| G | hads: thought you might be interested, finally got my hands on an mSATA drive and have confirmed for Intel that it isn't the OS at least (happens with Windows as well), hopefully they'll do something now | 03:23 |
|---|---|---|
| olly | rather sad they aren't interested in linux issues | 06:27 |
| olly | it seems a bit of hardware which is much better suited to linux than windows | 06:28 |
| G | olly: especially when you consider that one of the usecases they float about is digital signage et al | 07:34 |
| lifeless | G: huh. Did you source it locally? | 07:43 |
| lifeless | G: I'd seriously consider returning it as faulty. | 07:44 |
| G | lifeless: yeah, if I don't hear anything tomorrow morning (i.e. before US goes to the weekend) gonna take it back | 07:47 |
| G | the counterpoint is, that it feels obvious that there is a bad batch of something, kind of want to make sure that Intel acknowledge it for the good of everyone else, (the counterpoint to that is why should I when they appear to be ignoring two offers to help via two different channels) | 07:51 |
| lifeless | heh | 07:54 |
| lifeless | so Intel will feel it when they get returns | 07:54 |
| lifeless | even if they don't ack it, it costs a lot when product is returned faulty | 07:54 |
| lifeless | much more than 'best effort online support' | 07:54 |
| lifeless | I'm not suggesting that you should cost them money | 07:55 |
| lifeless | just that its a signal they will be looking at | 07:55 |
| G | If it was HP I wouldn't blink an eye at the idea, it's just I kind of like Intel | 07:56 |
| lifeless | oof that stings | 07:56 |
| G | plus surely processing via Warranty is more expensive, as they have to ship directly? :) | 07:57 |
| lifeless | [I work at HP these days :)] | 07:57 |
| G | lifeless: oh, you left Canonical? | 07:58 |
| lifeless | I did | 07:58 |
| G | lifeless: sorry then, just I've had a bad experience with HP in the past, and to be honest, when I look at laptops or printers in particular, I run the other direction from the HP stuff | 07:59 |
| lifeless | No need to apologise for genuine experience ;) | 07:59 |
| lifeless | And HP is a biiiiiig company, I have nothing to do with printers or laptops (sadly, I would -kill- for a samsung screen on this work laptop) | 08:00 |
| ibeardslee | HP generally seem to have decent stuff (ime), but just as useless as most providers supplying PCs and Laptops without legacy OSlicences | 08:00 |
| lifeless | OTOH this laptop is beautiful in every other way | 08:00 |
| G | (crazy thing is, as far as servers and that go, I'd say it'd be a toss up between HP or IBM for me) | 08:01 |
| ibeardslee | would love to find the right contact to keepbeating over the head until they make it easy to get legacy OS free laptops | 08:02 |
| lifeless | nothing under http://www.ubuntu.com/certification/make/HP/ grabs you ? | 08:02 |
| ibeardslee | plenty .. but I can't order it online with linux .. it's a special order and I can't do it online | 08:04 |
| lifeless | oh | 08:05 |
| lifeless | thats disappointing | 08:05 |
| ibeardslee | very | 08:05 |
| ibeardslee | it should be just part of the drop down list of OS options? | 08:06 |
| lifeless | any of those models are Ubuntu certified | 08:06 |
| lifeless | which means that HP and Canonical have a contract and its tested etc | 08:06 |
| lifeless | they don't generally do that unless they are going to market with the model with Ubuntu, because its not free | 08:07 |
| ibeardslee | doesn't mean you can buy it | 08:07 |
| lifeless | vendors often make stuff like this region specific | 08:07 |
| lifeless | so it may be that you can in the US, or in China, or Brazil | 08:07 |
| ibeardslee | which is retarded | 08:08 |
| lifeless | I have no info about what HP does in this regard, just general info. | 08:08 |
