[bind10-dev] initial memo on recursive server performance research

[Jerry Scharf]

On 07/05/2012 03:16 PM, JINMEI Tatuya / 神明達哉 wrote:
> At Thu, 5 Jul 2012 09:47:58 +0800,
> "Likun Zhang" <zlkzhy at gmail.com> wrote:
>
>>> Would it be possible to do the analysis for the actual cache hit rate
>>> rather than just the name appearing? Many of the popular web sites use
>>> short ttls (30-300s) for the main target. which reduces the
>>> effectiveness of caching.
>> Pre-fetch may helps here. the entries in cache will be refetched before they expire, in the last 10s, etc. but if the ttl is too short(less than 10s? we can give a min-ttl to stop it, :) ), it doesn't work.
> Actually, I suspect it's one of the latency vs throughput tradeoff
> issues.  prefetch could improve worst case latency, but it doesn't
> (at least directly) help improve throughput.  In fact, it could worsen
> overall throughput due to the more frequent attempt of external
> transactions (which are expensive themselves).
>
> I still think prefetch is a promising approach as latency is probably
> important for popular queries, too, but we need to carefully assess
> its pros and cons.
>
> ---
> JINMEI, Tatuya
> Internet Systems Consortium, Inc.
Jinmei,

My thinking of why it would help with performance was based on there 
being a processor available to service the fill independently of the 
process(es) answering queries. I always figure that if the query can run 
to completion and not have to schedule, the overall QPS throughput is 
higher. This may not be the case for the current software and certainly 
would slow things down if there was processor contention.

jerry