint of the multiprocessing code. But I defend my honor on the grounds that I was given statistically unlikely data.

So the issue is not multiprocessing confusion so much as 10.1 outperforming Mac 9 on networking to the degree that things formerly working now seem broken. However, I still occasionally get the ping miscount error of 1 second interval pings telling me times that are precisely 1 secand longer than their transmission. this is not multiprocessing specific.
if you're doing a flood ping across a router, strange things like that may well happen because most routing code will drop icmp first when faced with a saturated link.. I dunno if this is the case in your situation, or what the router code is link under os9..

but in any case, I always suspect dual proc support when I have strange things going on with my machine, and I'm usually right :) dualies will have issues that many developers will not encounter when they develop software on a uni proc box like threading issues and cpu deadlocks..
 
Top