| lifeless | I *imagine* that there is a minimum volume needed to make logistics for Ubuntu etc work in any region | 08:08 |
| ibeardslee | really like no-one can use an airplane or boat? | 08:09 |
| lifeless | no, like you won't sell in a region without sales staff training, support staff training, retail pipeline training and collateral | 08:10 |
| lifeless | so that you don't get J Random 'mom or pop' ringing up the support centre for this thing they bought they don't know how to use and won't run Word | 08:10 |
| lifeless | *and then* run into a clueless helpdesk person that has no idea whats going on. | 08:11 |
| G | lifeless: don't you mean, J Random why can't find the start button and getting the message "No boot disk found" :) | 08:11 |
| ibeardslee | so B Random calls and says apt-get install postgresql doesn't work? ;) | 08:11 |
| lifeless | folk who replace their OS are not in the regular consumer category, by definition :) | 08:12 |
| ibeardslee | no, but if use the default installed OS that won't work | 08:13 |
| G | to be perfectly honest, I get why the likes of HP/etc don't do "OS-less to distributor" shipments, particularly in this part of the world, it's the likes of Lenovo & Dell that are mainly Build-To-Order, surely it's a simplier process to forget to image the HDD and put a sticker on | 08:13 |
| lifeless | its not the bto aspect that makes it tricky - its all the rest I just mentioned | 08:15 |
| ibeardslee | and anyway, don'tthey send support calls to another region anyway? | 08:15 |
| G | lifeless: yep, but my main point there is also logistical concerns | 08:16 |
| lifeless | G: there are some, particularly non-BTO, but those can be dealt with just by allowing long lead times for those units | 08:16 |
| ibeardslee | But .. in the plus side we recently got a laptop on special order that ended up being <>$500 less than the same thing with the legacy OS | 08:17 |
| lifeless | nice | 08:21 |
| ibeardslee | but that's not usual .. e've regularly paid more for the FreeDOS version | 08:23 |
| ibeardslee | .. or at least had quotes for the FreeDOS version that was more expensive than the one with the legacy OS preinstalled | 08:24 |
| olly | morning | 21:51 |
| G | hads: curiously, which HDMI port do you use on your NUC? One that is next to the LAN or the one next to the USB? | 21:53 |
| lifeless | what res, and how many, can you run on a NUC? | 22:19 |
| G | lifeless: I think it depends on which one you get, the common one is 2xHDMI (so I'd imagine as far as res goes, it's whatever the Intel 4000 HD can push out over HDMI). The other has Thunderbolt, which may allow chaining of monitors | 22:34 |
| lifeless | oooh | 22:35 |
| lifeless | I want to give thunderbolt a serious workout someday | 22:35 |
| lifeless | the other thing I want to do is upgrade to 10Gb ethernet @ home, but I can't find a 5-port switch anywhere :( | 22:35 |
| G | lifeless: you have equipment at home capable of 10Gbit? | 22:36 |
| lifeless | G: not on motherboard, but they have pci 8x | 22:37 |
| lifeless | 2-port 10Gb pci8x cards are quite available | 22:37 |
| G | oh yeah, realised that, but I was just curious as to if you already had 10GbE stuff | 22:38 |
| lifeless | just the switch i'm lacking. I may need to just get 2x 2-port cards and throw them in my server | 22:38 |
| lifeless | ah | 22:38 |
| lifeless | so no, but I want to :) | 22:38 |
| lifeless | ahha DE-805TP may do the trick | 22:39 |
| lifeless | nope | 22:39 |
| lifeless | thats a bad blurb, 10M :( | 22:39 |
| G | yeah, I was gonna say | 22:40 |
| G | "Fastest switch in the west" | 22:40 |
| G | actually, it's not even a switch, it's a hub | 22:40 |
| G | lifeless: Netgear might have something in the Prosafe range | 22:41 |
