ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dmitriy Setrakyan <dsetrak...@apache.org>
Subject Re: Re: About IgniteAtomicSequence
Date Thu, 21 Jan 2016 07:11:39 GMT
On Wed, Jan 20, 2016 at 10:59 PM, 李玉珏 <18624049226@163.com> wrote:

> Hi:
>
> I in a very old machine simple to do the test, if open reserv mechanism,
> the ID generation speed can reach 50K/s. If turn off reserv mechanism,
> the performance will be dropped to 13K/s, performance degradation is
> very obvious, but off reserv will enable ID increase linearly with
> time, this aspect of the advantages and disadvantages can allow
> developers to choose their own.
>

The whole premise behind IgniteAtomicSequence is the reserve range,
however, you are right, the IDs will be unique, but will not increase
linearly.

If you turn off the reserve range, then IgniteAtomicSequence will not be
any different from IgniteAtomicLong, and every increment will entail a
network trip to update primary and backup copies.


>
> In addition, the performance of IgniteAtomicSequence may related to the
> performance of cache nodes, therefore,AtomicConfiguration, whether it is
> necessary to add an attribute, similar nodeFilter such, let developers
> control a distributed datastructure cache deployed on the nodes with
> high performance.
>

This is a good point. Do you mind creating a Jira ticket for it?


>
>
> Sent from Mail Master
>
>
> On 2016-01-20 04:29 , Dmitriy Setrakyan <dsetrakyan@apache.org> Wrote:
>
> The performance of AtomicSequence is very good as most of the operations
> are local. The way it works is that it will reserve a range of numbers by
> sending a request, but then will return all new IDs locally, until the
> reserved range is exhausted.
>
> More information here:
> https://apacheignite.readme.io/docs/id-generator
>
> D.
>
> On Tue, Jan 19, 2016 at 4:54 AM, 李玉珏@163 <18624049226@163.com> wrote:
>
> > Hi:
> >
> > Oh, sorry, wrong title.
> >
> > 在 16/1/19 16:50, Dmitriy Setrakyan 写道:
> >
> >> Yujue Li, what exactly would you like to benchmark? Is it the
> >> IgniteAtomicSequence or IgniteAtomicReference? Also, benchmarks are
> better
> >> viewed in comparison to something else. Is there something else in
> >> particular you would like to compare these data structures to?
> >>
> >> D.
> >>
> >> On Sun, Jan 17, 2016 at 7:31 PM, 李玉珏 <18624049226@163.com> wrote:
> >>
> >> Hi:
> >>>
> >>> About IgniteAtomicReference, I think the function in the existing
> >>> distributed ID generator advantage is very obvious, if converted to
> >>> hexadecimal, can do a very short。
> >>>
> >>> I would like to ask, is there a particular description about the
> >>> performance? Did the benchmark test on a specific hardware?
> >>>
> >>
> >
> >
>
>
>
> 来自 *网易**手机号码邮箱* -- 手机号码就是邮箱帐号,了解详情>
<http://shouji.163.com>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message