As promised, I'm adding the Jetstress test here to compare ReFS and NTFS. I tested once with a database and logs on one volume and 4 databases with separate log and database volumes (8 volumes). All volumes of the Windows VM were stored on the same XtremIO LUN. There were no other changes.
The first part can be found here:
Server 2016: Speed comparison between ReFS and NTFS (Part 1)
First of all, the tests with a database:
Jetstress 1 database
NTFS may start again. The drive was formatted with 64K blocks (without quick formatting). The ReFS drive was also created with 64K blocks without quick formatting. The FileIntegrity feature is switched off for ReFS.
NTFS
After the run, Jetstress reports 2459 IOs (read) and 2037 IOs (write):
The storage apparently calculates the IOs slightly differently here and displays 10,355 IOs:
The data transfer rate is 339 MB per second:
ReFS
ReFS is now just ahead of Jetstress, with 2526 IOs for reading and 2090 IOs for writing:
The storage view is also shown here for comparison, with even fewer IOs displayed at this point:
The bandwidth is also somewhat lower:
However, I can't guarantee that the screenshots were taken at the same time, so I would rather trust Jetstress here.
Jetstress 4 Databases
Here are the test results with 4 databases, each database was saved on a volume. The logs were also saved on separate volumes. This test therefore uses 8 volumes. All 8 volumes were each created with 64K blocks without quick formatting:
NTFS
4 databases on NTFS volumes achieve just under 3500 IOs when reading and 2900 IOs when writing:
The storage shows about 27000 IOs here:
The throughput is 868 MB per second:
ReFS
With 4 databases, each database delivers around 3400 IOs when reading and 2800 IOs when writing:
Conclusion
The results are very close together, so that no direct winner can be chosen. IOMeter showed advantages for NTFS, but Jetstress places both file systems almost equally. In the first part I had configured IOMeter with 4K IOs 75 % read and 25 % write. In my opinion, Jetstress provides a better overview here, as different IO sizes are retrieved.
Since ReFS is the recommended file system for Exchange databases and log files, there is nothing to be said against it, at least from a performance point of view.
Appendix
The Jetstress HTML files can be downloaded directly for a better overview:
Natürlich geht es um die Partition innerhalb einer VM. Was du in Hyper-V machst (also wo der Container liegt), bekommt die VM ja gar nicht mit.
Ich lese und höre immer wieder, dass ReFS mittlerweile für Exchange 2016 empfholen wird. Jedoch finde ich keine offizielle Quelle, die das schreib. Lediglich in den Systemanforderungen steht „können optional als ReFS“
Festplattenpartitionen, die nur die folgenden Dateitypen enthalten, können optional als ReFS formatiert werden:
Partitionen, die Transaktionsprotokolldateien enthalten
Partitionen, die Postfach-Datenbankdateien enthalten
Partitionen, die Inhaltsindizierungsdateien enthalten
Quelle: https://technet.microsoft.com/de-de/library/aa996719(v=exchg.160).aspx
Hallo Franky,
gilt die Empfehlung für ReFS auch innerhalb einer VM? Der physische Host ist mit ReFS formatiert, die Partition innerhalb der VM bisher NTFS. Danke