home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!sun-barr!news2me.ebay.sun.com!exodus.Eng.Sun.COM!ennoyab.Eng.Sun.COM!beepy
- From: beepy@ennoyab.Eng.Sun.COM (Brian Pawlowski)
- Newsgroups: comp.protocols.nfs
- Subject: Re: NFS I/O Ops/seconds
- Date: 26 Jul 1992 08:45:47 GMT
- Organization: Sun Microsystems Inc., Mountain View, CA
- Lines: 72
- Distribution: usa
- Message-ID: <l74phrINNcus@exodus.Eng.Sun.COM>
- References: <1992Jul22.061146.15641@u.washington.edu> <64081@hydra.gatech.EDU> <nkkvj9s@twilight.wpd.sgi.com>
- NNTP-Posting-Host: ennoyab
- Summary: nfs nhfsstone laddis
-
- [A public follow-up].
-
- I agree with Rick Jones. My take on the various status I've seen from
- both Sun and other vendor reps is that most of what was mentioned for
- reporting rules in Coolidge's posting is not a problem.
-
- Reporting rules are simple: You must report all configuration details
- which affect NFS server performance. SPEC is nailing down specifically
- for NFS what the reporting rules are, so there is not misunderstanding,
- along with guidelines--I believe--for vendors who may have non-standard
- configurations. The August SPEC meeting in Lisle will continue this
- work.
-
- "Run rules" are more contentious. Reporting rules are where you get to
- say how you set up and ran the tests. Run rules, on the other hand, are
- the "rules" you must adhere to when running the tests. This has been
- the source for alternating lively and heated discussion. I have
- disagreements with counterparts. Variety is the spice of life.
-
- Also, to my highly informed knowledge, Sun has never engaged in
- reporting numbers for nhfsstone and PRE-LADDIS where we have modified
- the mix. Some field people may have done this in an attempt to match a
- customer's measured mix (that's why the knobs were put in in the first
- place).
-
- Different versions of nhfsstone under construction (*before* it was
- called PRE-LADDIS) where the benchmark load characteristics varied from
- release to release has confused some prior reporting. That's why the
- LADDIS guys and SPEC got a little rigid regarding reporting results
- from the beta evaluation version (i.e., you can't)--things were already
- confused enough. The whole point behind this SPEC process of trying to
- hammer out an agreed to NFS (server) benchmark is to bring some sanity
- and conformity to the process. I believe SPEC is helping in this
- regard--getting players in the NFS server market to hash out their
- issues in a server benchmark to provide a tool for the industry.
-
- You mentioned that SPEC or LADDIS process is political. LADDIS
- evaluation is being complicated by different participant vendor
- concerns. My hope is that with sufficient feedback from the beta
- evaluation, and SPEC's structure of "balance of powers", that LADDIS
- will not suffer (and actually improve) as part of process. I firmly
- believe that the changes occurring since January when it was funneled
- into SPEC have been for the better. It's not perfect (what is).
-
- I've given up looking for perfection.
-
- Is SGI a SPEC member? (I do believe so) If so, I *strongly* encourage
- you to raise your objections to your SPEC rep! Now is the time. If you
- want, I'll put forward any concerns you have. I'll simply add them to
- my voluminous list. Anyone should feel comfortable evaluating
- PRE-LADDIS and getting comments back to either Bruce Keith
- at DEC (SPEC PRE-LADDIS Project Leader) or any SPEC representative.
- [Do we need a repost of the PRE-LADDIS info?]
-
- Most of my concerns--which are somewhat nebulous--center on "workload
- model". Does PRE-LADDIS (or for that matter nhfsstone) define
- sufficiently an NFS workload to justify its use as a benchmark? I tend
- to believe "yes". I and several other engineers (outside of Sun)
- recently decided that PRE-LADDIS is a very powerful tool to aid in NFS
- server analysis. Regardless of the SPEC outcome, or acceptance of
- PRE-LADDIS as the mother-of-all NFS server benchmarks, the work as it
- now stands (evolved from nhfsstone from Legato) provides a great tool
- to accomplish analysis of NFS servers. I use it regularly to subject
- servers to a reproducible load to analyze effectiveness of changes to
- software or hardware. Way cool.
-
- Rusty Sandberg did a great job on nhfsstone. Mark Wittle (DG),
- Ken Teelucksingh (DEC) and John Corbin (Sun) did major efforts
- to get PRE-LADDIS (highly evolved from nhfsstone) to where
- it is today. Thanks.
-
- Brian Pawlowski
-