Re: Number of Spindles Exchange + SQL
- From: <workinghard@xxxxxxxxxxxxxx>
- Date: Mon, 11 Dec 2006 22:57:31 +0100
Hello,
I can see some of your concerns ... but considers some of these taughts ...
Never seen Multiple disk failures? Great! I have, several times ... all disk
from the same batch with a production fault. Servers that are located in
harsh conditions ... Servers that had been moved / transported a couple of
times ... I to have/know customers who have never seen a RAID 0/1/5/10/3/4/6
fail ... well they should count their blessings. But we can all hope for the
best while we prepare for the worst.
The reads are now with the big caches less imporant than the writes ...The
writes are becoming the limiting factor in that respect and battery cached
controllers or technologies like lazy writing can only sustain what they are
designed to ... and that is dependend upon the price and quality of the
controller. And under real heavy loads it becomes a bottle neck.
Logs are writing ... and lots of it (depending on the load and usage
naturally ....) seperating them from random reads is good for performance
A hot spare is perhaps a luxury but one will not allways be in a situation
where some one is available to swap the disk (time, location, type of
client)
When a disk fails RAID 5 is horrible. RAID 5 is is good for one thing cheap
redundancy. If you want performant redundancy under all conditions RAID 1
or RAID 10.
Also Raid one: Break a mirror, install SPX or new app Y if all is well
recreate mirror ... if not hey use other disk to get back to begin
situation: fast, effective ... With one big raid 5 you can't do that and
you'll have to restore backups, images ...
Granted, He might not hit te bottle neck in his SBS setup but it can and if
it does .... RAID 1 is better (due to ever more caching for reads). In the
end a look at what the customers need and what their I/O is will help make
the decision. If the load is small enough you can have it all run on one
mirror and never notice ... but RAID 5. No sir, if possible I avoid it for
all te above but I have built 'm yes when "ordered" or "forced" (diskspace,
money) to do it it's better than no redundancy (as I stated that i will take
redundancy over performance in an single point of failure setup).
Cheers
"Leythos" <void@xxxxxxxxxxx> wrote in message
news:vLjfh.4165$ja6.2935@xxxxxxxxxxxxxxxxxxxxxxxxx
In article <u#gMQfWHHHA.4804@xxxxxxxxxxxxxxxxxxxx>,
workinghard@xxxxxxxxxxxxxx says...
When does this setup become worth it's money: when on monday afternoon at
13:45 disk 1 dies and 2 hours later disk 4. Changes are they are in
different mirrors (if it's one big RAID 5 you're toast) and until you get
a
spare disk in there the perfomance will hold up. If they have the money:
add
a global hot spare ... that way they regain redundancy and you or the
vendor
have more time to fix it before the entire situation turns into a
disaster
recovery.
Chances of having 2 disks go bad in the same server during the same day,
without any external factors, is very slim - never seen it happen in all
the years I've worked with low/high end servers.
Also, even if they can't do a hot-spare, I always order a spare drive
for the arrays and have it on the shelf.
Most database access is reads, at least for the generic user, 80% reads
with 20% writes, so, RAID-5 is the better option for performance.
If a user has an application that is more than 50% writes, then the
solution changes, but, I can't think of any applications off the top of
my head that is 60+% writes that users access all day long.
--
spam999free@xxxxxxxxxx
remove 999 in order to email me
.
- Prev by Date: Re: Active Sync Controls in GP
- Next by Date: Login attemps
- Previous by thread: Re: Active Sync Controls in GP
- Next by thread: Re: Number of Spindles Exchange + SQL
- Index(es):
Relevant Pages
|
