Gino Ruopolo wrote:
I have attempted to measure CPU overhead due to inter-zone networking using various tools, including DTrace and an internal network benchmark tool. Any overhead was so small that it could not be measured, and was always less than 1%.

Could be present a "very high" network latency compared to the single zone setup? Anyway I can assure that switching from 2 zone to a single zone decrease system cpu% by 18-20%.
No paging. No resource control.

Network performance can be characterized by latency and bandwidth metrics, and both are useful. Here's one simple method to measure throughput:

1) In one zone (zoneA), create a sufficiently large file in /tmp.
2) In zoneA, use ftp to send the file to /tmp in the same zone (zoneA). You can use the IP address of the zone, or the loopback address ftp will display the total throughput.
3) In zoneA, use ftp to send the file to /tmp in another zone (zoneB).

Here are my results of that test, using a 75MB file of random data in /tmp, using a 64-bit Athlon laptop:

1. Global zone to global zone using 125 Mbytes/s
2. Global zone to global zone using 192.x.y.z: 124 Mbytes/s
3. Global zone to non-global zone:             124 Mbytes/s
4. Non-global zone to itself:                  118 Mbytes/s
5. Non-global zone to another non-global zone: 122 Mbytes/s

CPU utilization did not vary much during those tests.

I would not read too much into those values, but I think that it is safe to conclude that there is little difference in network performance among those situations. The two situations you have described are (4) and (5).

I have measured network latency in similar situations, using DTrace, and found similar results.

If we can rule out network performance, we must look elsewhere. Would you send the zonecfg info for the two zones you have been using?

Also, how many zones are running on this system?

Jeff VICTOR              Sun Microsystems            jeff.victor @
OS Ambassador            Sr. Technical Specialist
Solaris 10 Zones FAQ:
zones-discuss mailing list

Reply via email to