Slow versus fast VAXen

[John Klos]

One more thing - what is different between the 4000/30 and the 4000/60
with regards to timing? Both are running netbsd-6, but the 4000/30 is
showing ping times in increments of 10 ms:

claire: {6} ping google.com
PING google.com (74.125.224.99): 52 data bytes
64 bytes from 74.125.224.99: icmp_seq=0 ttl=57 time=20.000 ms
64 bytes from 74.125.224.99: icmp_seq=1 ttl=57 time=30.000 ms
64 bytes from 74.125.224.99: icmp_seq=2 ttl=57 time=30.000 ms
64 bytes from 74.125.224.99: icmp_seq=3 ttl=57 time=30.000 ms
64 bytes from 74.125.224.99: icmp_seq=4 ttl=57 time=30.000 ms
64 bytes from 74.125.224.99: icmp_seq=5 ttl=57 time=40.000 ms
64 bytes from 74.125.224.99: icmp_seq=6 ttl=57 time=30.000 ms
64 bytes from 74.125.224.99: icmp_seq=7 ttl=57 time=30.000 ms

A 4000/60, on the other hand... but it jumps to 200 ms when the network is
completely idle:

gaia: {1} ping google.com
PING google.com (74.125.224.133): 52 data bytes
64 bytes from 74.125.224.133: icmp_seq=0 ttl=56 time=108.149 ms
64 bytes from 74.125.224.133: icmp_seq=1 ttl=56 time=189.728 ms
64 bytes from 74.125.224.133: icmp_seq=2 ttl=56 time=37.478 ms
64 bytes from 74.125.224.133: icmp_seq=3 ttl=56 time=210.956 ms
64 bytes from 74.125.224.133: icmp_seq=4 ttl=56 time=51.152 ms
64 bytes from 74.125.224.133: icmp_seq=5 ttl=56 time=56.896 ms
64 bytes from 74.125.224.133: icmp_seq=6 ttl=56 time=201.141 ms
64 bytes from 74.125.224.133: icmp_seq=7 ttl=56 time=201.070 ms
64 bytes from 74.125.224.133: icmp_seq=8 ttl=56 time=181.716 ms
64 bytes from 74.125.224.133: icmp_seq=9 ttl=56 time=200.309 ms
64 bytes from 74.125.224.133: icmp_seq=10 ttl=56 time=200.432 ms
64 bytes from 74.125.224.133: icmp_seq=11 ttl=56 time=200.096 ms
64 bytes from 74.125.224.133: icmp_seq=12 ttl=56 time=200.038 ms


Hmmm... what's also strange is that another 4000/60 on the same ethernet
segment as an OS X box is showing very different ping times than the OS X
machine (the OS X machine is pinging the VAX while the VAX pings the OS X
machine):

maia: {3} ping6 chococat
PING6(56=40+8+8 bytes) 2001:470:a068:2a::90 --> 2001:470:a068:2a:6233:4bff:fe0c:6d3e
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=0 hlim=64 time=3.236 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=1 hlim=64 time=198.183 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=2 hlim=64 time=198.052 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=3 hlim=64 time=249.127 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=4 hlim=64 time=5.438 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=5 hlim=64 time=197.265 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=6 hlim=64 time=197.565 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=7 hlim=64 time=197.519 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=8 hlim=64 time=197.493 ms
16 bytes from 2001:470:a068:2a:6233:4bff:fe0c:6d3e, icmp_seq=9 hlim=64 time=197.556 ms


OS X machine:

Chococat:~ john$ ping6 maia
PING6(56=40+8+8 bytes) 2001:470:a068:2a:6233:4bff:fe0c:6d3e --> 2001:470:a068:2a::90
16 bytes from 2001:470:a068:2a::90, icmp_seq=0 hlim=64 time=2.461 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=1 hlim=64 time=1.708 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=2 hlim=64 time=1.709 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=3 hlim=64 time=1.729 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=4 hlim=64 time=1.679 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=5 hlim=64 time=2.119 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=6 hlim=64 time=1.779 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=7 hlim=64 time=4.386 ms
16 bytes from 2001:470:a068:2a::90, icmp_seq=8 hlim=64 time=1.650 ms


Hmmm...

John

[Anders Magnusson]

Post by John Klos
One more thing - what is different between the 4000/30 and the 4000/60
with regards to timing? Both are running netbsd-6, but the 4000/30 is

Hm, it seems to be a bunch of things that may interact here.

Some VAXen do not have a fine-granularity timer, so the lowest time
difference is 10ms. MV2 is one, and obviously the 4000/30.

The 4000/60 has a better timer, but something seems to be wrong, either
in the use of the timer, or somewhere else. The near-200ms delay may be
a hint on what to search for.

-- Ragge

[Toby Thain]

Post by Anders Magnusson

Post by John Klos
One more thing - what is different between the 4000/30 and the 4000/60
with regards to timing? Both are running netbsd-6, but the 4000/30 is

Hm, it seems to be a bunch of things that may interact here.
Some VAXen do not have a fine-granularity timer, so the lowest time
difference is 10ms. MV2 is one, and obviously the 4000/30.

Hm, my MVII reports down to about 3ms, which is, I believe,
approximately the context switch time.

--Toby

Post by Anders Magnusson
The 4000/60 has a better timer, but something seems to be wrong, either
in the use of the timer, or somewhere else. The near-200ms delay may be
a hint on what to search for.
-- Ragge

[Matt Thomas]

That's because the 30 doesn't have a register to read to get the sub-ms time
where are the 60 and 90 do.

As for the 200ms, that sounds like interrupt mitigation (or a bug) in the ze driver.