Re,

concat : le plus moyen a mon sens, t'as 3 disques, de 10Go donc une partition de 30Go
de 0-9go disk1 de 10->19 disk2 20->29 disk3,
du coup si ta partition est rempli de 15G au debut seul le disque 1 et 2 sont utilise, pour un serveur web par exemple apres l'os et les applis, les pages vont sur le disque 2 qui est toujours utlise pendans que les autres attendent.

strip, toujours les meme trois diques.
le fichiers sont tous repartis sur les trois disque en "petit morceaux" (defini par la taille du stripe), ce qui fait que pour peu que le fichier soit un peu gros tous les disques travaillent, et donc les lectures se font plus vite.


dans les deux cas, si le mot magique mirroir n'est pas dans les specs.....il faut directement:

- prends le bus
- descendre chez sa mère
- pleurer .......

a+

http://docs.sun.com/app/docs/doc/805-5962/6j5kgj6kh?a=view

Le 12 juil. 05 à 13:16, Gerard Henry a écrit :

bonjour,
y a t il une difference entre le stripe et le concat?
en relisant un vieux fil de discussion, qqun disait qu'il valait mieux concatener que stripper: Comme je suis pas sur d'avoir tout compris, je mets ici une partie du mail en copie "It would seem with 12,000 users accessing mailboxes via POP and similar,
and with a steady stream of incoming messages, what you'd really want
would be good random-access performance. Striping is going to give you
good sequential performance for one-request-at-a-time kind of loads,
which is not what you have.

Also, there is a cost to striping: when joeuser logs in via POP
(or IMAP or whatever), he'll typically download his whole mailbox.
Unless he has a very fast network connection, this will not happen
in one burst but will be a few I/O requests scattered over the
course of a few seconds (or sometimes minutes). If you use
striping, his mailbox is likely to be close to contiguous
and will thus span multiple spindles in the stripe. So, it would
seem that every time he needs to download his 250k mailbox, if you
have 5 disks in the stripe, every single disk head will have to
move to the same general area, where his mailbox is. This seems
kind of silly to me. Fetching his data as fast as possible
isn't what's desired; what you need is the ability to fetch his
data with minimal impairment to your ability to fetch others'
data at the same time.

Imagine you concatenate instead. Then, most likely (although
not definitely), joeuser's mailbox lies all on one disk.
If you have 5 concatenated disks, maybe you can get lucky and
have 5 people logged in at once, with each one fetching stuff
from their mailbox off one spindle and not disturbing others.

In fact, were it not for the extra administration headache,
I'd be really tempted to suggest making a separate filesystem
for each disk and splitting up /var/spool/mail so that different
users' mailboxes are hashed onto different disks.

So, it seems to me that perhaps a good compromise would be to
create a series of two-way mirrors and then concatenate the
mirrors. This would allow joeuser's mailbox activity to affect
only one spindle, and every mailbox would be accessible through
one of two spindles, whichever is most convenient."
_______________________________________________
Solaris_fr liste de diffusion en français pour Solaris, sur toutes architectures
Solaris_fr@x86.sun.com
http://x86.sun.com/mailman/listinfo/solaris_fr


_______________________________________________
Solaris_fr liste de diffusion en français pour Solaris, sur toutes architectures
Solaris_fr@x86.sun.com
http://x86.sun.com/mailman/listinfo/solaris_fr

Répondre à