trafficserver-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zirui" <>
Subject Re: Ram cache not utilized?
Date Thu, 23 Dec 2010 17:38:01 GMT
 I've tried with 2.1.4 and it performed the same. I think your guess is
 probably right. The algorithm wont promote twice-hit object to mem.</:includetail>
 Is there any choice of ram cache algorithm?</:includetail>
 BTW, I have compared 2.1.0 and 2.1.4, found that some config files</:includetail>
 like nntp.config and filter.config are no longer existing in 2.1.4. Are</:includetail>
 those functions removed in 2.1.4 ? Are those functions working</:includetail>
 properly in 2.1.0? </:includetail>
  ------------------ Original ------------------
  From:  "Leif Hedstrom"<>;
 Date:  Fri, Dec 24, 2010 01:27 AM
 To:  "users"<>; 
 Cc:  "Zirui"<>; 
 Subject:  Re: Ram cache not utilized?

On 12/23/2010 10:08 AM, Zirui wrote:
> Hi guys,
> I have installed ATS-2.0.1 and done some stress test with web
> polygraph. I found a interesting thing that when I set server
> sending 100% cachable objects and client to made 50% revisit
> to the url it has visited, object size is about 15KB.
> Both ATS and Squid have perfect hit rate. However, squid runs
> faster because it has many TCP_MEM_HIT meanwhile ATS seems
> always read&write disks (cache.db). I used systemtap to trace
> traffic_server and found I/O on cache.db about 10MB/s, which is
> considerable on my SATA disk. Squid seldom reads or writes
> disk.
> I am sure I modified following config:
> proxy.config.cache.ram_cache.size 536870912
> Disk cache is 1GB.
> Is there any other thing I have to change to enable ATS utilize
> more memory?

That should do it (you have it right, in that you need both a disk cache 
and RAM cache). The one thing I can think of is that objects don't get 
promoted to RAM cache directly, so if you only hit the object twice, you 
wouldn't see any RAM cache hits I think.

Assuming however that you are hitting each objects more than twice, I'm 
not sure what is going on. Is there any chance you can try v2.1.4 
instead and see if the same problem occurs there ?

-- Leif

View raw message