OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
From: Wietse Venema (wietseporcupine.org)
Date: Fri Apr 27 2001 - 14:46:45 CDT

  • Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

    deimoslowrider.lewman.org:
    > On Fri, Apr 27, 2001 at 12:46:27PM -0400, wietseporcupine.org spewed 1.6K bytes in 41 lines about:
    > : If you have 10 boxes, and DNS lookups produce a semi random result,
    > : then work is distributed over all boxes equally. If one box is
    > : much busier than the others, then something is wrong with that box
    > : or something is wrong with the DNS.
    >
    > This assumes a constant workload.

    Nope. This assumes that a large load is made up from many "light
    load" requests. Only then will a large load distribute equally
    over multiple machines.

    > If I have 10 massive mailings, it
    > would be faster to send the 10 massive mails to 10 different boxes (such th
    >at one per box), such that I can assure that each mailing is being given equ
    >al time (assuming all boxes are equal). Our mail lists range from 12k recip
    >ients to over 2 million. Most of the time-sensitive emails are the larger l
    >ists. I feel it would be faster to send the next mailing to the least busy b
    >ox, assuming some threshold. If all boxes hover around 97% capacity, the le
    >ast busy is still 97% busy.

    If your load is made up from a small number of "heavy load" requests,
    then round robin DNS is not going to help you to distribute a heavy
    load over multiple machines.

    > : The lower cost alternative is to use removable disks. When a box dies,
    > : pull the disks from the corpse and stick them into a standby machine,
    > : tell your router to flush its ARP table, and off you go.
    >
    > Correct. I was thinking of sheer volume of I/O requests and most re
    >movable disk systems wouldn't be as fast as a correctly configured san with
    >volumes spread out across multiple disk arrays.

    Removable disks are exactly as fast as non-removable ones, and you
    can stripe data over multiple removable disks.

    If you drop everything on that big SAN, then your SAN will have to
    suffer the load from 10 boxes pounding on it. And if you make the
    mistake to stripe multiple hosts over the same set of disks then
    performance will be terrible.

            Wietse

    -
    To unsubscribe, send mail to majordomopostfix.org with content
    (not subject): unsubscribe postfix-users