Re: [exim] Nhash and Imap

Páxina inicial
Borrar esta mensaxe
Responder a esta mensaxe
Autor: Torsten
Data:  
Para: exim-users
Asunto: Re: [exim] Nhash and Imap
Hi there,

here is a wee field report of an loadbalanced 2 node RHEL 4 CS GFS
active/active pop3/imap cluster:

First I must say it runs excellent.

Some details:
Loadbalancing: Nortel Alteon

Hardware:
        Nodes:
                   - Intel Dual Core Xeon 5130 (2 x 2000MHz)
                   - 4G Ram


         Shared Storage:
                   - SAS Raid 10


Software:
- running OS RHEL 4 with Clustersuite and Gfs
- running Exim for incoming mails
- running Dovecot (pop3(s) and imap(s)) for serving data to mua`s
- running Mysql as database: serving user accounts and maildirs
- 32k pop3/imap accounts
- 7 gigs of incoming data per day (from the daily exim statistic of
2008.08.21 - 22)
- 90k incoming messages per day (from the daily exim statistic of
2008.08.21 - 22)
- 98 gigs of mail data stored

What else can I say, it runs since 1 year without any unwanted downtime
and load problems.

Regards
Torsten



> My first 'large scale' email relied on twinned IBM 3080, later 3090...
> ... but I have learned not to be a believer in clustering for MTA.
>
> Too many eggs in one basket - even if it is a strong and well-proven

basket.
> Which IBM mainframes were, but RH GFS is not (not yet, anyway...)
>
> WHEN, not IF, s**t breaks, I prefer it breaks in manageable chunks.
>
> Unless you are taking on Yahoo, MSN, or Gmail, a by-domain split

ordinarily
> serves even for large multi-domain ISP's, 'coz there are seldom that

many in any
> single domain.tld (or division thereof).
>
> YMMV
>
> Bill Hacker



Torsten wrote:/
>> RedHat *expects* 'performance issues'?
>>
>> Refreshing change, but I am otherwise not going to go there.
>
> This performence issues are only mentioned with GFS clustering (locking
> a file and so on).
> Do you know another way to set up a loadbalanced active/active
> clustering with the same data (same storage)?
> I think GFS (Global File System) is the only way to setup two or more
> systems to read/write on the same data.
> If not, tell me.
> By the way this GFS is so "hot" that the RedHat guys are telling me that
> they need 5 days to setup this "loadbalanced active/active clustering
> with the same data". This is like a workshop (with knowledge transfer).
> By the way, they don't configure my Exim or Imap, it`s only for setting
> up GFS and clustering.
> So, let`s see. /