From owner-freebsd-hackers Tue Aug 27 04:05:05 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id EAA29239 for hackers-outgoing; Tue, 27 Aug 1996 04:05:05 -0700 (PDT) Received: from ns.eu.org (valerian.glou.eu.org [193.56.58.251]) by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id EAA29208 for ; Tue, 27 Aug 1996 04:04:56 -0700 (PDT) Received: (from uucp@localhost) by ns.eu.org (8.7.3/8.7.1/951117) with UUCP id NAA23503; Tue, 27 Aug 1996 13:02:03 +0200 (MET DST) Received: (from regnauld@localhost) by tetard.glou.eu.org (8.7.5/8.7.3/tetard-uucp-2.7) id MAA09631; Tue, 27 Aug 1996 12:40:59 +0200 (MET DST) From: Philippe Regnauld Message-Id: <199608271040.MAA09631@tetard.glou.eu.org> Subject: Re: -current kills harddrives To: joerg_wunsch@uriah.heep.sax.de Date: Tue, 27 Aug 1996 12:40:59 +0200 (MET DST) Cc: hackers@freebsd.org (hackers) In-Reply-To: <199608270659.IAA23681@uriah.heep.sax.de> from J Wunsch at "Aug 27, 96 08:59:30 am" X-Mailer: ELM [version 2.4ME+ PL15 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8bit Sender: owner-hackers@freebsd.org X-Loop: FreeBSD.org Precedence: bulk J Wunsch écrit / writes: > As Philippe Regnauld wrote: > > > BTW, I'm writing up a "Hardware maintenance" section -- could be > > useful in the handbook (proper ventilation, do's and dont's), or > > does this seem too way out of line with FreeBSD ? > > The extended cooling requirements of 7200 rpm drives should of course > be mentioned there. They are :-) Comments and critics welcome. ->8--->8--->8--->8--->8--->8--->8--->8--->8--->8--->8--->8--->8--->8--->8-- Hardware maintenance guidelines FreeBSD is a high performance operating systems, and as such, it is often used as the driving force behind high performance servers. Unfortunately, experience shows that selecting the right OS and the right components (See "PC Hardware Compatibility") is sometimes not enough; there are several technical issues involved that will (hopefully) be covered here, such as correct cabling, proper insulation and ventilation, thermal constraints, etc... These are mostly tips and hints acquired from personal experience in working environments -- they are not in any case absolute rules (i.e.: your mileage may vary). * COMPONENT ASSEMBLY * Casings If you have ever assembled a PC yourself, you may have noticed some aberrations in the design and/or basic construction of the casing. Remember that the majority of these were designed with cheapness in mind, not fault tolerance. To start with, try and avoid all forms of "small tower" designs, as they have a tendency to double as ovens -- this is mainly due to the amount of cabling generally present in your average PC (see Cabling). High towers are recommendable for three reasons: o they offer room for more devices and disks in particular; o they provide better airflow as more space is available; o there usually is room for additional fan units Depending on the make and model of the casing, they may or may not offer facilities that are often useful : o ventilation slots on the underside of the cabinet o clip-on facade or hinged front door provides easy access to the disk units o screwed in motherboard mount-plate avoids excess manipulation of the motherboard (i.e.: unclipping nylon columns, sliding the board out). Some manufacturers do make these kinds of casings, though they are usually not cheap (count at least $150). For real easy access to cabling, disks, bus cards, I can only recommend rack-mount units, with its reliability (redundant power supplies, UPS support)... and price! Also, always make sure that the cables of the power supply and led/switch connectors are long enough to reach the mainboard (I'm serious about this, it happens often !), and that there are enough mainboard support holes, so that half of it doesn't hang unsupported. Expansion cards There are not many restrictions concerning expansion card installation and/or manipulation other than the usual static precautions and placement constraints due to available cabling. Still, some points should be respected as a safe measure, though they should not be required with quality power supplies and mainboards. Note: for those of you using 486s with VLB (good luck...), remember that most, if not all, mainboards that offer 3 VLB slots only allow 2 of those to do bus mastering. This may not seem like a big issue, but I have seen peak disk I/O rise from 2 Mb/sec to 3 Mb/sec by simply moving the I/O card to another VLB slot. Placement of the expansion cards, whether they be ISA, EISA, PCI or VLB should try and follow these recommendations: o avoid placing video cards too close to the power supply. It might cause static or other kind of video disturbances with badly shielded power supplies, or sensitive RAMDACs on cheap cards. If it does occur, try moving the card further 'down', away from the power supply. If this is not possible (PCI cards), get a bigger casing, or switch video cards; o same advice for audio cards: they are known to be sensitive to electromagnetic disturbances, which may cause humming and electric noises in the speaker output. Once again, try to move the sound card as far away from the power supply, cables