Search Postgresql Archives

Re: Disk arrangement in a cheap server

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Why the hell would you buy a 1U chassis in the first place when perfectly good cheap 4U chassis exists that will take 8 or more drives?

1U motherboards are a pain, 1U power supplies are a pain and 1U space for drives sucks.

Most tests I've seen these days show that there is very little actual benefit from seperating pg_xlog and tablespace if you have a half decent controller card.  Infact you are better off putting it all on one nice RAID 10 to get the good read performance that splitting it up will loose.

if you don't have a decent controller card, RAID 0 will suck too.  Namely onboard SATA RAID often sucks.

Alex

On Nov 24, 2007 12:06 PM, Steve Atkins < steve@xxxxxxxxxxx> wrote:

On Nov 24, 2007, at 8:17 AM, Ron Johnson wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 11/24/07 09:12, Scott Marlowe wrote:
>> On Nov 24, 2007 5:09 AM, Clodoaldo
>> < clodoaldo.pinto.neto@xxxxxxxxx> wrote:
>>> I will build a cheap server and I'm in doubt about what would the
>>> the
>>> best for performance:
>>>
>>> 1 - everything in one lonely fast 10,000 rpm Raptor HD;
>>>
>>> 2 - two cheap 7,200 rpm 16MB cache HDs like this:
>>>
>>>     disk 1 - system and pg_xlog

This doesn't really buy you much. The supposed advantage of having
pg_xlog on its own drive is so that the head doesn't need to seek. If
it's on the system drive it'll be competing with, at least, syslog.

>>>     disk 2 - pg_data without pg_xlog
>>>     or a better arrange suggested by you;
>>>
>>> 3 - The two cheap HDs above in Raid 0.
>>
>> From a DBA perspective, none of those seem like a good choice, as
>> there's no redundancy.
>>
>> I'd make the two 7200 RPM drives a RAID-1 and have some redundancy so
>> a single disk failure wouldn't lose all my data.  then I'd start
>> buying more drives and a good RAID controller if I needed more
>> performance.

It depends on what the box is used for, but for most cases where the
data
is valuable, that sounds like a much better idea.

For batch data crunching, where the data is loaded from elsewhere then
processed and reported on, the cost of losing the data is very low, and
the value of the machine is increased by RAID0-ing the drives to make
the crunching faster... RAID0 could be good. That's probably not the
case
here.

>
> Remember: disks are *cheap*.  Spend an extra US$250 and buy a couple
> of 500GB drives for RAID 1.  You don't mention what OS you'll use,
> but if you really need cheap then XP & Linux do sw RAID, and FreeBSD
> probably does too.
>

Disks aren't necessarily cheap. Disks are fairly expensive, especially
when you need more spindles than will fit into the servers chassis
and you
need to move to external storage. Disk n+1 is very expensive, likely
more expensive than the cheap 1U server you started with.

Two, though, does seem to be false economy for a server that'll be
running a database, when you can get a 1U chassis that'll take 4 drives
pretty cheaply.

Cheers,
  Steve


---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
      choose an index scan if your joining column's datatypes do not
      match


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [Postgresql Jobs]     [Postgresql Admin]     [Postgresql Performance]     [Linux Clusters]     [PHP Home]     [PHP on Windows]     [Kernel Newbies]     [PHP Classes]     [PHP Books]     [PHP Databases]     [Postgresql & PHP]     [Yosemite]
  Powered by Linux