How's your server load?

Paolo Cravero pcravero at
Thu Dec 20 06:36:23 EST 2012

anyone is willing to comment on this Cyrus server load, or share theirs?

Virtual machine on vSphere 5 with 2x 2 GHz  vCPU;
cyrus-imapd RPM 2.4.16;
kernel 2.6.18 x64_64;

There are 150 simultaneous IMAP connections on average mostly coming 
from a webmail (openXchange), with 240 imapd processes. No POP3. Just 
one instance of cyrus. Messages come over LMTP at a rate of 1 
msg/second, average of course.

Data and metadata partitions are local to the virtual machine (not NFS), 
connected to a SAN via FC.
The system has 20 data partitions (and 20 metadata), for a total of 
3.8TB, of which 550GB are used. Mailboxes (accounts) sum up to 2000.

What is bugging me is the system load that averages at 10 during peak 
hours. CPU usage is equally shared between user/system/idle. As reported 
by vmstat:

us sy id wa st
33 31 32  4  0

These figures obviously vary each second, but don't go too far during 
peak hours.

/proc/meminfo reports:
MemTotal:      4043792 kB
MemFree:         34416 kB
Buffers:        305456 kB
Cached:        2505672 kB
SwapCached:          0 kB
Active:        2393176 kB
Inactive:      1083692 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:      4043792 kB
LowFree:         34416 kB
SwapTotal:     4192956 kB
SwapFree:      4192724 kB
Dirty:           86608 kB
Writeback:           0 kB
AnonPages:      665816 kB
Mapped:          95900 kB
Slab:           445736 kB
PageTables:      55056 kB
NFS_Unstable:        0 kB
Bounce:              0 kB
CommitLimit:   6214852 kB
Committed_AS:  1018640 kB
VmallocTotal: 34359738367 kB
VmallocUsed:    265672 kB
VmallocChunk: 34359472247 kB
HugePages_Total:     0
HugePages_Free:      0
HugePages_Rsvd:      0
Hugepagesize:     2048 kB

The userbase is going to grow. Another group of 3500 users with 2TB/year 
traffic is going to go on a dedicated cyrus server, similar to the one 
detailed above.

Not having a direct experience on large and active userbase under Cyrus, 
the load seems high compared to other solutions we know of. I would 
appreciate any input about the apparent ability of the current server to 
hold more users, or which debugging/characterization steps should be 
undertaken to further optimize our asset.
The "Performance Notes" page has been studied and applied when possible.

Thank you for reading so far and Season's Greetings, if applies.

More information about the Info-cyrus mailing list