Re: Dying net6501 servers
David Ruggiero <[email protected]> Thu, 26 Jan 2017 11:32:30 -0800
| Newsgroups | gmane.comp.hardware.soekris.technical |
|---|---|
| Message-ID | <CAM9qDQdZ14oY5T1QBnkMPTm9xkqLM-NZ0obU5XFpDvTkM8ywmQ@mail.gmail.com> |
--===============1520073298== Content-Type: multipart/alternative; boundary=001a114fca423d187d0547046a24 --001a114fca423d187d0547046a24 Content-Type: text/plain; charset=UTF-8 Apologies for the update to the fossilized thread here, but another very-close-to-home data point. Our last remaining net6501 died week before last - it refused to boot, had no console message, and would only show the red error light. We RMA'd it to Soekris but they replied with a terse, "sorry, can't repair, and also out of warranty". This unit saw only light duty in a cool environment, was run with a Soekris-supplied PSU on a good AC line filter, had the bigger (aluminum) heatsink from day one, and was deliberately _underclocked_ to 600mhz for its whole life (for power consumption and supposed reliability). The case also had an active cooling fan in it the entire time and the CPU temp was monitored regularly. Unless we could have fed it milk and cookies we couldn't have babied it more than we did - and yet it's still mysteriously dead and unfixable as so many others are. Thanks all for the alternative replacement strategy ideas (below). We will likely be buying SuperMicros in the short-term, but really think the Raspberry Pi 3 idea has a lot of potential if we can get the IO working (time to bone up on VLAN-fu here). cheers, David ---- On Sat, Sep 3, 2016 at 5:18 PM, David Ruggiero <[email protected]> wrote: You might laugh, but for embedded applications where we only need 1-3 hard-wired Ethernet ports, and don't need full GigE, we're actually starting to investigate using....Raspberry Pi 3s . (More and more often now, 802.11n/ac wireless is the primary connection medium for a lot of our client devices, so less need these days for wired Ethernet.) The RPi3 includes 802.11n, 10/100 ethernet, 4-port USB 2.0 (supporting Eth-to-USB converters), and Bluetooth low-power - all standard. Cost is incredibly cheap - just $35 qty 1, plus some small change for whatever you decide you need for a case and PSU. And their performance is astonishingly high - no firm benchmarks yet that I've seen, but looks like it's going to pen out at ~50% faster than the net6501-70 on single-threaded tasks, and even more on multi-threaded ones (given it has four (!) cores [*] ). At those price/performance numbers, it becomes harder and harder to justify a platinum-priced, zero-support, non-updated-in-umpteen-years platform like Soekris. (One might argue the Soekris is more robustly built - whatever that means in the Real World, given all the failure reports here - but the price of a single net6501-70 will buy over a half-dozen RPi 3s to have around as spares, including extra cases and PSUs. ) We haven't quantified the real power consumption yet, but everything points to the RPi 3 using less, sometimes a LOT less, juice than the net6501. We'll post our experiences as we evaluate and test. > >> "I'm disappointed to find that Soekris has been so quite about these problems. If this was a preventable > >> heat issue with the CPU, I'm doubly disappointed since I asked about CPU power management when > >> I first got the boards. Soekris provided no solution but to run the CPUs at full power when they're very > >> capable of running at much lower power with "normal" intel BIOS configurations." Yup. We are pretty well through with Soekris at this point unless there's no other choice, given the hardware reliability (witness this thread), zero communication, and total lack of updates (hardware, bios, and documentation). And that's all before the hugely out-of-whack price/performance/power ratio is considered. We can also get support in a heartbeat from literally thousands of expert users of the platform - quite different. [[ * See: https://blog.pivotal.io/labs/labs/faster-soekris-net6501-raspberry-pi - a 2013 comparison with between the old RaspPi Model B vs the Soekris 6501-70. Soekris at that time was 1.7x the speed of a RPi model B for a single-threaded task. But the new RPi 3 is 2.5x+ faster than the model B for single-threads and 3x+ faster for multiple-threads, so given those numbers it's reasonable to think it's going to be 50% faster than the net6501-70 in the real world, at a bare minimum.]] > On Mon, Mar 28, 2016 at 3:33 AM, Ben Kochie <[email protected]> wrote: > > We endedup replacing our soekris boards being used as routers with VyOS with Ubiquiti EdgeRouter Lite since we only needed 2 router ports for this install. We used a 1U bracket so we could install 2 of them in 1U, saving a bit of space. > > https://www.rfarmor.com/index.php/toughswitch-mounts/1uurspoe.html > > I'm disappointed to find that Soekris has been so quite about these problems. If this was a preventable heat issue with the CPU, I'm doubly disappointed since I asked about CPU power management when I first got the boards. Soekris provided no solution but to run the CPUs at full power when they're very capable of running at much lower power with "normal" intel BIOS configurations. > > On Fri, Mar 18, 2016 at 8:46 AM, Andreas Steinel <[email protected]> wrote: > >>> Can anyone on the list share which vendor, model, and OS to which they have switched? > >> We switched to APU Boards http://www.pcengines.ch/apu.htm --001a114fca423d187d0547046a24 Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: quoted-printable <div dir=3D"ltr">Apologies for the update to the fossilized thread here, bu= t another very-close-to-home data point. Our last remaining net6501 died we= ek before last - it refused to boot, had no console message, and would only= show the red error light. We RMA'd it to Soekris but they replied with= a terse, "sorry, can't repair, and also out of warranty".