Questions before researching a move from squid

[Kingsley Foreman <ki...@internode.com.au>]

Hi Guys,

I've been using squid in a reverse proxy environment for some time now, and am pretty unimpressed with its performance and lack of scalability and im looking at giving Traffic Server a bit of a trial. But before I do that I need to know if it supports a couple of things because if it doesn't then there is no point me testing. And yes some of these things are annoying and silly but out of my control.

1. Large files i need the ability to cache files 8gb and up (it is for a remote edge server and these happen every now and then).

2. range_offset_limit like setting, There is a client, that does 206 queries starting at byte range 0 (and of course others), i need them to still cache.

3. ICP based on lowest latency

4. nice but not required would also be esi

I guess the question is is Traffic Server for me, and should I give it a decent trial?

Kingsley

Re: Questions before researching a move from squid

[Leif Hedstrom <zw...@apache.org>]

On 01/03/2011 03:43 PM, Kingsley Foreman wrote:
> Thanks John, they are the 2 show stoppers anyway,
>
> Do you know if there is much chance of the 206 behaviour changing eventually to allow smarter retrieval, eg if it does a 206 lookup on a file that is cached as a full 200 to lookup part of the file only rather then a whole new file?

Note sure I understand this, but, if the object is cached in full, we 
will serve 206 range responses just fine out of cache. Eric has made 
some improvements too, which are on trunk. The problem is that, afaik 
(ebalsa knows this best) is that we don't cache partial objects.

To comment on your other questions:

3) ICP is "supported", but I think it's currently broken. IMO, our 
clustering feature is vastly superior to ICP anyways.

4) ESI is (partially)  implemented at Yahoo as a TS plugin. I don't know 
if they will release it as open source or not. Their implementation is 
actually pretty nice, e.g. it will make parallel fetches of components 
(compared to Varnish for example, which afaik will fetch each component 
sequentially).

Cheers,

-- leif

RE: Questions before researching a move from squid

[Kingsley Foreman <ki...@internode.com.au>]

Thanks John, they are the 2 show stoppers anyway,

Do you know if there is much chance of the 206 behaviour changing eventually to allow smarter retrieval, eg if it does a 206 lookup on a file that is cached as a full 200 to lookup part of the file only rather then a whole new file?
 
Kingsley

-----Original Message-----
From: John Plevyak [mailto:jplevyak@apache.org] 
Sent: Saturday, 1 January 2011 4:28 AM
To: users@trafficserver.apache.org
Subject: Re: Questions before researching a move from squid

I can answer 2 of these.

On 12/29/2010 5:16 PM, Kingsley Foreman wrote:
> > Hi Guys,
> >
> > I've been using squid in a reverse proxy environment for some time now, and am pretty unimpressed with its performance and lack of scalability and im looking at giving Traffic Server a bit of a trial. But before I do that I need to know if it supports a couple of things because if it doesn't then there is no point me testing. And yes some of these things are annoying and silly but out of my control.
> >
> > 1. Large files i need the ability to cache files 8gb and up (it is for a remote edge server and these happen every now and then).
> >
ATS 2.1+ can cache files up to the size of a partition (a disk/raid device) up to .5PB (500TB).

> > 2. range_offset_limit like setting, There is a client, that does 206 queries starting at byte range 0 (and of course others), i need them to still cache.
You can cache range requests by setting:
proxy.cache.http.cache.range.lookup to 1.

Range requests are cached separately from whole documents currently,
so the range request must match to be found in the cache.

> >
> > 3. ICP based on lowest latency
> >
> > 4. nice but not required would also be esi
> >
> > I guess the question is is Traffic Server for me, and should I give it a decent trial?
> >
> > Kingsley

Re: Questions before researching a move from squid

[John Plevyak <jp...@apache.org>]

I can answer 2 of these.

On 12/29/2010 5:16 PM, Kingsley Foreman wrote:
> > Hi Guys,
> >
> > I've been using squid in a reverse proxy environment for some time now, and am pretty unimpressed with its performance and lack of scalability and im looking at giving Traffic Server a bit of a trial. But before I do that I need to know if it supports a couple of things because if it doesn't then there is no point me testing. And yes some of these things are annoying and silly but out of my control.
> >
> > 1. Large files i need the ability to cache files 8gb and up (it is for a remote edge server and these happen every now and then).
> >
ATS 2.1+ can cache files up to the size of a partition (a disk/raid device) up to .5PB (500TB).

> > 2. range_offset_limit like setting, There is a client, that does 206 queries starting at byte range 0 (and of course others), i need them to still cache.
You can cache range requests by setting:
proxy.cache.http.cache.range.lookup to 1.

Range requests are cached separately from whole documents currently,
so the range request must match to be found in the cache.

> >
> > 3. ICP based on lowest latency
> >
> > 4. nice but not required would also be esi
> >
> > I guess the question is is Traffic Server for me, and should I give it a decent trial?
> >
> > Kingsley