web workload time units and byte rates

[tdeneau]

In the raw files produced by the web workload, specifically SPECweb_Support.xxx.raw, there are result fields labelled AVG_REQ_TIME, TOTAL_TIME, RESP_TIME_(MAX), etc.  What are the time units of these fields?

Also in the run rules, I see
   "Very large static files (i.e. Support downloads) use specific byte rates as their QoS requirements."

What are these "specific byte rates"?

Also, where in the java source code would one look to see the code that validates the QOS requirements listed in the run rules doc?

[DavidSchmidt]

Hi. More details on the webserver workload can be found in the SPECweb2009 Design Guide (see: https://www.spec.org/web2009/docs/design/SPECweb2009_Design.html). Also, more details on the support workload used by SPECvirt_sc2013 can be found here: https://www.spec.org/web2009/docs/design/SupportDesign.html.

To answer your specific questions, the different fields for response times are measured in mSecs.

Regarding the byte rates, you can see that in the design guides and in the validation code.

The source code for the QoS requirements is located in src/org/spec/specweb/Validator.java.

[tdeneau]

Thanks, David.

and I guess the field called AVG_REQ_TIME is in seconds rather than ms.

[tdeneau]

I was surprised that average byte rate for the download requests seems to be about 100,000 bytes/sec regardless of the number of tiles.  I would think that a smaller number of concurrent web servers on an SUT would result in higher byte rates.

What does it mean when it says "SPECweb2009 uses a rated receive mechanism for simulating connection speeds. The simulated connection speeds have a maximum rate of 100,000 bytes/sec."

-- Tom

[DavidSchmidt]

Hi Tom. Sorry for the late response here, but it took some digging to answer your question since I wasn't involved in the original development of the SPECweb2005/2009 benchmark. The workload stresses a system by creating a certain number of "simultaneous session" (for SPECvirt_sc2013 it's 500). Each session is throttled so that it cannot exceed 100KB/s worth of receive requests. This is what is meant by rated receive mechanism. There is a good write-up of this mechanism in this SPECweb99 whitepaper: (https://www.spec.org/web99/docs/whitepaper.html, section 4.2 and 4.3).

Keep in mind that the benchmark this workload was based on, SPECweb2005, started its development quite some time ago (2001-2002), so the typical "real-world" client connection speed was still quite low compared to today, so 800Kb/100KB wasn't an unreasonable target. When it was adopted for a SPECvit_sc2013 workload, we didn't change that portion of the code, but rather just increased the number of simultaneous sessions and required SSL encryption to beef up the CPU requirements.