Re: Fwd: [squid-users] aufs cache_dir growing beyond configured limits?

Hi Henrik,

sorry to hijack the topic.

On Tue, 1 Jan 2008, Henrik Nordstrom wrote:

> On mn, 2007-12-31 at 08:48 -0800, Neil Harkins wrote:
>> On 12/30/07, Henrik Nordstrom <henrik@henriknordstrom.net> wrote:
>>> 950 MB == 972800 KB
>>
>> Yes, Adrian pointed that out a few weeks ago.
>> Got any other recommendations on the points below?
>
> No, there is not much more to be said.
>
> Remember to issue "squid -k rotate" periodically, to let Squid compact
> the swap.state index/log.
>
> And always leave about 30% free space in the cache partition, or the
> filesystem will degrade noticeably.

Is it really REQUIRED to leave 30% of free space
on the partition / under what condition would it degrade the filesystem
if less than 30% free space on partition. I'm using reiser v3 and only
have about 10% free space on both of my cachedir partitions and have not
yet seen any performance degrade. I'm getting 45-50% byte hit and similar
request hits. I was short on disk space so I'm pushing it hard but it
would be great to know the pitfalls of my scenario and would decrease the
cache size if required.

client_http.requests = 136.414047/sec
client_http.hits = 68.765346/sec
client_http.errors = 0.000000/sec
client_http.kbytes_in = 99.904748/sec
client_http.kbytes_out = 1027.876928/sec
client_http.all_median_svc_time = 0.273318 seconds
client_http.miss_median_svc_time = 1.311657 seconds
client_http.nm_median_svc_time = 0.000911 seconds
client_http.nh_median_svc_time = 1.242674 seconds
client_http.hit_median_svc_time = 0.010350 seconds

Connection information for squid:
         Number of clients accessing cache: 447
         Number of HTTP requests received: 107420517
         Number of ICP messages received: 2747673
         Number of ICP messages sent: 2748084
         Number of queued ICP replies: 0
         Request failure ratio: 0.00
         Average HTTP requests per minute since start: 3577.4
         Average ICP messages per minute since start: 183.0
         Select loop called: 891765974 times, 2.020 ms avg
Cache information for squid:
         Request Hit Ratios: 5min: 50.4%, 60min: 47.9%
         Byte Hit Ratios: 5min: 49.9%, 60min: 46.8%
         Request Memory Hit Ratios: 5min: 14.3%, 60min: 16.0%
         Request Disk Hit Ratios: 5min: 38.4%, 60min: 37.8%
         Storage Swap size: 62259262 KB
         Storage Mem size: 1048816 KB
         Mean Object Size: 15.68 KB
         Requests given to unlinkd: 0
Median Service Times (seconds) 5 min 60 min:
         HTTP Requests (All): 0.27332 0.61549
         Cache Misses: 1.31166 1.31166
         Cache Hits: 0.01035 0.00865
         Near Hits: 1.24267 1.24267
         Not-Modified Replies: 0.00091 0.00091
         DNS Lookups: 0.00278 0.00278
         ICP Queries: 0.00000 0.00000

these are my cache partitions:
[root@dns2 ~]# df -h
Filesystem Size Used Avail Use% Mounted on
/dev/sda1 35G 31G 3.5G 90% /sda1
/dev/sdb1 35G 32G 3.0G 92% /sdb1

cache_dir aufs /sda1/cache0 8000 20 256
cache_dir aufs /sda1/cache1 8000 20 256
cache_dir aufs /sda1/cache2 8000 20 256
cache_dir aufs /sda1/cache3 8000 20 256
cache_dir aufs /sdb1/cache0 8000 20 256
cache_dir aufs /sdb1/cache1 8000 20 256
cache_dir aufs /sdb1/cache2 8000 20 256
cache_dir aufs /sdb1/cache3 8000 20 256

I'm using squid 2.6S17 on centOS.

Thanks.

Manoj

--