Protocol Analysis Expert Q&A
A few weeks ago, the FCIA hosted a live webcast on Protocol Analysis for High-Speed Fibre Channel Fabrics in the Data Center. For those of us within the FCIA, we also referred to it as “Saving You SAN and Your Sanity!” Well, you get the picture.
If you missed the live webcast you can watch it on-demand at your convenience. We received some thought provoking questions at the live event and our expert presenters, Yamini Shastry and David Rodgers, have answered them here.
Q. Any advice on tools that can be used?
A. Some available tools we are aware of include:
- Generation/IO Tools -> SanBlaze emulation solutions, Viavi solutions FC Generator/Load tester platforms, Spirent’s test center benchmarking solutions, Medusa Labs Test tools.
- Analyzers -> Teledyn Lecroy’s Fibre Channel analyzers, Viavi solution’s Fibre Channel analyzers
- FMUs: Cisco’s UCS or Brocade’s tools, or a third party like SolarWinds “
This may not be a comprehensive list. We encourage you to do your own research and due diligence for the tools that are right for you.
Q. You mentioned that the FCIA has a webcast on the Flow Control. Where can I watch that?
A. Yes, the FCIA hosted a webcast called “Fibre Channel Performance, Congestion, Slowdrain and Over Utilization.” You can view it on-demand at your convenience.
Q. Where is the piece around ISLs?
A. In this webinar, due to time constraints, we were only able to talk about how to measure fabric congestion. ISL issues could be a separate webinar, if there is enough interest.
Q. How does the discussed analysis apply to mixed traffic – FCP-SCSI vs. FC-NVMe?
A. Fibre Channel provides a one-to-one mapping of upper layer protocol exchanges. So, the same analysis will apply to both FCP-SCSI and FC-NVMe.
Q. How are Yamini’s graphs displayed by a trace?
A. The graphs discussed in the presentation are produced by a post capture analysis engine that statefully processes the frames in the trace and creates the latency metrics. In case of exchange age latencies, most trace viewers have a transaction view or exchange view that allows you to look at the latencies of individual exchanges. The user can plot them. Stateless errors such as physical/link layer errors are easily visible in traces whereas stateful errors or latency timings takes lot of manual time or scripting to dig through and better suited for an analysis engine.
Q. How can a user view the trace when using these graphs on an analyzer?
A. If the graphing tool allows the user to go back and forth to the trace, you should be able to jump to the trace at the timestamp of the issue and look at the events around that time range. Typically, throughput dips show as large inter frame gaps in trace.
Q. List examples of customer environment issues that can be replicated with the jammer?
A. The idea of error injection testing is to create error situations and test the DUT’s recovery capabilities. Here are some examples:
- Create one lost credit i.e. Jam one R_RDY and test if the devices recover from it in time without causing latency issues.
- Jam credits to the receive credit count and cause a out of credit situation.
- Create a frame error and do not correct the CRC to test the recovery at link layer.
- Modify a field in the frame to an unexpected value for example the command code and correct the CRC to test the recovery at the SCSI layer.
- Modify a good status to a check condition to see a retransmission of the exchange on the wire.
We hope these questions helped. In addition to this webcast being available on-demand, you can also download the webcast slides. Please follow us on Twitter @FCIAnews for information on all upcoming FCIA webcasts and events.
