You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@trafficserver.apache.org by Ron Tsoref <ch...@gmail.com> on 2011/11/15 17:44:03 UTC
Real-world numbers
Has anybody released numbers of ATS performance in real-world scenario? I
mean, performance of ATS serving images, CSS/JavaScript files, etc. (I
don't mean 100 bytes objects servering performance, as this the only thing
I found.)
Anything related to the subject is more than welcome!
Ron
Re: Real-world numbers
Posted by "ming.zym@gmail.com" <mi...@gmail.com>.
http://people.apache.org/~zym/benchmark/20111116-cn62.txt
this is the status of reverse proxy working in my production env, normal
time.
FYI
在 2011-11-16三的 09:56 +0700，Hung Nguyen写道：
> We're using ATS in production too. In our environment, we use ATS as
> reverse proxy, serving thousands of connection for social games and
> website, behind IP virtual server. Till now, I dont have anything to
> complaint about ATS but, look at result of our benchmarking and
> service monitoring, ATS is as good as SQUID and Varnish currently do,
> I meant 3 are good and I dont see TS is better.
> We dont use clustering function, because it's (maybe) broken or
> untested, undocumented, and maybe not stable, we cannot bring it to
> production environment.
> I attach to this email screenshot of IP virtual server, it shows
> number of connection ATS is serving for our social game. Hope it's
> worth to let you give ATS a try.
>
>
> On Wed, Nov 16, 2011 at 12:31 AM, Mav Peri <ma...@optasports.com>
> wrote:
> We are using ATS in production and we have seen it serving up
> to around up to 5k requests per second on dynamic content.
> This was in a cluster of EC2 instances and not a single
> physical server with several objects being in the kb.
>
>
> It works well but if the origin server becomes congested ATS
> is missing functionality to serve stale if congested. I have
> observed that this results in ATS being unable to catch up
> under heavy load even if the origin server is not congested.
> Eventually it does catch up but the behaviour is far from
> desirable.
>
>
> Another big limitation at present is that the ICP
> functionality is broken and there is no proper documentation
> on implementing parent http (if that would be suitable) so you
> will have to rely on trial and error and any help the guys on
> this mailing can provide. The lack of ICP is not to be
> underestimated. It is a waste of resources which increases
> load on your origin servers.
>
>
> From: Ron Tsoref <ch...@gmail.com>
> Reply-To: <us...@trafficserver.apache.org>
> Date: Tue, 15 Nov 2011 18:44:03 +0200
> To: <us...@trafficserver.apache.org>
> Subject: Real-world numbers
>
>
>
> Has anybody released numbers of ATS performance in real-world
> scenario? I mean, performance of ATS serving images,
> CSS/JavaScript files, etc. (I don't mean 100 bytes objects
> servering performance, as this the only thing I found.)
>
>
> Anything related to the subject is more than welcome!
>
>
> Ron
>
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security
> System.
> For more information please visit
> http://www.messagelabs.com/email
> ______________________________________________________________________
>
>
>
>
> --
> ---
> hungnv
>
--
zym, Zhao Yongming.
aka: yonghao @ taobao.com
Re: Real-world numbers
Posted by Hung Nguyen <it...@gmail.com>.
We're using ATS in production too. In our environment, we use ATS as
reverse proxy, serving thousands of connection for social games and
website, behind IP virtual server. Till now, I dont have anything to
complaint about ATS but, look at result of our benchmarking and service
monitoring, ATS is as good as SQUID and Varnish currently do, I meant 3 are
good and I dont see TS is better.
We dont use clustering function, because it's (maybe) broken or untested,
undocumented, and maybe not stable, we cannot bring it to production
environment.
I attach to this email screenshot of IP virtual server, it shows number of
connection ATS is serving for our social game. Hope it's worth to let you
give ATS a try.
On Wed, Nov 16, 2011 at 12:31 AM, Mav Peri <ma...@optasports.com> wrote:
> We are using ATS in production and we have seen it serving up to around up
> to 5k requests per second on dynamic content. This was in a cluster of EC2
> instances and not a single physical server with several objects being in
> the kb.
>
> It works well but if the origin server becomes congested ATS is missing
> functionality to serve stale if congested. I have observed that this
> results in ATS being unable to catch up under heavy load even if the origin
> server is not congested. Eventually it does catch up but the behaviour is
> far from desirable.
>
> Another big limitation at present is that the ICP functionality is broken
> and there is no proper documentation on implementing parent http (if that
> would be suitable) so you will have to rely on trial and error and any help
> the guys on this mailing can provide. The lack of ICP is not to be
> underestimated. It is a waste of resources which increases load on your
> origin servers.
>
> From: Ron Tsoref <ch...@gmail.com>
> Reply-To: <us...@trafficserver.apache.org>
> Date: Tue, 15 Nov 2011 18:44:03 +0200
> To: <us...@trafficserver.apache.org>
> Subject: Real-world numbers
>
> Has anybody released numbers of ATS performance in real-world scenario? I
> mean, performance of ATS serving images, CSS/JavaScript files, etc. (I
> don't mean 100 bytes objects servering performance, as this the only thing
> I found.)
>
> Anything related to the subject is more than welcome!
>
> Ron
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email
> ______________________________________________________________________
>
--
---
hungnv
Re: Real-world numbers
Posted by Mav Peri <ma...@optasports.com>.
Please note I am not part of the team that looks after ATS. I started
using ATS soon after it was made available.
