As you can imagine I am getting daily emails about the status of our GMS replacement. In my last blog on the subject I stated that we would probably be shipping that product this month. Since that time and, as is often the case with software projects, things have not run entirely to plan. We had already put the product through our superlab for scalability testing, and refactored some of the code in response. We also had some additional work to do to change the underlying database that we using. So far everything looked good in our lab - we were getting good throughput (1000 users/devices in the test environment) and stability was good.
Then we got to the internal rollout. This is where we hit some problems - lab data and real life data are two very different things. Some users would sync with no issue at all, and some users could not get even the first item to sync. Product stability took a hammering too. Our engineering teams have been working through the issues and are making good progress, to the point where we are now syncing what seemed to be our most problematic mailboxes. In that process we have seemingly hit the item limit of an iPhone (9,999 emails), and we have Palm, Windows Mobile, Nokia and Android devices working against the server too.
In response to some of the early testing that we, and others have done, we have made a fairly major product change. We have consolidated our platform support matrix down to a single configuration. That supported configuration is 64bit SLES 11, using the PostgreSQL database. There are a few reasons that we did this, but the one that will help our customers the most is that we are delivering it as an Add-on CD for SLES. This means that you install it much like OES, and the wizard should take care of most of the configuration for you. This should allow you to get it up and running more quickly and reliably, and will cut down on the install types of questions that our support team gets.
We also looked at delivering the product as an appliance for XEN and VMWare, however, we chose to push that deliverable off until after we ship this first version. A question to you is how many of you are ready for a virtualized appliance, what kind would you want - and just as importantly, how many of you are not ready?
After all of the work that we did over the last 2 months we put the product through our superlab again, and results are very encouraging (caveat - test environment/data only). We seemed to comfortably scale to well over 1000 users on the configuration mentioned above. We also benchmarked it against GMS on the same hardware, and performed much better, both under normal load and very heavy load. On our test harness we had 1500 users configured and syncing with no real issue (CPU util, memory consumption, Disk I/O and response times were all within acceptable limits). We will get more real world data as we roll out more broadly internally, which is the real litmus test.
Our next milestone is to deliver the connector to our current set of Gradenko customers and let them validate it. After that we will go to closed beta, and then public beta. As I mentioned on NGWList yesterday I am expecting to get to public beta or FCS within the next 3 months or so, depending on closed beta experiences.
Dean recently blogged on the devices that will be supported here.
I am sorry that we are still not able to deliver the product, but I would rather we lived through this initial pain and get it fixed, rather than deliver something of low quality to our customers