|
|
Back to UserFriendly Strip Comments Index
|
Who makes good hard drives? | by woodchip | 2003-11-06 19:08:31 |
|
You've never heard of infant mortality? | by Slamlander | 2006-11-19 12:55:59 |
|
Promise? | by Myke | 2003-11-06 23:10:00 |
|
>snirk< hehe build a bootable array | by Slamlander | 2003-11-06 23:15:57 |
| Do you know what you're talking about? |
by Myke |
2006-11-19 12:55:59 |
My RDBMS is running from a RAID 5 array, pure software. Do try to know the facts before you deny them...
please, how often to you saturate your SCSI or PCI busses? If so, then you need bigger hardware, period. Besides... what is hardware RAID? Why! It's just software RAID running on a small/slower processor than your CPU! My goodness! What is hardware RAID giving me? hmmm.... more latency due to moving processing to a slower chip, which may or may not be more loaded than my CPU and is probably subject to more blocking operations than the CPU... I *DO* get sheilding from a hardware failure / bus errors... whereas if I do software RAID, I'm likely to keep the more important data in L1/L2 cache where it's useful. Not only that... caching beyond L1 & L2 is *BAD*, when the OS commits data, it should be committed, you run a higher risk of corrupting your filesystem if your OS thinks the filesystem is stable... especially if you've got multiple volumes / controllers with different loads and cache algorithms, god know where your data is post-commit... because the OS sure as hell don't. So, when you keep your indices on one volume, your big tables on another etc... when your power fails or system bus faults, and you've got differening states of data across your volumes, you're rolling back more than you need to.
Cache bad. Slower process bad. Desynchronization of commits bad for recovery (good for performance tho) |
|
[ Reply ] |
|
My, my, such a reaction... I do know the facts | by Slamlander | 2003-11-07 00:12:01 |
|
|
[Todays Cartoon Discussion]
[News Index]
|
|