Up to a couple of months ago we couldn't get ATS to work with ICP and the
responses I got from the list was that it is still broken.
On 17/11/2011 07:59, "Jan Algermissen" <al...@mac.com> wrote:
>Mav,
>
>On Nov 15, 2011, at 6:31 PM, Mav Peri wrote:
>
>> We are using ATS in production and we have seen it serving up to around
>>up to 5k requests per second on dynamic content. This was in a cluster
>>of EC2 instances and not a single physical server with several objects
>>being in the kb.
>>
>> It works well but if the origin server becomes congested ATS is missing
>>functionality to serve stale if congested. I have observed that this
>>results in ATS being unable to catch up under heavy load even if the
>>origin server is not congested. Eventually it does catch up but the
>>behaviour is far from desirable.
>>
>> Another big limitation at present is that the ICP functionality is
>>broken and there is no proper documentation on implementing parent http
>>(if that would be suitable) so you will have to rely on trial and error
>>and any help the guys on this mailing can provide. The lack of ICP is
>>not to be underestimated.
>
>I did not find this in the issues list and the docs[1] suggest ICP is
>there. You say this as if it was a well-known issue - can you explain or
>send a pointer?
>
>Jan
>
>
>
>[1] http://trafficserver.apache.org/docs/v2/admin/hier.htm#ICPPeering
>
>
>
>
>> It is a waste of resources which increases load on your origin servers.
>>
>> From: Ron Tsoref <ch...@gmail.com>
>> Reply-To: <us...@trafficserver.apache.org>
>> Date: Tue, 15 Nov 2011 18:44:03 +0200
>> To: <us...@trafficserver.apache.org>
>> Subject: Real-world numbers
>>
>> Has anybody released numbers of ATS performance in real-world scenario?
>>I mean, performance of ATS serving images, CSS/JavaScript files, etc. (I
>>don't mean 100 bytes objects servering performance, as this the only
>>thing I found.)
>>
>> Anything related to the subject is more than welcome!
>>
>> Ron
>>
>> ______________________________________________________________________
>> This email has been scanned by the MessageLabs Email Security System.
>> For more information please visit http://www.messagelabs.com/email
>> ______________________________________________________________________
>
>
>______________________________________________________________________
>This email has been scanned by the MessageLabs Email Security System.
>For more information please visit http://www.messagelabs.com/email
>______________________________________________________________________
Re: Real-world numbers
Posted by Jan Algermissen <al...@mac.com>.
Mav,
On Nov 15, 2011, at 6:31 PM, Mav Peri wrote:
> We are using ATS in production and we have seen it serving up to around up to 5k requests per second on dynamic content. This was in a cluster of EC2 instances and not a single physical server with several objects being in the kb.
>
> It works well but if the origin server becomes congested ATS is missing functionality to serve stale if congested. I have observed that this results in ATS being unable to catch up under heavy load even if the origin server is not congested. Eventually it does catch up but the behaviour is far from desirable.
>
> Another big limitation at present is that the ICP functionality is broken and there is no proper documentation on implementing parent http (if that would be suitable) so you will have to rely on trial and error and any help the guys on this mailing can provide. The lack of ICP is not to be underestimated.
I did not find this in the issues list and the docs[1] suggest ICP is there. You say this as if it was a well-known issue - can you explain or send a pointer?
Jan
[1] http://trafficserver.apache.org/docs/v2/admin/hier.htm#ICPPeering
> It is a waste of resources which increases load on your origin servers.
>
> From: Ron Tsoref <ch...@gmail.com>
> Reply-To: <us...@trafficserver.apache.org>
> Date: Tue, 15 Nov 2011 18:44:03 +0200
> To: <us...@trafficserver.apache.org>
> Subject: Real-world numbers
>
> Has anybody released numbers of ATS performance in real-world scenario? I mean, performance of ATS serving images, CSS/JavaScript files, etc. (I don't mean 100 bytes objects servering performance, as this the only thing I found.)
>
> Anything related to the subject is more than welcome!
>
> Ron
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email
> ______________________________________________________________________
Re: Real-world numbers
Posted by Mav Peri <ma...@optasports.com>.
We are using ATS in production and we have seen it serving up to around up
to 5k requests per second on dynamic content. This was in a cluster of EC2
instances and not a single physical server with several objects being in the
kb.
It works well but if the origin server becomes congested ATS is missing
functionality to serve stale if congested. I have observed that this results
in ATS being unable to catch up under heavy load even if the origin server
is not congested. Eventually it does catch up but the behaviour is far from
desirable.
Another big limitation at present is that the ICP functionality is broken
and there is no proper documentation on implementing parent http (if that
would be suitable) so you will have to rely on trial and error and any help
the guys on this mailing can provide. The lack of ICP is not to be
underestimated. It is a waste of resources which increases load on your
origin servers.
From: Ron Tsoref <ch...@gmail.com>
Reply-To: <us...@trafficserver.apache.org>
Date: Tue, 15 Nov 2011 18:44:03 +0200
To: <us...@trafficserver.apache.org>
Subject: Real-world numbers
Has anybody released numbers of ATS performance in real-world scenario? I
mean, performance of ATS serving images, CSS/JavaScript files, etc. (I don't
mean 100 bytes objects servering performance, as this the only thing I
found.)
Anything related to the subject is more than welcome!
Ron
______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email
______________________________________________________________________