Re: [vchkpw] How many inodes is enough?
Zitat von Ajai Khattri <[EMAIL PROTECTED]>: Hi Ajai, > The new mail server is running Linux and qmail+vpopmail+MySQL :-) > > All this is great but I need to somehow calculate how many file system > inodes I need to move these mailboxes and support 2800 users. Does > anyone have any formulas/rules-of-thumb/tips ? Also, what would be an > appropriate block size for the fs? > You should use a filesystem with dynamic Inode allocation. I use XFS - it is incredibel fast and a journaling filesystem. It is even NFS-proof, so you can export your vpopmail-home to multiple hosts. IMHO you should not use ext[23], because hosting 1.500.000 emails, stored in Maildirs, you will not see a good performance. Reifers is also a filesystem with dynamic Inode allocation, but in my stress tests it fails under heavy SMP-Load and it has problems with NFS. Last it was unusable to be a cluster FS, because the standby host didn't see any file the origin hosts sees. So feel free to make your own tests - this is meant to be my experience. Cheers Jens Jahr
Re: [vchkpw] How many inodes is enough?
At 07:40 AM 6/11/2003 +0200, you wrote: You should use a filesystem with dynamic Inode allocation. I use XFS - it is incredibel fast and a journaling filesystem. It is even NFS-proof, so you can export your vpopmail-home to multiple hosts. IMHO you should not use ext[23], because hosting 1.500.000 emails, stored in Maildirs, you will not see a good performance. Jens Jahr I can confirm this to some extend, due to a loop in a antivirus product, i ended up filling a little over 100gb with 5kb mails (thats roughly 20.000.000.000 emails) in the postmaster@ account. on the 1.13ghz p4 512mb Qmail didnt have any problems handeling the mails in que, and delivering them to the account, but sqwebmail timed out when trying to index / count them, also the ext3 file system took so long on a "ls" that i after a long time eventually gave up and killed the process, and rm -rf'ed the "new" dir. the lesson is as Jens points out, if you are going to use maildirs, you should use a file system thats geared to handeling huge amounts of small files. Kris Cream[DONut] - www.donut.dk www.nethouse2000.dk - admin
Re: [vchkpw] How many inodes is enough?
> Reifers is also a filesystem with dynamic Inode allocation, but in my > stress tests it fails under heavy SMP-Load and it has problems with NFS. > Last it was unusable to be a cluster FS, because the standby host didn't > see any file the origin hosts sees. Hmm -- I am running a 15k-user ISP mail system on Reiserfs on SW RAID1, exporting the 'domains' directory via NFS to the actual SMTP/POP3/IMAP4 servers, all over IPSec. The edge servers handle all the virus/spam scanning and deliver directly into the NFS mounted Maildirs, and the users can contact any edge server (usually routed to the closest one via a little DNS magic) to pick up and send their mail. No SMP here, but no problems so far, either, even under heavy NFS load. I'd eventually like to have the mail spool server as a two or three node (geographically distant) cluster, but I may have to move to AFS for that. I was originally looking at various DFSs but CODA and Intermezzo are both way too immature at this point, and AFS had no decent documentation. I was also under the impression that XFS was not a suitable FS, but I have to admit I did not look too hard. Do you have some resources on XFS and what to look out for? Regards, Andrew
Re: [vchkpw] How many inodes is enough?
Zitat von Andrew Kohlsmith <[EMAIL PROTECTED]>: Hi Andrew, > No SMP here, but no problems so far, either, even under heavy NFS load. I'd exaclty that was the point "under heavy SMP-load" it broke with several error messages working as an NFS-Server ( kernel-space ). Ok, I didn't repeat the tests know for 15 month, because I decided to use XFS, so why should do these work again ? Maybe they are now fixed. But I really dont care and as I said it was meant to be _my_ experience. But you may search the reifers archives where you can find a lot of articles about this issue. It is (was) a known problem. > eventually like to have the mail spool server as a two or three node > (geographically distant) cluster, but I may have to move to AFS for that. AFS is fast and secure - I agree, but AFAIK volumes are limited to 8 GB > I was originally looking at various DFSs but CODA and Intermezzo are both CODA also broke under heavy SMP-load in my tests. > also under the impression that XFS was not a suitable FS, but I have to > admit I did not look too hard. > Do you have some resources on XFS and what to look out for? http://oss.sgi.com/projects/xfs/ You will find a lot of docu. XFS also supports full POSIX - ACL's. There are a lot of system tools that come with xfs ( diagnostic, dumps, restore, resize ) and which make life very easy and confortable. ( And of course I am familiar with XFS, because of my IRIX background ;-)) ). Feel free to use you FS of your choice, so do I. Cheers Jens
Re: [vchkpw] How many inodes is enough?
Cream[DONut] wrote: on the 1.13ghz p4 512mb Qmail didnt have any problems handeling the mails in que, and delivering them to the account, but sqwebmail timed out when trying to index / count them, also the ext3 file system took so Doesn't sqwebmail use IMAP or POP3? In which case, it was the IMAP/POP3 server timing out - what server were you using? -- Aj. Systems Administrator / Developer
Re: [vchkpw] How many inodes is enough?
On Wed, 11 Jun 2003, Ajai Khattri wrote: > Cream[DONut] wrote: > > > on the 1.13ghz p4 512mb Qmail didnt have any problems handeling the > > mails in que, and delivering them to the account, but sqwebmail timed > > out when trying to index / count them, also the ext3 file system took so > > Doesn't sqwebmail use IMAP or POP3? In which case, it was the IMAP/POP3 > server timing out - what server were you using? Nope, that's why sqwebmail is so damn fast. It reads maildirs directly. Ugly, but fast. Charles > -- > Aj. > Systems Administrator / Developer > > >
