1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Hardware performance and Proxmox

Discussion in 'Mail Gateway: Installation and configuration' started by tom, Oct 12, 2007.

  1. tom

    tom Proxmox Staff Member
    Staff Member

    Joined:
    Aug 29, 2006
    Messages:
    11,001
    Likes Received:
    48
    The filtering process triggers quite a lot of database connections and several postfix connections. Therefore, the performance of the hard disk system is one very important parameter, especially fsyncs/sec. We introduced a small benchmark tool (proxperf) which you can run on the console (use it when the server is not under load).

    Here is a sample output (one cpu, single sata harddrive):
    ________________________
    proxmox:~# proxperf
    CPU BOGOMIPS: 4289.64
    REGEX/SECOND: 410814
    HD SIZE: 6.89 GB (/dev/sda2)
    BUFFERED READS: 116.38 MB/sec
    AVERAGE SEEK TIME: 8.09 ms
    FSYNCS/SECOND: 1084.51
    DNS EXT: 46.26 ms
    DNS INT: 1.05 ms (domain.com)
    DNSBL: 35.47 ms (zen.spamhaus.org)
    proxmox:~#
    _______________________

    for actual hard drives, the result for FSYNCS/SECOND should be around 1000 or higher. Especially when you use hardware raid without write cache (and batteries backup), the performance concerning FSYNCS/SECOND could be quite bad (around 100) and the overall performance not as good as you expect.

    To get the best performance out of Proxmox:
    • Fast CPU, multi core cpu´s are recommended
    • Hardware raid controllers with batteries backup and write cache enabled and fast hard drives (Raid1, Raid5 or Raid10)
    • Fast DNS servers (maybe locally cached)
     
    #1 tom, Oct 12, 2007
    Last edited: Dec 7, 2007
  2. alex-schaefer

    alex-schaefer New Member

    Joined:
    Oct 13, 2007
    Messages:
    8
    Likes Received:
    0
    I can assertively confirm the importance of write cache enabled on the RAID systems. We have installed 2xProxmox 2.0 Anti Spam Gateways on HP DL380 G4 servers, with:
    1x3,0 GHz Xeon CPU,
    4 GByte RAM,
    3x72 GByte (15.000 RPM) SCSI HDDs,
    configured in RAID 5 with internal SmartArray 6i SCSI controller, without BBWC option kit (BBWC - battery backuped write cache). As result we can complain about realy bad I/O performance of the system:
    promox01:~# proxperf
    CPU BOGOMIPS: 12004.51
    REGEX/SECOND: 408455
    HD SIZE: 132.55 GB (/dev/cciss/c0d0p2)
    BUFFERED READS: 205.13 MB/sec
    AVERAGE SEEK TIME: 8.88 ms
    FSYNCS/SECOND: 123.90
    DNS EXT: 75.34 ms
    DNS INT: 2.31 ms (burkert.com)


    As comparison test we did install the same Proxmox configuration on my Windows XP workstation as VMWare appliance running under VMWare server 1.0.3 with 512 MB virtual RAM. The workstation is an HP PC, with 2,0 GHz Pentium4 CPU, 1 GByte RAM and 1xSATA HDD with 8 MByte cache(!!!). The performance (FSYNCS/SECOND) is much more better then on the high speed RAID configuration:
    FSYNCS/SECOND: 516.10

    Of course, after such blame comparison results we did order the BBWC option kit for both of our Proxmox mail gateways. I will report the new benchmarks results, after BBWC installation.
     
  3. alex-schaefer

    alex-schaefer New Member

    Joined:
    Oct 13, 2007
    Messages:
    8
    Likes Received:
    0
    Performance vlaue with BBWC installed

    As promissed, here is the new proxperf output, after we have installed the Battery Backuped Write Cache option on our servers:

    CPU BOGOMIPS: 12004.33
    REGEX/SECOND: 245574
    HD SIZE: 132.55 GB (/dev/cciss/c0d0p2)
    BUFFERED READS: 170.66 MB/sec
    AVERAGE SEEK TIME: 7.41 ms
    FSYNCS/SECOND: 1124.47
    DNS EXT: 376.00 ms
    DNS INT: 2.58 ms


    OK, the valur FSYNC/SECONDS varies, depending on current storage usage, because the server is running in production environment with over 120.000 mails/day.
     

Share This Page