2

I am running a Nexentastor server with the follow specs:

dual core (core2duo) cpu 4GB of ram 1x 160GB SATA2 drive for the OS 6x 1.5TB Drives Sata2 drives in a RaidZ1

copying files is fine, fairly fast, I can normally max gigabit so no complaints.

But when it comes to manipulating files especially deleting files that is when it gets VERY painfull.

For example, Deleting a 705MB folder that has about 30 files in it took 2 minutes and 30 seconds (using the time command)

any commands I can run at the same time as rm to narrow this down?

thank you.

Luma
  • 1,460
  • 4
  • 19
  • 31
  • What model of SATA drives, what controller, and what cache/battery? – Chris Thorpe Mar 26 '11 at 06:27
  • The drives are ST31500341AS using the on board ICH10 controller. ZFS is supposed to be a replacement to those expensive raid controllers so I figured I would try it. – Luma Mar 26 '11 at 21:05
  • IMHO ZFS is unequivocally *better than* expensive hardware raid controllers, but it isn't necessarily cheaper to deploy optimally. – Skyhawk Mar 27 '11 at 02:41

2 Answers2

3

If you do not have a separate slog device, disabling ZIL will speed writes tremendously if you can afford to lose delayed writes in the event that power is lost. It seems likely that you can, because your specs indicate that you can afford to lose your entire storage server if your OS drive fails!

ZFS random read and write performance is utterly phenomenal, even with 7200 rpm drives, when the storage server has plenty of horsepower and RAM and both L2ARC and ZIL are enabled using fast SSDs. Write performance is not impressive if you are using spindles only, unless ZIL is disabled.

Skyhawk
  • 14,200
  • 4
  • 53
  • 95
  • Hello Miles, Thanks for the info, I might try disabling ZIL. The server is indeed very non-critical. It is something I use at home to store media and backups of my desktop since I use Raid 0 in my desktop. I surely can't afford striped SSD's for L2ARC and mirrored SSD's for ZIL :) With Zil disabled if the power does go out mid write will I loose data? the Server is on a UPS which shuts the box down in the event of a power outage. Thank you again. – Luma Mar 26 '11 at 21:03
  • As you've indicated, your server is non-critical and you care more about performance and budget than fault tolerance. Go ahead and disable ZIL and see what happens. Pending writes will be stored only in RAM, instead of on disk, and I suspect you will see a tremendous performance boost. By the way, you need nothing so fancy as striped or mirrored SSDs to implement L2ARC and ZIL: even a single Intel X25M for L2ARC and a single Intel X25E for ZIL will do the job. You would not lose data immediately if your X25E failed unless you lost power too, because all pending writes are also stored in RAM. – Skyhawk Mar 26 '11 at 21:26
1

RAIDZ1 will limit your write speed to the equivalent of single disk per vdev.

705MB in 150 seconds is 4.7MB/s which seems normal (i.e. horrible) for SATA disks and highly random workloads.

https://blogs.oracle.com/bonwick/entry/raid_z

alanc
  • 1,500
  • 9
  • 12
gtirloni
  • 5,746
  • 3
  • 25
  • 52
  • I don't agree that 4.7MB/s is "normal" for SataII Disks. I don't expect high end SSD speeds but give me a break. This Machine is doing nothing else, It is in my home and serves files to a single workstation which was doing nothing else at the time but deleting this. these disks formatted with other file systems are much faster. is ZFS really that bad? what is the hype about then? – Luma Mar 26 '11 at 05:13
  • Agree. Especially that DELETING A FILE IS NOT OVERWRITING IT. It is just changing some directory entries. – TomTom Mar 26 '11 at 06:09
  • 1
    [This is a good explanation of why ZFS write performance can be slow](http://fixunix.com/solaris-rss/582874-solaris-zfs-synchronous-writes-zil-explained.html) unless you set up a dedicated ZIL device -- preferably a write-optimized SSD -- and/or disable ZIL entirely. Note that disabling ZIL is *never* an option on a critical production server. However, in non-critical home/test environments where performance and budget are more important than fault tolerance, it is a reasonable option to consider, so long as you're making an educated decision. – Skyhawk Mar 26 '11 at 21:34