QuickAssist hardware decision



  • Hi
    I am trying to make a decision about which motherboard to buy.

    Option 1
    Supermicro X11SBA-LN4F
    N3700 quad core 6 watt
    AES-NI support
    Quad intel I210 lan

    Option 2
    Supermicro A1SRi-2558F
    C2558 quad core 15 watt
    AES-NI support
    QuickAssist support
    Quad intel i350 lan

    I use a openVPN client for PIA with a 150/20 connection with. I'm planning on using packages  such as Squid and Snort.
    Both board seems closely matched for running pfSense.

    My question
    Will the extra cost of the C2558 to get the QuickAssist help Or am I better off with the N3700 and save money and power??

    Thanks
    Dave



  • hi,

    well this is always in your hands to decide if the price difference make sense or not. When ASE-NI will be fully supported for sure it will reduce CPU stress alot. In any case Pfsense is not cpu intense if you don't have HAVP + ids/ips stuff on. this may increase the usage alot.

    https://forums.servethehome.com/index.php?threads/pfsense-and-openvpn-performance-recommendations.5625/

    It may also bring advantages on other things like snort and compression jobs:
    http://marc.info/?l=openbsd-misc&m=143773664731133&w=2

    OpenVPN and quickassis:
    https://community.openvpn.net/openvpn/ticket/301
    so it's in 2.4 version

    read this thread :
    https://forum.pfsense.org/index.php?topic=107329.msg606808#msg606808



  • get the atom 2558 board. not much more money, for a lot more



  • The C2558 will be much more offering such options, functions and is well proven tested
    and reported here in that forum. It will be serving more benefit for your entire network
    in my eyes. This might be not really interesting today, but if the pfSense team is bringing
    that Intel QuickAssist up and working well it might be the better choice.



  • It depends on how much of a premium you pay. If it's a couple of bucks, get the C2558. If it's double, then get the N3700 and then upgrade in a year when skylake SOCs have better availability. You're basically making a bet that the QAT on the C2558 (a three year old part which is currently useless) will be more valuable in the future, and I think that's a bad bet. The maximum performance of QAT is lower than the real world AES-NI performance on current intel skylake cores, and the QAT implementation on the Atom chips is different than the QAT add in cards (and probably any future SOCs with QAT) meaning that it's relatively unlikely that they'll see a lot of future development investment from intel. It's been true for a long time in the industry that unless a hardware crypto part gets you an immediate benefit that you can't live without, it's a bad investment.