pinkerton wrote:Just got off a phone call with one some Veeam technican (3rd party) discussing WAN acceleration. At the end we talked about ReFS issues and he told me that Veeam now recommends to use 4K formatting for ReFS volumes instead of 64K. Is that true? We implemented ReFS a year ago and IIRC the recommendation was 64K back then. We also have problems with our ReFS setup and are waiting for the hotfix. However, if 64K is also cause of the issue we are going to change to 4K as well.
I'm not sure why someone from support would say that. We recommend 64K not just due to performance of merges/synthetics, although that is a factor, but primarily due to the fact that using 4K clusters leads to the problems documented in this forum thread much more often and in worse ways, even with much smaller datasets. For example, with 4K clusters I've personally seen ReFS lockups on repositories as small as a few TBs in size, and common (nightly), lockups with 100TB repositories when they contained <10TB of data. On the other hand, with 64K clusters, these smaller repos were stable as long as they had enough memory. Of course, 64K clusters did not eliminate all of the issues, as this thread well documents, it just made them less likely.
If the fixes for ReFS in the latest patches fix all of the resource consumption and performance issues even with repositories at scale, even when used with 4K clusters, it might be possible for us to change this recommendation in the future, but that will only be proven over time in the field as we see the results. Initial results from customers that I know were using 4K clusters, and were testing the patches, seems promising, but it will take some time and more results before we really know.
For now, let's just all test with these new ReFS updates and hope that we may finally be near the end of this long road with ReFS stability and please post your results here. It's a great technology when it works, but it's been a bumpy ride to this point.