On Jun 16, 2011, at 20:29, Jesper Krogh wrote:
> On 2011-06-16 17:09, Haestan wrote:
>> I am evaluating hardware for a new PostgreSQL server. For reasons
>> concerning power consumption and available space it should not have
>> more than 4 disks (in a 1U case), if possible. Now, I am not sure what
>> disks to use and how to layout them to get the best performance.
> What is your data:memory-size ratio? Can you afford to have everything
> in memory and only have the disks to be able to sustain writes?
Yes, I can definitely affort to have everything in memory. Right now,
the main database is about 10GB in size including bloat (around 4GB
without). And there are some more db's of about 4GB in size. So in
total around 14GB at the moment and slowly rising.
I was planning to put in at least 16GB RAM or probably even 24GB to be
safe. The problem is that the data of the main db is more or less
constantly updated or deleted/reinserted throughout the day. It seems
to me that the resulting bloat and the constant new data is really
hurting the cache hit rate (which is now around 90% in the main appl.).
It's those remaining queries that read from the disk that I really
would like to speed up as best as possible.
>> Furthermore, the LSI MegaRAID 9261 offers CacheCade which uses SSD
>> disks a as secondary tier of cache for the SAS disks. Would this
>> feature make sense for a PostgreSQL server, performance wise?
> I have one CacheCade setup... not a huge benefit but it seems
> measurable. (but really hard to test). .. compared to a full
> SSD-setup I wouldn't consider it at all.
Thanks for that input. What I've read from you and others, the SSD
cache doesn't seem a viable option for me.