<di= v><br></div><div>This unit saw only light duty in a cool environment, was r= un with a Soekris-supplied PSU on a good AC line filter, had the bigger (al= uminum) heatsink from day one, and was deliberately _underclocked_ to 600mh= z for its whole life (for power consumption and supposed reliability). The = case also had an active cooling fan in it the entire time and the CPU temp = was monitored regularly. Unless we could have fed it milk and cookies we co= uldn't have babied it more than we did - and yet it's still mysteri= ously dead and unfixable as so many others are.</div><div><br></div><div>Th= anks all for the alternative replacement strategy ideas (below). We will li= kely be buying SuperMicros in the short-term, but really think the Raspberr= y Pi 3 idea has a lot of potential if we can get the IO working (time to bo= ne up on VLAN-fu here).</div><div><br></div><div>cheers,=C2=A0</div><div>Da= vid</div><div><br></div><div>----<br><div><br>On Sat, Sep 3, 2016 at 5:18 P= M, David Ruggiero <<a href=3D"mailto:[email protected]">thatseatt= [email protected]</a>> wrote:<br>You might laugh, but for embedded applica= tions where we only need 1-3 hard-wired Ethernet ports, and don't need = full GigE, we're actually starting to investigate using....Raspberry Pi= 3s . (More and more often now, 802.11n/ac wireless is the primary connecti= on medium for a lot of our client devices, so less need these days for wire= d Ethernet.)<br><br>The RPi3 includes 802.11n, 10/100 ethernet, 4-port USB = 2.0 (supporting Eth-to-USB converters), and Bluetooth low-power - all stand= ard. Cost is incredibly cheap - just $35 qty 1, plus some small change for = whatever you decide you need for a case and PSU. And their performance is a= stonishingly high - no firm benchmarks yet that I've seen, but looks li= ke it's going to pen out at ~50% faster than the net6501-70 on single-t= hreaded tasks, and even more on multi-threaded ones (given it has four (!) = cores [*] ).=C2=A0 At those price/performance numbers, it becomes harder an= d harder to justify a platinum-priced, zero-support, non-updated-in-umpteen= -years platform like Soekris.<br><br></div><div>=C2=A0(One might argue the = Soekris is more robustly built - whatever that means in the Real World, giv= en all the failure reports here - but the price of a single net6501-70 will= buy over a half-dozen RPi 3s to have around as spares, including extra cas= es and PSUs. )<br><br>We haven't quantified the real power consumption = yet, but everything points to the RPi 3 using less, sometimes a LOT less, j= uice than the net6501. We'll post our experiences as we evaluate and te= st.<br><br></div><div>> >> "I'm disappointed to find that= Soekris has been so quite about these problems.=C2=A0 If this was a preven= table<br>> >> heat issue with the CPU, I'm doubly disappointed= since I asked about CPU power management when<br>> >> I first got= the boards.=C2=A0 Soekris provided no solution but to run the CPUs at full= power when they're very<br>> >> capable of running at much lo= wer power with "normal" intel BIOS configurations."<br><br>Y= up. We are pretty well through with Soekris at this point unless there'= s no other choice, given the hardware reliability (witness this thread), ze= ro communication, and total lack of updates (hardware, bios, and documentat= ion). And that's all before the hugely out-of-whack price/performance/p= ower ratio is considered.=C2=A0 We can also get support in a heartbeat from= literally thousands of expert users of the platform - quite different.<br>= <br><br>=C2=A0[[ * =C2=A0See: <a href=3D"https://blog.pivotal.io/labs/labs/= faster-soekris-net6501-raspberry-pi">https://blog.pivotal.io/labs/labs/fast= er-soekris-net6501-raspberry-pi</a> =C2=A0 - a 2013 comparison with between= the old RaspPi Model B vs the Soekris 6501-70. Soekris at that time was 1.= 7x the speed of a RPi model B for a single-threaded task. But the new =C2= =A0RPi 3 is 2.5x+ faster than the model B for single-threads and 3x+ faster= for multiple-threads, so given those numbers it's reasonable to think = it's going to be 50% faster than the net6501-70 in the real world, at a= bare minimum.]]</div><div><br></div><div>> On Mon, Mar 28, 2016 at 3:33= AM, Ben Kochie <<a href=3D"mailto:[email protected]">[email protected]</a= >> wrote:<br>> > We endedup replacing our soekris boards being use= d as routers with VyOS with Ubiquiti EdgeRouter Lite since we only needed 2= router ports for this install.=C2=A0 We used a 1U bracket so we could inst= all 2 of them in 1U, saving a bit of space.<br>> > <a href=3D"https:/= /www.rfarmor.com/index.php/toughswitch-mounts/1uurspoe.html">https://www.rf= armor.com/index.php/toughswitch-mounts/1uurspoe.html</a><br>> > I'= ;m disappointed to find that Soekris has been so quite about these problems= .=C2=A0 If this was a preventable heat issue with the CPU, I'm doubly d= isappointed since I asked about CPU power management when I first got the b= oards.=C2=A0 Soekris provided no solution but to run the CPUs at full power= when they're very capable of running at much lower power with "no= rmal" intel BIOS configurations.<br><br>> > On Fri, Mar 18, 2016= at 8:46 AM, Andreas Steinel <<a href=3D"mailto:[email protected]">a.s= [email protected]</a>> wrote:<br>> >>> Can anyone on the list= share which vendor, model, and OS to which they=C2=A0have switched?<br><br= >> >> We switched to APU Boards <a href=3D"http://www.pcengines.ch= /apu.htm">http://www.pcengines.ch/apu.htm</a><br></div></div></div> --001a114fca423d187d0547046a24-- --===============1520073298== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech --===============1520073298==--