MDS performance on R80.20 or above with xfs

[Kaspars_Zibarts]

Just wondering - whats the general feeling regarding R80.20 MDS file system for those who have upgraded from R80.10 and EXT3 - is the XFS filesystem faster and more efficient? We will be upgrading shortly from R80.10 (current take level) and were hoping for some increased performance in disk read/write.

[PhoneBoy]

From what I've been told, xfs is faster/more efficient than ext3.
Have not seen any specific performance data, though.

[Kaspars_Zibarts]

It's interesting as when i run some basic dd commands to compare two (existing production VM with R80.10 / ext3 and newly built VM running R80.20 with xfs, both exactly the same VMs parameter wise), it was rather mix bag of results - large files seemed to perform better on ext3 whilst smaller on xfs. That's why I wanted to ask as we will be upgrading shortly.

[Timothy_Hall]

My understanding is that XFS is well-suited for handling a large number of threads trying to access the disk path simultaneously (parallelized access via allocation groups ensures multiple threads can perform I/O simultaneously on the same volume).  This is constantly the case on an SMS/MDS with incoming firewall logs being written, SOLR log indexing, SmartEvent correlation, security policy configuration-related postgres database operations, etc.  XFS is also optimized for working with very large files (like postgres database extents) and works especially well with SSDs which are starting to become more and more common on Check Point appliances. 

It is doubtful that a single dd I/O operation will reflect these XFS advantages.

[Kaspars_Zibarts]

Upgraded our MDSes a week ago. One thing definitely stands out - creating and restoring backups. Not only that backup size has reduced but also seems that disk read/write has much better performance at least in this area.  😎