>
> Hello guys,
> currently, I have
> Mean Object Size: 21.66 KB
>
> At the moment, I have 3 SCSI disks, 36 GB, 10K, so my cache_dir is:
>
> cache_dir aufs /cache/cache1 28000 60 256
> cache_dir aufs /cache/cache2 28000 60 256
> cache_dir aufs /cache/cache3 28000 60 256
>
> There is no RAID. Because my Mean Object Size is 21 K, it means, there
> is 28000000/21/60/256 = 88 files in each second-level directory (which
> is few, I didn't expect the mean object size so high, but I
> don't think
> it is a disadvantage).
>
> My boss just gave me another two SCSI disk, 70 GB, 15K - he
> wants me to
> maximise "Byte hit ratio" and I believe to have bigger cache is the
> right way (I have already set LFUDA replacement policy etc).
>
> I am going to add these two disk to the box, how would you
> use them, so
> you get the best performance?
>
> 1) add 2 new cache_dirs
> 56 GB cache on every from these two 70GB disks?
>
> cache_dir aufs /cache/cache4 56000 60 256
> cache_dir aufs /cache/cache5 56000 60 256
>
> If my mean object size stays 21 kb, there will be average 173 files
> (56000000/21/60/256) in each second-level directory.
>
> 2) add 4 new cache_dirs
> two 28 cache on every from these two 70GB disks?
>
> cache_dir aufs /cache/cache4 28000 60 256
> cache_dir aufs /cache/cache5 28000 60 256
> cache_dir aufs /cache/cache6 28000 60 256
> cache_dir aufs /cache/cache7 28000 60 256
>
>
> I prefer the first solution, but does it realy matter?
>
> I also wonder if there will be any trouble caused by my mix of three
> 36GB disk 10K only and two 70GB disks 15K working together...
>
Don't make your cache (dirs) arbitrary large. It has no
additional value and you will need more phys mem. to support
such a configuration (see FAQ).
The total cache size should be about one week of normal generated
traffic.
Byte hit also, depends on user's profile and habbits. Suppose
everyone hit's sites with no-cache pragma all the time. What then ?
So, in theory, your boss should perhaps address the user
community instead of you ... (?)
M.
Received on Fri Dec 10 2004 - 00:42:07 MST
This archive was generated by hypermail pre-2.1.9 : Sat Jan 01 2005 - 12:00:02 MST