Re: [PATCH v2] fetch-pack: grow stateless RPC windows exponentially

From: Stefan Beller <sbeller@google.com>
Date: Jul 19, 2016, 19:17 UTC
Message-ID: <CAGZ79kZxN9uYprMUXMuWpO3mKfw6WTa3uoG=SXWny7St1OfGzA@mail.gmail.com>
In-Reply-To: <CAGf8dgL3t7uX7yAux0xc2QMJJdmnM0262Quj4o6gDehwA+4JqQ@mail.gmail.com>

On Tue, Jul 19, 2016 at 12:03 PM, Jonathan Tan <jonathantanmy@google.com> wrote:

Show 25 quoted lines

> On Tue, Jul 19, 2016 at 9:46 AM, Stefan Beller <sbeller@google.com> wrote:
>> Care to elaborate on why you choose 11/10 as growth factor?
>>
>> (As someone who has a tick in micro optimizing:
>> 9/8 is roughly the same exponent, but the division
>> by 8 is easier as it is just a shift by 3. Similar 17/16)
>
> I don't have a specific reason for 11/10 as opposed to, say, 9/8 - I
> think that the time taken to execute this line is negligible compared
> to what's done in the calling code, but I'll change it to 9/8 if there
> is another reason for me to send another patch.
>
>> I guess one design criterion was 10 being a round number?
>> Does it make sense to experiment with the factor at all?
>> Digging into that, LARGE_FLUSH originates from 6afca450c3f,
>> (2011-03-20, fetch-pack: progressively use larger handshake windows),
>> and before we only had a linear growth.
>>
>> So I guess what I do not understand is why we need to slow down the
>> exponential growth at all?
>
> The current code has an exponential (a' = a * 2) then a linear (a' = a
> + 1024) growth. I'm not slowing down the exponential growth - that
> part is retained. I'm replacing the linear growth with another
> conservative exponential growth (a' = a * 11 / 10).

Sorry for the miss understanding. Once we have the new conservative exponential, we'd have a fast exponential first (a=2*a) and then after a while a slower exponential (a=1.1*a). So we have 2 exponential curves with different growth properties.

So my question could be reworded as: Why do we need two different exponential growth phases (as they are both exponential)? And I answered myself with: Oh, the exponents are different, so that's why.

But then I tried to understand why we choose the 2 exponents as of this patch. (Why start with 2 and drop back to 1.1?) Could one exponential phase be sufficient with an exponent in between (e.g. a=1.3*a with a larger starting a to compensate for the reduced early growth) ?

Or to reword it another way: Using just one exponential growth is simpler than 2 different exponential growth phases, so how do we explain the added complexity?

Thanks, Stefan