On Sun, Sep 28, 2014 at 03:00:35PM +0000, Edward Ned Harvey (lopser) wrote: > > From: Derek Balling [mailto:[email protected]] > > Sent: Sunday, September 28, 2014 10:46 AM > > > > Wild-ass Speculation: With Amazon's commitment to cloud computing, and > > with SDN use on the rise... is it possible they're rebooting "virtualized > > network gear"?
> Extremely possible. But that ... might not? ... explain the > wildly variable network performance during times when everything > is up and operational, yet horribly performant. (Such as > downloading/uploading files, sometimes going 30Mbit and > sometimes 15KB/sec, such as ping response being sometimes 20ms, > and sometimes 800ms or straightup timeout). Hi Edward, I feel your pain. Seems like you need more data. sar could show the server perspective on what was happening on the hardware over time. If the link is bouncing, congestion, retransmissions, long wait times for disk, etc. sar is usually configured to keep 7 days of history. traceroute to your workstation could be run as a cronjob for data on how the intermediate nets behave. clink could evaluate node congestion but it's more effort and typically takes hours to collect data. Wish I knew a metric to expose CPU cycles you didn't get because some other VM was sucking them up. Maybe L2 cache misses or TLB misses would be revealing. Maybe low CPU utilization when you should see it getting busy? HTH, -- Charles Polisher _______________________________________________ Tech mailing list [email protected] https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
