Following Part I, I did some more tests. It was a similar format to last time, but I though it worth looking at the results, and I also spoke to Microsoft Support about this latency question.

So, similar results to last time. The trial tenant was a little slower, but this was probably an artefact of the slower sampling rate used in this experiment.

I also drew up a chart of upload time of day vs how long indexing took. This chart only shows the results from our company tenant.

You’ll note that times can be quite long during the day, but then overnight indexing seems to run much more quickly. This is important (see below).

Microsoft Support

I discussed the indexing speed with Microsoft support; the 7 hours or longer we saw in the first round of tests was sufficiently long I felt there was an issue. I was told by the support engineer (who was pretty helpful given this question being flung at him) that:

Microsoft consider 8 hours or less to be a normal indexing time.

Cripes! 8 hours or less is normal? That’s disappointing. The good news is that all documents in all tests were indexed within that time. It’s just a shame it’s so long; I’m sure I remember reading a much shorter time in the service description.

We also discussed what factors might make our machine slower, and he suggested that it could be the volume of content, the number of changed documents/items, or simply contention on the server that our tenant resides on.

Conclusions

Well, I don’t believe the volume of content would make any damn difference for an incremental crawl. For a full crawl, absolutely, but not an incremental.

Similarly, I don’t believe the number of changed documents to be that high, so I don’t believe that was the issue here.

However, contention with other users – that fits perfectly. During off-peak times, latency was much better. If you look at the graph of indexing times you can see it’s high during office hours, drops overnight, and it looks like it’s just starting to rise again the next morning.

Now, the impact of contention is difficult to judge – it’s all down to what the other tenants on your machine(s) are doing. And this is a key point of using Office 365; your service is bound by what resources Microsoft choose to throw at you. Using the cloud does not mean ‘as much as you like’, but rather ‘as much as you as you receive’, and all you can do is decided if the service level suits you, or not.

So when you’re deciding to use Office 365, you need to consider if what they’re offering is good enough, and if you’re willing to give up the control you have. And just be aware that the search crawler can be somewhat slow.