| lifeless | G: so my thinking is that NAS density is silly high now | 22:44 |
| G | lifeless: yeah, I'd agree with that | 22:45 |
| lifeless | G: and if I iSCSI things up (or even SMB/NFS) I should be able to get better than local disk performance, if I have > 5Gb b/w. Lots of SME NAS kit is now 10Gb ready | 22:45 |
| lifeless | [or roll your own NAS with esata enclosures etc etc] | 22:45 |
| G | not sure I'd use eSATA, I'd like go with some sort of PCIe SATA controller over eSATA | 22:46 |
| G | *likely | 22:47 |
| G | of course, that depends (like always) on the onboard chipset really | 22:47 |
| lifeless | G: so that means a new computer case :) - may as well buy a stock nas and flash ubuntu on it for that. | 22:47 |
| G | hmmm based on Netgear's ProSafe Switch Poster, the only 10GbE stuff they have is the L3 switches | 22:48 |
| lifeless | G: anyhow, thats a relatively minor detail | 22:48 |
| G | (tip: don't make my mistake and get one of those side-load HDD cage, cases | 22:48 |
| G | but yeah, with the whole NAS/SAN+Hypervisor model we are going with, it's silly not to have all your storage in one box (and hope it doesn't fail) and use iSCSI/etc to make it available to the rest of the machines | 22:51 |
| lifeless | s/one box/two boxes/ [thanks ceph :P] | 22:51 |
| G | okay, two boxes then :) | 22:51 |
| lifeless | I'm thinking if I put a decent chunk of effort in I can spend a litle now and never have to give a toss about storage logistics at home for another decade | 22:52 |
| lifeless | just rolling upgrade the disks as needed | 22:52 |
| G | yeah, the likes of ceph/gluster/etc sure do make that handy, but isn't there going to be a point where SATA 6Gb/s isn't going to cut it? | 22:53 |
| G | or are going to be crazy expensive because it's legacy (like DDR1 RAM) | 22:54 |
| lifeless | yeah - so | 22:55 |
| lifeless | this is where iscsi vs NFS/SMB makes a difference | 22:55 |
| lifeless | with iscsi you don't need much/any caching on the storage nodes | 22:55 |
| lifeless | its all cached by the consuming device as os page cache entries | 22:55 |
| lifeless | so consider a hard disk - 64MB cache or so often, right ? | 22:56 |
| G | Thinking back 10 years, we didn't really have SATA (not until 2003), that was only 1.5Gbps, we've only recently started getting an uptick in 6Gb/s SATA, so a lot has changed, I'd say 10 years is an unrealistic goal, I think you'd need to add a refresh in at the 5 year mark | 22:56 |
| G | yeah, the ones I brought the other month were 64MB Cache iirc | 22:56 |
| lifeless | so infiniband is up at 40Gb and beyond | 22:57 |
| lifeless | 2003 was 1.5Gbps SATA, 2008 gave us 6Gbps | 22:58 |
| G | yeah, but we didn't really get a large number of 6Gbps equipment until the last year or so | 22:59 |
| lifeless | sata express can do - wtf - 16Gbps | 22:59 |
| lifeless | http://en.wikipedia.org/wiki/Serial_ATA (3.2 section) | 22:59 |
| G | yeah, I was reading that | 23:00 |
| * G starts reading the presentation | 23:00 | |
| G | hmmm the SATA folk reckon volume shipping was in Jul `10, so thats more 7 years since SATA 1 came out | 23:01 |
| G | (for 6Gbps SATA that is) | 23:01 |
| lifeless | http://www.sata-io.org/technology/sataexpress.asp links to a sata express faq | 23:03 |
| G | lifeless: hmm the reference Wikipedia use tehre (the pdf/presentation) is interesting, convert a PCIe slot on the motherboard to two normal SATA ports, or one super SATA Express port it seems | 23:03 |
| lifeless | yeah | 23:04 |
| lifeless | I read that pdf just now too :) | 23:04 |
| lifeless | the faq is useful, expands on reasoning | 23:04 |
| lifeless | The specification is expected to start SATA-IO member review by the end of 2012. It may be | 23:05 |
| lifeless | ratified as early as Q1 2013 | 23:05 |
