storage config min value

[Kevin Giles <kg...@gmx.com>]

Hi,

 I turned enabled the debugging for the cache, I noticed that the cache size is always 65M less than the value I specified in the storage.config
 and also if I set it to a value less than 65, 64 for example traffic_server throws a segmentation fault on startup. Is this a bug?

 Does anyone know why it always takes 65M away?

 Thanks and Regards,

 Kevin.

Re: storage config min value

[Igor Galić <i....@brainsware.org>]

----- Original Message -----
> 
> 
> I was trying to get a better understanding of the caching algorithm
> and what items it would
> evict when the cache was full and how long it would take to fill the
> cache. Hence the small cache size, faster to fill.
> 
> I wasn't going to report the issue, but a segmentation fault is
> probably something that should be looked at as you pointed out.
> 
> Anyway, as pointed out in previous threads, the cache is a circular
> buffer and the least recently used items get evicted
> first, as long as the item is cacheable the headers do not affect how
> long it remains in the cache, the LRU determines this.
> 
> As for how long it takes to fill the cache, if you subtract 65M from
> the value specified for the cache size
> in the storage config. If I set the cache size to 128M in the storage
> config, then the cache is 63M, according to the debug statements in
> the traffic.out
> and in practice this seems to be the case. Is this 65M for the ram
> cache or the http header store else, or something else, I don't
> know.

I would say that the cache size scales much better with
the storage size than 1:1.

RAM cache is, as the name suggests, stored in RAM. But
the RAM is also used to map the directory of the storage.
 
> If the the compression is not turned on then it is roughly a one to
> one mapping between the
> http object sizes and what the cache stores, there is a little
> overheard per http item though, is this attributed
> to the cache key and that http objects are stored as fragments?

There's a lot of loss here depending on varying parameters:
* object size
* compression efficiency
* header variation
* File System overhead
* etc

But please don't quote me on this, as my understanding
of the cache is incomplete, at best.
John should have a more sound explanation, and once
he delivers it, I shall copy/paste it into the docs ;)

> Thanks,
> 
> Kevin.

i

-- 
Igor Galić

Tel: +43 (0) 664 886 22 883
Mail: i.galic@brainsware.org
URL: http://brainsware.org/

Re: storage config min value

[Kevin Giles <kg...@gmx.com>]

I was trying to get a better understanding of the caching algorithm and what items it would
evict when the cache was full and how long it would take to fill the cache. Hence the small cache size, faster to fill.

I wasn't going to report the issue, but a segmentation fault is probably something that should be looked at as you pointed out.

Anyway, as pointed out in previous threads, the cache is a circular buffer and the least recently used items get evicted
first, as long as the item is cacheable the headers do not affect how long it remains in the cache, the LRU determines this. 

As for how long it takes to fill the cache, if you subtract 65M from the value specified for the cache size
in the storage config. If I set the cache size to 128M in the storage config, then the cache is 63M, according to the debug statements in the traffic.out
and in practice this seems to be the case. Is this 65M for the ram cache or the http header store else, or something else, I don't know.

If the the compression is not turned on then it is roughly a one to one mapping between the
http object sizes and what the cache stores, there is a little overheard per http item though, is this attributed
to the cache key and that http objects are stored as fragments?

Thanks,

Kevin.

On 28 Jun 2011, at 16:45, Igor Galić wrote:

> 
> 
> ----- Original Message -----
>> Hi,
>> 
>> I turned enabled the debugging for the cache, I noticed that the
>> cache size is always 65M less than the value I specified in the
> 
> People will usually complain about their multi-TB sized caches
> getting a little wonky at times. But I've never heard someone
> complain about 65M - that sounds terribly small for a cache..
> 
>> storage.config
>> and also if I set it to a value less than 65, 64 for example
>> traffic_server throws a segmentation fault on startup. Is this a
>> bug?
> 
> However, a Segmentation fault always sounds like a bug to me..
> 
> 
>> Does anyone know why it always takes 65M away?
> 
> Lets start out by having you submit that as a bug:
> https://cwiki.apache.org/confluence/display/TS/Filing+useful+bug+reports
> 
>> Thanks and Regards,
>> 
>> Kevin.
> 
> i
> 
> --
> Igor Galić
> 
> Tel: +43 (0) 664 886 22 883
> Mail: i.galic@brainsware.org
> URL: http://brainsware.org/

Re: storage config min value

[Igor Galić <i....@brainsware.org>]

----- Original Message -----
> Hi,
> 
>  I turned enabled the debugging for the cache, I noticed that the
>  cache size is always 65M less than the value I specified in the

People will usually complain about their multi-TB sized caches
getting a little wonky at times. But I've never heard someone
complain about 65M - that sounds terribly small for a cache..

>  storage.config
>  and also if I set it to a value less than 65, 64 for example
>  traffic_server throws a segmentation fault on startup. Is this a
>  bug?

However, a Segmentation fault always sounds like a bug to me..


>  Does anyone know why it always takes 65M away?

Lets start out by having you submit that as a bug:
https://cwiki.apache.org/confluence/display/TS/Filing+useful+bug+reports
 
>  Thanks and Regards,
> 
>  Kevin.

i

-- 
Igor Galić

Tel: +43 (0) 664 886 22 883
Mail: i.galic@brainsware.org
URL: http://brainsware.org/