Latest interface: 0.3.1
Latest system: 010
coopers1862
User

47 posts

Posted on 25 November 2011 @ 10:17
http://img72.imageshack.us/img72/6941/diskbenchada1.png
http://img52.imageshack.us/img52/3024/diskbenchada2.png
http://img502.imageshack.us/img502/2136/diskpoolconfiguration.png
http://img189.imageshack.us/img189/8752/diskpooltank.png

Server:

HP N36L
8 GB RAM (zfs autotuning)
2 x 500 GB (RAID 0) - tank
1 x 250 GB - bsd - ZFSGuru OS
SAMBA

Client:

Windows XP & MacBook

Network:

1 GB netgear cg3100 cable modem/router

Speeds:

Copying large media file ("copy and paste") from client to server - 30 MB/sec
Copying large media file ("copy and paste") from server to client - 50 MB/sec

Any ideas what the problem could be?
Jason
Developer

806 posts

Posted on 25 November 2011 @ 19:38
Hey coopers,

According to your screenshots, you using an older web-interface version and I suspect an older system version as well (9.0-001 perhaps?). You should definitely not use the experimental 9.0-001 system image! Only use 8.3-001 (production) or the latest 9.0-003 (experimental) system images.

Your tuning is also done with just 1GiB memory, probably inside Virtualbox? You should reinstall on your final system with 8GiB memory, so the tuning is done for that amount of memory.

I suspect you used an older USB stick to install, right? In that case
1) boot from the USB stick again
2) update to 0.1.9 final (System->Update)
3) format your disks with GPT on Disks page (this inserts new boot code)
4) create a pool (250GB disk; the 500GB can be done later)
5) Install to your 250GB disk on the System->Install page, using either 8.3-001 or 9.0-003 system image, which you will need to download.

After installation, you can boot from your 250GB disk, and you should have modern system image, modern boot code and modern web-interface. The tuning should be updated for 8GiB RAM and your performance should be a lot better.
coopers1862
User

47 posts

Posted on 28 November 2011 @ 12:17edited 12:18 43s
I did everything from scratch like you said. Unfortunately, the speeds are only marginally better. I am basically "copying and pasting" a single 5 GB mkv file and manually timing the completion transfer speeds. There is probably a more efficient way of testing this!

I then set up the 2 x 500 GB under raid 0 with no redundancy.

These are my results, using SAMBA via OSX & Windows XP clients

Copying 5 GB mkv from client to server - ~40 MB/sec
Copying 5 GB mkv from server to client - ~50 MB/sec

These are my results, using NFS via OSX clients

Copying 5 GB mkv from client to server - N/A write permission due to funky GUID of OSX
Copying 5 GB mkv from server to client - ~50-60 MB/sec

Unfortunately, I don't have a Windows 7 or 8 machine, so there might be some speed limitations of Windows XP. I've also read Samba should be slower under OSX.

I'll try to borrow a Windows 7 or 8 laptop off my friend to do some further testing.

Totally stumped :(
coopers1862
User

47 posts

Posted on 29 November 2011 @ 00:45
These are the bench marks ran again on current stable. I think my graph's look a bit diff to others on here.

http://img100.imageshack.us/img100/3962/diskbenchmark.png
http://img593.imageshack.us/img593/9707/poolbenchmark.png
Adam
User

19 posts

Posted on 29 November 2011 @ 07:10
I like your little hp server. I probably would have purchased this if I had not build my own with a low power amd 160u 1.8ghz processor but our hardware is very similar. I too have slowish Samba shares and am wondering about tweaks and tips. Will certainly update here if I find something
Adam
User

19 posts

Posted on 30 November 2011 @ 02:31
I tried setting StrictHostKeyChecking to no. Default is 'ask' and the server will 'check' the client every time. Set to 'yes' and only particular machines will be able to access the server. My home network is behind a freebsd router with no wireless on the inside. So I'm comfortable opening access to all machines plugged into my switches.

Also I set loglevel to 2 rather than default '3' to cut down on overhead.

These changes seem to have quickend SAMBA for me. from 20ish MB/s to 30+. Would like to see 60 to 80 ultimatiely. So far this is simply pushing files from win7 based computers to zfsguru box through a gigabit switch.
Adam
User

19 posts

Posted on 1 December 2011 @ 02:14
Autotuning with prefetch enabled

Large files...Backups movies etc. (GiBish)

Client to server via Samba and gigabit switch 70 to 80 MB/s
Server to Client 50ish (probably limited by 750GB Seagate)

Happy ;) Will upgrade computers to ssd drives and use zfsguru for storage. All random Spinners can jbod for backup to zfsguru.
coopers1862
User

47 posts

Posted on 1 December 2011 @ 02:31edited 02:33 40s
On NFS, I'm now getting

75 MB/sec from server to client.

I can't test client to server because of the write permissions and GUID on OSX unfortunately. Theoretically, client to server should be faster right?