permitting; In general, if a problem occurs that seems to be of a hardware nature (stray IRQs, erratic and/or intermittent response of a device or plain refusal to boot), try and switch the cards around. Sadly, this has worked in many cases for me with cheap mainboards or expansion cards. Cabling - Power cabling - When setting up your system, try to envision future needs you may have: it is highly recommended that you buy a power supply with enough connectors in the first place, rather than use "Y" power splitters. These are often of bad manufacture and can cause hard disk locks, bus resets and other mishaps through inconsistent of faulty power supply. If you do have to use them, ALWAYS check if the colors match (yellow to yellow, red to red, etc...): it has once happened that 5V (red) was cross-wired to 12V (yellow) on such an adapter :-( - Drive cabling - There are simply two cases for disk configurations: non-SCSI and SCSI configurations :-) It is not that SCSI is a sensitive or otherwise unreliable subsystem, but it is a high performance one, and should be treated with the corresponding awareness. Regarding IDE disks : IDE disks sometimes seem to behave strangely (no probing at boot, slave drive is not seen) depending on which drive is master or slave, and which one of these is first on the cable (though officially this should not have any influence on the way the disks operate). The best here is to try different configurations and see the results. See the section "What is SCSI" for the description of the SCSI subsystem, bus types and cabling recommendations. Keep in mind the following points : o always terminate both ends of the bus with the right terminator -- this means SCSI-I (big centronics-type connector) terminators for SCSI-I busses (also known as passive terminators), and SCSI-II (small connector) for busses containing SCSI-II devices (active terminators); o internal termination by the drives/devices themselves (i.e.: jumper activated) should be avoided -- use clip-on terminators that fit on the ribbon cable like any device; o use the right cabling for external devices : SCSI-I cables and SCSI-II cables do not have the same electrical specifications. Also respect maximum lengths - Ribbon cables - Always try and use cables that are long enough, especially for disk units: it is better to have spare cable rather than subject a barely long enough one to stress by twisting or folding it. Also, long enough cables will let you push them out of the way a bit, which is important for correct ventilation and air flow. * PLACING STORAGE DEVICES * This is probably the most important point here. It is crucial that you let storage devices (disks, CD Rom units, tape drives) have enough air for them to function correctly. Without ventilation, you can expect your 7200RPM drives to last a couple of hours at the most, if they do not have built-in thermal switches (recent ones do). There are a few things to know about ventilation: o ventilation is not simply exposing your unit to open air -- that may help, but it will in most cases only heat up the air around it; o air acts as a fluid, and it must circulate to be efficient. A closed cabinet with two fans blowing outward is a good way to achieve this; o having the fans simply blow INTO the cabinet will not achieve cooling: this will not help the hot air escaping. The idea is to have the air be pushed out of the cabinet. Maximum effect is obtained with two fans, one extracting and one intaking, if and only if those are running at the same speed: if the intaking fan is rotating faster, the resulting flow will be less efficient (turbulence). Also, intaking fans push a LOT of dust into the cabinet. Proper ventilation is guaranteed by respecting the following points: o do not place disks that you know are running very hot (more than 55°C) next to each other -- this will only make disks close to each other overheat much faster; o leave some space between the units (1/2" to 1") -- this allows better air flow; o leaving 5"1/4 front bays opened up creates a "leak" in the air circuit, which could result in less efficient cooling; o try to achieve good thermal dissipation. This means: - mount the disks directly on metal whenever you can -- the metal frame inside the cabinet usually provides average to good thermal dissipation - avoid disk drawers and other "easy extraction" mecanisms. Since they are made of plastic, they trap heat (if your disk died suddenly, you'll still probably have to restore a backup, and saving 2 minutes in opening the casing won't make that much difference) = = = If you follow these guidelines, you will already avoid many of the problems of building and keeping running a PC based server or workstation -- a process which Jordan rightly describes as 'Tapdancing through a minefield'. Most PC hardware is indeed not among the best in quality, but through careful selection AND correct placement, it is possible to approach the performance and reliability of much more expensive dedicated hardware. -- -[ Philippe Regnauld / regnauld@eu.org / +55.4N +11.3E @ Sol3 / +45 31241690 ]- -[ "To kärve or nøt to kärve, that is the qvestion..." -- My sister ]-