Very Computer

by **John Smit** » Mon, 24 Jan 2005 04:42:26

Hello All,

I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
Ethernet. The V480 talk to each other over the Catalyst. My application is
complaining about excessive latency between the two 480's.

The problem is that this latency problem (to the order of 45 seconds for a
message to reach from one 480 to another) only happens once in a while. For
most part the data flow is very fast. Now, my question is this. Is there a
utility that I can use to monitor network latency between the two machines?
Something that will perhaps measure the roundtrip delay between the two
machines.

I played around with ntop but I couldn't get it to work (I didn't spent too
much time on this either), and I realized that ntop is not meant for
measuring latencies. Unless I am wrong. I also used netperf but it measures
the throughput.

Regards,
JS

by **Barry Margoli** » Mon, 24 Jan 2005 12:12:18

Quote:> Hello All,

> I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
> Ethernet. The V480 talk to each other over the Catalyst. My application is
> complaining about excessive latency between the two 480's.

> The problem is that this latency problem (to the order of 45 seconds for a

Did you really mean "45 seconds", not something like 45 milliseconds? I
can't imagine any networking device taking that long to deliver a
message -- not even a satelite transmission.

Quote:> message to reach from one 480 to another) only happens once in a while. For
> most part the data flow is very fast. Now, my question is this. Is there a
> utility that I can use to monitor network latency between the two machines?
> Something that will perhaps measure the roundtrip delay between the two
> machines.

ping

--

Arlington, MA
*** PLEASE post questions in newsgroups, not directly to me ***

by **Filippos** » Mon, 24 Jan 2005 13:59:19

Smokeping
http://people.ee.ethz.ch/~oetiker/webtools/smokeping/

If it is indeed 45 seconds then I suggest you:

a) install latest drivers/updates and/or look for bugs
b) check cables
c) if everything else fails, contact Sun

The V480 is not a *brand* new machine and a serious problem like this
(if it really exists) would be fixed by now.

HTH,

Filippos

> Hello All,

> I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
> Ethernet. The V480 talk to each other over the Catalyst. My
application is
> complaining about excessive latency between the two 480's.

> The problem is that this latency problem (to the order of 45 seconds
for a
> message to reach from one 480 to another) only happens once in a
while. For
> most part the data flow is very fast. Now, my question is this. Is
there a
> utility that I can use to monitor network latency between the two
machines?
> Something that will perhaps measure the roundtrip delay between the
two
> machines.

> I played around with ntop but I couldn't get it to work (I didn't
spent too
> much time on this either), and I realized that ntop is not meant for
> measuring latencies. Unless I am wrong. I also used netperf but it
measures
> the throughput.

> Regards,
> JS

by **Filippos** » Mon, 24 Jan 2005 14:03:35

I forgot to mention, you can "watch" for network traffic as well, with
a sniffing tool such as "snoop" on the Solaris hosts, and from within
the Cisco router's telnet interface. This way you can somehow eliminate
possible physical problems such as cables and gain a better insight of
the problem

by **Richard B. Gilber** » Mon, 24 Jan 2005 14:52:21

>Hello All,

>I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
>Ethernet. The V480 talk to each other over the Catalyst. My application is
>complaining about excessive latency between the two 480's.

>The problem is that this latency problem (to the order of 45 seconds for a
>message to reach from one 480 to another) only happens once in a while. For
>most part the data flow is very fast. Now, my question is this. Is there a
>utility that I can use to monitor network latency between the two machines?
>Something that will perhaps measure the roundtrip delay between the two
>machines.

>I played around with ntop but I couldn't get it to work (I didn't spent too
>much time on this either), and I realized that ntop is not meant for
>measuring latencies. Unless I am wrong. I also used netperf but it measures
>the throughput.

>Regards,
>JS

ntpd will measure network delays if you set one system up as a server
and the other as a client.

I don't think measuring network delays is going to help you much
though. I suspect you have either serious hardware problems, software
problems or both. What sort of retry timeouts come into play if you
have a dropped packet? Are your patch cables, plugs, sockets, etc,
rated for gigabit speeds?

Are the problems only between these two machines? Does either machine
have problems talking to the rest of the world?

by **bart.smaald..** » Mon, 24 Jan 2005 15:34:37

I have seen weird problems when some machines plumb and unplumb
interfaces; these trigger spanning-tree calculations in the router and
it's
out to lunch for a while. 45 seconds seems like a long time though.
See if it correlates to machines going up and down...

- Bart

by **Gregory Toome** » Mon, 24 Jan 2005 16:30:24

> Hello All,

> I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
> Ethernet. The V480 talk to each other over the Catalyst. My application is
> complaining about excessive latency between the two 480's.

> The problem is that this latency problem (to the order of 45 seconds for a
> message to reach from one 480 to another) only happens once in a while.
> For most part the data flow is very fast. Now, my question is this. Is
> there a utility that I can use to monitor network latency between the two
> machines? Something that will perhaps measure the roundtrip delay between
> the two machines.

> I played around with ntop but I couldn't get it to work (I didn't spent
> too much time on this either), and I realized that ntop is not meant for
> measuring latencies. Unless I am wrong. I also used netperf but it
> measures the throughput.

> Regards,
> JS

Could be a reverse dns timeout. They manifest themselves in strange ways.

gtoomey

by **Colin B** » Wed, 26 Jan 2005 02:53:50

> Hello All,

> I have two Sun V480s connected to a Cisco Catalyst 6500 via Gigabit
> Ethernet. The V480 talk to each other over the Catalyst. My application is
> complaining about excessive latency between the two 480's.

Duplex. Check your speed and duplex, and make sure that they're all either
autonegotiating properly, or being forced to the same thing everywhere.
Check for network errors with netstat -i.

by **Rick Jone** » Wed, 26 Jan 2005 09:06:23

> I also used netperf but it measures the throughput.

The _default_ test type in netperf is a TCP_STREAM test that indeed is
a througput test. However, there is also the TCP_RR test which is a
test very much like ping, only from application to application.

$ netperf -t TCP_RR -H <host>

is the simplest version of that. It will report average transactions
per second over the test interval. You can also use the -r option to
set the request and/or response size:

$ netperf -t TCP_RR -H <host> -- -r 128,256

will send 128 byte requests and receive 256 byte responses. A number
with no comma sets both to the same value. The default is one byte
each way.

If you compile netperf with -DHISTOGRAM it will keep a "logarithmic"
histogram of individual round-trip times and display that histogram if
you set the verbosity level to two. Here is an example of a test
between a machine in Palo Alto and one in Cupertino just to make it
more interesting:

TCP REQUEST/RESPONSE TEST to tardy.cup.hp.com : histogram
Local /Remote
Socket Size Request Resp. Elapsed Trans.
Send Recv Size Size Time Rate
bytes Bytes bytes bytes secs. per sec

16384 87380 1 1 10.00 926.78
32768 32768
Alignment Offset
Local Remote Local Remote
Send Recv Send Recv
8 0 0 0
Histogram of request/response times

UNIT_USEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
TEN_USEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
HUNDRED_USEC : 0: 0: 0: 0: 0: 0: 0: 0: 1540: 2823
UNIT_MSEC : 0: 4759: 123: 24: 2: 1: 0: 0: 0: 0
TEN_MSEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
HUNDRED_MSEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
UNIT_SEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
TEN_SEC : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0

Quote:>100_SECS: 0

HIST_TOTAL: 9272

Of course, it won't necessarily tell you what is going on, just that something
was going-on with that test at that time.

happy benchmarking,

rick jones
--
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...

Very Computer

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency

Network Latency