On SAMBA, the Windows XP & OSX clients are still stuck on 40 MB/sec. I'm beginning to think SAMBA is very slow on OSX & XP machines.

Still trying to borrow Windows 7/8 laptop off my mate, maybe SAMBA performance will improve with a newer OS?
coopers1862
User

47 posts

Posted on 3 December 2011 @ 07:06edited 07:13 39s
Okay!

I've done a mixture of tests on OSX, XP, and Win7 machines.

8 GB RAM - Performance tuning (use most ram + prefetching)
1 x 250 GB HDD - BSD operating system
2 x 500 GB HDD - RAID 0 - SAMBA SHARE

These are the rough speeds using all 3 OS's.

Copying from client to server - 30-35 MB/sec
Copying from server to client - 45-50 MB/sec

Using NFS for OSX

Copying from client to server - 30-35 MB/sec
Copying from server to client - 65-75 MB/sec

Note: Shouldn't server to client be faster? :O

Here are the HDD benchmarks

Test size: 32.000 gigabytes (GiB)
Test rounds: 1
Cooldown period: 2 seconds
Sector size override: default (no override)
Number of disks: 2 disks
disk 1: gpt/disk2
disk 2: gpt/disk3


  • Test Settings: TS32; TR1;

  • Tuning: KMEM=11.9g; AMIN=1.6g; AMAX=4g;

  • Stopping background processes: sendmail, moused, syslogd and cron

  • Stopping Samba service


  • Now testing RAID0 configuration with 1 disks: cWmRd@
    READ: 69 MiB/sec = 69 MiB/sec avg
    WRITE: 65 MiB/sec = 65 MiB/sec avg

    Now testing RAID0 configuration with 2 disks: cWmRd@
    READ: 140 MiB/sec = 140 MiB/sec avg
    WRITE: 137 MiB/sec = 137 MiB/sec avg

    Now testing RAIDZ configuration with 2 disks: cWmRd@
    READ: 75 MiB/sec = 75 MiB/sec avg
    WRITE: 62 MiB/sec = 62 MiB/sec avg

    Now testing RAID1 configuration with 2 disks: cWmRd@
    READ: 118 MiB/sec = 118 MiB/sec avg
    WRITE: 65 MiB/sec = 65 MiB/sec avg

    Done

    My speeds going from the server to the client are much faster than client to server. Is this unusual?

    I am thinking it is either my hard drives that is the problem, or I have a slow modem/router.

    Thoughts?
    danswartz
    User

    252 posts

    Posted on 3 December 2011 @ 07:55
    I'm assuming client and server on the same LAN, so modem/router should be moot. Try setting sync=disabled and retest.
    coopers1862
    User

    47 posts

    Posted on 3 December 2011 @ 08:32edited 08:33 15s
    danswartz wrote: I'm assuming client and server on the same LAN, so modem/router should be moot. Try setting sync=disabled and retest.
    Hi, thanks for the reply. Yes, client and server on the same LAN hooked up to single modem/router.

    But, can you explain how I can set sync=disabled? Is this in the modem/router, or somewhere in zfsguru?
    danswartz
    User

    252 posts

    Posted on 3 December 2011 @ 17:21
    Do do it at the zfsguru command line. Like this:

    zfs set sync=disabled tank

    or whatever the pool name is...
    coopers1862
    User

    47 posts

    Posted on 4 December 2011 @ 07:02
    danswartz wrote: Do do it at the zfsguru command line. Like this:

    zfs set sync=disabled tank

    or whatever the pool name is...

    is that the same thing as vfs.zfs.zil_disable? I think it made the speeds even slower :O

    I'm beginning to think I've got dodgy HDD's. I'll post the SMART results, tell me if you think these could impact performance.
    coopers1862
    User

    47 posts

    Posted on 4 December 2011 @ 07:07edited 07:07 32s
    Here are the results. I've even zero'd the HDD but the SMART still returns the same results

    http://img694.imageshack.us/img694/4139/ada1smart.png
    http://img141.imageshack.us/img141/9273/ada2smart.png

    ADA1, I've got the cable errors. But the number has never increased after HDD usage, so I think this is a good thing.

    Ditto for ADA2.

    Not sure if this could be the performance problem?
    Last Page

    Valid XHTML 1.1