Hi all, I've been looking around for specific sizing guidelines but haven't had a lot of luck. I've seen a number of resources that will tell you everything about their system but nothing about their user base (size, e-mail traffic, simultaneous sessions etc). The reverse is also true (here are our users but not much hardware detail). Also, some sites seem to have 90,000 users running on a gutless system and others seem to have a cluster for 6,000.
Based on what I've read I've come up with this solution for a client.
They want to host 6,000 users with IMAP and Web Mail access (IMP). It is
unknown how the users will utilize the system (will half of them sit
there idle all day taking up memory?), but initially the traffic will
include a low volume of small e-mails. It is unknown how the traffic
will go from there. I want to err on the side of caution and assume the
service will be a big and immediate hit. The client has the following
already:
- A BIG IP load balancer with hardware SSL support
- A full SAN setup
- A T3
Here is what I've got so far:
Back End Servers (2 of them)
- Dual PIII-933
- 4GB of Memory
- Gigabit Ethernet
- ATA-100 20GB Drive (for OS and swap)
- SAN Interface (for mail)
- Linux HA redundant (4 boxes total for 2 backend servers)
- Provides IMAP store
- Postfix for incoming SMTP from Front end servers
- Should I run a replica of LDAP here?
Front End Servers (5 to start)
- AMD Athlon 1.2Ghz
- 1.5 GB of Memory
- 10/100 NIC
- ATA-100 20GB Drive
- Linux
- Will host IMAP proxy
- Apache and IMP with Zend accelerator
- Postfix for incoming SMTP from internet
- Mail virus checking
- Should I run a replica of LDAP here?
SAN Drives
- Mirrored 75GB drives (that should be enough to start)
Cluster Control/Monitoring/Config Host (1)
- Haven't decided on specs yet
- LDAP master
- Logging host. No logging done on individual servers. Sends log to
here
- RSYNC or NFS filesystems for front end servers
- NetSaint & MRTG monitoring
- ACAP server if needed
- Probably using Linux HA
I based the mem requirements on 2MB per active IMAP session and an extra
1MB for IMP. So there would be 2MB taken up on the back and end 3MB
taken up on the front end unless the proxy is super light. This cluster
would allow for 2300 on the front end servers where more could be added
until the limit of the backend it reached (3800, allowance made for OS
and cache). Another back end system could be added if needed. I based
disk requirements on 10MB quotas for most users with a little left over
for power users and admins. Is the processor too much on the back end?
I get the feeling that I'm overpowered here, but I'd like to know for
sure. Thanks!
Neil
P.S. Sorry if I missed some completely obvious sizing document. Please
point me in that direction if one exists.
smime.p7s
Description: S/MIME Cryptographic Signature
