...
Host | State | last seen | Status |
---|---|---|---|
pinger.cern.ch | Pingable, web server does not respond. Sent email. Bad disk. Working again 9/17/2018. | 9/11/2018. | Fixed 9/17/2018 |
pinger.cs.ubru.ac.th | Access to http://pinger.cs.ubru.ac.th/cgi-bin/ping_data.pl from SLAC forbidden starting October 1, 2018. Charnsak found that the raspberry had been moved. He restored it to working order 10/4/2018. | 9/30/2018 | Fixed 10/4/2018. |
nsi.ampath.net | Unable to gather data. Responds to ping and http://nsi.ampath.net/cgi-bin/ping_data.pl, however, no new data is available. Site responded and it is working again as of 10/3/2018. | 9/24/2018 | 9/30/2018, reminder 10/3/2018. Fixed 10/3/2018. |
pinger.stanford.edu | Unable to ping the host. Due to building power maintenance. | 9/21/2018 | 10/3/2018 Fixed |
pinger.daffodilvarsity.edu.bd | Unable to ping host, email sent 10/14/2018 | 10/8/2018 | |
pingersonar-um.myren.net.my | No response | 6/26/2018 | |
pinger.isra.edu.pk | Down, it came up Sep 2-4, 2018. It is pinging, however, all the targets are not responding after 31 tries, email sent to Wajahat 9/18/2018. As of 10/14/2018 it is not pingable. | March 6, 2018 | |
pingeramity.in | It has been working since 28th July. It is unclear how stable it is. It is down again 9/5/2018. Worked again from 9/24/2018 to 9/28/2018. Emailed Amity requesting clarification 10/3/2018. They aer looking at iy. | April 27, 2018 |
Context:
- Is there any statistical difference between ICMP and TCP Ping? Context here is Internet (not data center). This is important because the network stack is different (e.g., MPI over infiniband) and latencies are significantly less.
Questions:
- Why should we focus on minimum RTT instead of average RTT
- Min RTT essentially reflects fixed delay, while average RTT subsumes variations and path load
- Are the R plots generated using minRTT?
- Averages and computed. Min RTT is available. Scripts need to be updated to use minRTT.
- What is the breakdown of latency between endpoints?
- Latency for an echo packet to travel up the stack and back down is about 3.75 micro seconds (see StackMap https://www.usenix.org/conference/atc16/technical-sessions/presentation/yasukata). As expected, this is negligible when considered with milli second latencies.
- To replicate use system tap. See: https://unix.stackexchange.com/questions/419449/how-can-i-determine-if-a-latency-is-due-to-a-driver-or-the-scheduler
- The remaining components would be propagation and queuing delay. As we cannot breakdown the two in a public network without using an active look like pathchirp etc, we'll continue to consider these as a single component.
- Latency for an echo packet to travel up the stack and back down is about 3.75 micro seconds (see StackMap https://www.usenix.org/conference/atc16/technical-sessions/presentation/yasukata). As expected, this is negligible when considered with milli second latencies.
- If there is a difference, is it because of the type or location of the source? What if the source of traffic was not SLAC?
- Is there a correlation with distance between the end points?
- Are the differences limited to a particular region?
- It may be that end hosts which are farther away have larger variances and thus the pronounced differences.
- How do we determine/understand if traffic prioritization is implemented?
- Test in a controlled environment to avoid variables such as traffic prioritization, queuing delay due to cross traffic.
- Review the time series of latencies for both ICMP and TCP ping, instead of averages?
- Is there a difference between IPv4 measurements vs IPv6.
Next Meeting
Next meeting: Tuesday, November 6th 8 pm Pacific time (Nb now on winter time); Wednesday, November 7th, 2018 9:00 am Pakistan time; 12:00 noon Malaysian & Guangzhou time; and 11 am Thailand time.
...