1

Just got off the phone with Dell, who told me that there's not much difference from 256MB to 512MB on their Perc6 controllers.

I'm aiming for a very fast MySQL server with 4 data disks in RAID10 15k, and 2 root disks RAID1.

Any advice pls.

Dennis Williamson
  • 62,149
  • 16
  • 116
  • 151
jDempster
  • 143
  • 1
  • 8

4 Answers4

6

Difference depends on particular scenario. I would always go with more memory since that means that there is higher chance of having my data already available for reading. And do not forget, if you have battery backed memory, that memory helps with data writes too.

However, there is always critical amount that gives best performance/price ratio. Double the memory and you may only improve performance by 20%.

If there is enough money, I would go for more memory.

Josip Medved
  • 1,734
  • 2
  • 18
  • 18
  • +1 - Depends on price but it seems every time I limit myself on memory/cache/buffer/etc it always comes back to bite me in the ass. Its just a matter of WHEN. – KPWINC Aug 19 '09 at 14:01
3

A larger buffer in the controller will help with overall write speed by gathering a number of writes that are scattered and organizing them so they go faster. In some cases, it may even eliminate a write sequence if the block was modified before it was committed to disk. If you're putting enough load to fill it, more memory will help.

Without any workload info, that's probably the best I can give you.

Jeff Ferland
  • 20,547
  • 2
  • 62
  • 85
  • Yes, and remember that in a raid 10 configuration all writes are done twice, once to each stripe in the mirror. So the more organized they are the better. – Brettski Aug 19 '09 at 13:31
0

The whole thing basically comes down to your particular usage scenario. If you have the money, go for the extra memory, as throwing hardware at it won't hurt but not having enough can affect performance.

If you can upgrade the memory later on and are short on cash go for the 256 and see what your stats show in actual performance for your particular workload and usage.

People ask questions about whether XYZ is enough for a web server or web BBS...well, there's a Commodore 64 running a web server on the Internet. It's enough for them. Except when slashdotted, otherwise it chugs along just fine for their needs.

Your disk usage depends on how hard you're slamming the disk controller. Do you have any current stats you could share to estimate your load?

Bart Silverstrim
  • 31,172
  • 9
  • 67
  • 87
0

512MB will improve the odds of a cache hit somewhat. How much effect this has will depend on the working set of your workload. The working set is the set of data that your application is using frequently enough for cache misses to cause significant delays.

In an extreme case such as a table scan of a 1TB table there is no locality of reference (likelihood of the same blocks being hit more than once), so no reasonable amount of cache memory will make a significant difference. At the other extreme the whole database could fit into the cache.

The only way to get a clear view of the benefits is to test. However, unless the 512MB cache module is a lot more expensive, get the biggest cache you can afford.

ConcernedOfTunbridgeWells
  • 8,820
  • 2
  • 32
  • 52