From: Stefan Beller <sbeller@google.com>
To: Jonathan Tan <jonathantanmy@google.com>
Cc: git <git@vger.kernel.org>
Subject: Re: [RFC PATCH] fetch-pack: space out sent "haves" in negotiation
Date: Tue, 22 May 2018 12:01:59 -0700 [thread overview]
Message-ID: <CAGZ79kbe0mAQW-2iGX2Q+WEWJ-MEoBcXtyBOb8a-7iWApp7pqw@mail.gmail.com> (raw)
In-Reply-To: <20180522114440.0b1aa23a780efe6a68770d9b@google.com>
Hi Jonathan,
> I wouldn't characterize the errors as "off by one errors".
Yes, I put it in quotes but realized that would not convey it very well.
> They are
> more like...let me use a diagram:
>
> A
> |\
> B D
> | |
> C E
>
> Suppose we know that the server does not have A, has C, and may or may
> not have E (we sent "have E" but didn't get a response yet). My method
> restarts the walk at all the parents of A (that is, B and D), but D is
> irrelevant to the situation (and should not be walked over - this is the
> error).
D is irrelevant to the A, C situation, but it is still a useful probing point?
So I would not call it an error but an inefficiency?
> In this patch, I wrote the new algorithm and deleted the old one.
...
> You're proposing that if I proceed with this, I split the patch into 2 -
> one to move the negotiation algorithm, and one to update it? If yes,
> normally I would agree, but the current negotiation algorithm is not
> very sophisticated (and does not take up much code), so I think it's not
> worth it.
ok, in that case I'll just dive into the code.
>
>> > +struct fetch_negotiator {
>> > + struct sent_commit **sent_commits;
>> > + size_t sent_commit_nr, sent_commit_alloc;
>> > + struct prio_queue candidates;
>> > +};
>>
>> Maybe we can just declare the struct fetch_negotiator here and not
>> define it, such that nobody outside the actual implementation tries
>> to access its internals?
>
> That's possible - I wanted to allow allocation of this on the stack (to
> save a malloc), but perhaps that's over-optimization.
Ah good call. Please keep it that way then.
>> So even if we do not use the skip commit logic, this would be a benefit for any
>> http(-v0) and v2 users of the protocol?
>
> It would conserve bandwidth, yes, but storing all the commits sent with
> additional metadata for each would require more memory.
I did not see the memory requirements here as a problem until now.
Are you saying this memory might be too much to keep around?
Stefan
next prev parent reply other threads:[~2018-05-22 19:02 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-05-21 20:43 [RFC PATCH] fetch-pack: space out sent "haves" in negotiation Jonathan Tan
2018-05-21 22:57 ` Stefan Beller
2018-05-22 18:44 ` Jonathan Tan
2018-05-22 19:01 ` Stefan Beller [this message]
2018-05-23 1:08 ` Junio C Hamano
2018-05-23 3:42 ` Junio C Hamano
2018-05-29 16:58 ` Jonathan Tan
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAGZ79kbe0mAQW-2iGX2Q+WEWJ-MEoBcXtyBOb8a-7iWApp7pqw@mail.gmail.com \
--to=sbeller@google.com \
--cc=git@vger.kernel.org \
--cc=jonathantanmy@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).