The Zimbra team has just posted the PDF of a benchmarking study that we recently completed with HP. This particular study targets a service provider (Telco, ISP) consumer-facing deployment. (We are also at work on benchmarks for large and medium business profiles that you can expect to see shortly.) The end result was that ZCS delivered consistent and excellent response times while scaling up to 50,000 concurrent/active users and multiple 100,000s of provisioned users per dual-core/dual CPU PC-class server! The Zimbra Community should be thrilled by these results—we are effectively not giving up anything in terms of performance and scale to existing service provider solutions, and we have a dramatically richer feature set to boot.
For those that may be curious how Zimbra has come so far so quickly, the key architectural justifications for these successes can be found in “Scaling-up Zimbra”, “Innovation to validation in less than half the time”, and “Ajax impact on server scaling”. Particular kudo’s for all of the open source infrastructure upon which the Zimbra Collaboration Suite depends.
While there is much more detail in the document, please be forewarned that (1) this is a comparative benchmark, and as such is not intended to be directly used for capacity planning, and (2) that this result is for a consumer profile (e.g., 8 sends/8 receives per day of 32K average sized messages) rather than that for a business/power user profile. For more on the latter, please stay tuned!
And for any that may be concerned that it is hard for a software solution like Zimbra to meet the broad needs of both organizations (businesses, .edu’s, .gov’s, and .org’s) as well as those of service providers (hosting for consumers and businesses), I would humbly point out that we are following the successful path of the open source infrastructure that Zimbra is built upon—Linux, Apache, and so on, whose value across this spectrum has already been proven out.