| lifeless | SATA | 23:05 |
| lifeless | Express will be available to the general public in the next version of the SATA specification, for which no | 23:05 |
| lifeless | schedule has yet been set. | 23:05 |
| G | so I guess there is your 10 year plan, just amke sure you get a MB with a lot of PCIe slots :) | 23:05 |
| lifeless | so | 23:05 |
| lifeless | :) | 23:05 |
| lifeless | the goal isn't to run the fastest possible hardware at any point | 23:05 |
| lifeless | its to not be stuck with otherwise great stuff that is inflexible on too-short timeframes | 23:06 |
| lifeless | consider this sketch - 4 x 4-port 6G esata port multiplying enclosures, 2x storage hosts w/4GB RAM each, 2-port 10Gb network connectivity | 23:07 |
| lifeless | running in no-redundancy mode (for simple reasoning), an iscsi device mapped across both hosts, with say 4MB stripes | 23:07 |
| lifeless | there are 16 disks, and linear IO will switch disk -> disk every 4MB | 23:08 |
| lifeless | you want that to go from (storage node / esata port / disk) 0/0/0 then 1/0/0 then 0/1/0 then 1/1/0 then 0/0/1 and so on | 23:10 |
| lifeless | e.g. each time you switch stripe, go from node to node, then enclosure to enclosure w/in node, and disk to disk within enclosure, to keep utilisation of buses from being a choke point | 23:10 |
| G | yeah, that'd work quite well I'd imagine | 23:11 |
| G | but surely because you are using so many disks, and non-redundancy (I get the reasoning by the way), wouldn't you just be upping your risk for something to go wrong, through the roof? | 23:12 |
| lifeless | totally | 23:14 |
| lifeless | lets just check the math | 23:14 |
| lifeless | assume my games box is a regular 10Gb single port machine | 23:15 |
| lifeless | so we guarantee 10Gb between the two storage nodes for doing replication / redundancy traffic when we go to add that in | 23:15 |
| lifeless | we have 12Gb total esata b/w within each node | 23:15 |
| lifeless | so in reading-mode we can saturate the 10Gb to the games box from either node and still have 2Gb spare to do maintenance operations, but if the load is well distributed we have 5Gb games box traffic per, and 7Gb for maintenance (or another client machine) | 23:16 |
| lifeless | within each enclosure we have 6Gb total, and 4 disks, which is somewhat oversubscribed for linear IO, but probably plenty of excess for random IO unless I put SSD's in there | 23:17 |
| lifeless | quick google suggests peak transfer rates are stil ~150MBps or 1.2Gbps for spinning platters | 23:20 |
| G | yeah, without doing the math (seriously, math on a weekend? ;)), I'd say it'd work pretty well | 23:20 |
| lifeless | SSD's are in the 250MBps range, or 2Gbps, so 4 SSD's would flatline esata. | 23:20 |
| lifeless | G: math is fun :) | 23:21 |
| lifeless | so, *if* I have 10Gbps ethernet, I can drive more disks in parallel from more machines than if each machine is separate. | 23:21 |
| G | lifeless: of course it is, but not on a weekend ;) | 23:22 |
| lifeless | If I have 1Gbps ethernet, my peak will be less than half that of a single SSD | 23:22 |
| lifeless | I guess I could bond 3x1Gbps on each machine... buts thats getting into way too hard territory. | 23:22 |
| lifeless | My laptops will always be non-virtualised for my working set I suspect | 23:22 |
| G | yeah, well I suspect that I was nowhere near maxing the Read capabilities of my WD Blacks, nor the Write of my new mSATA drive, my limitation was the 1Gbps network | 23:23 |
| lifeless | but if I can get good enough b/w to the backend I can at least shuffle that around without dying of old age. | 23:23 |
| lifeless | I've got a 2 WD RED 3TB drives, waiting on an RMA for the other 2. | 23:23 |
| lifeless | they are -so- quiet its not funny | 23:23 |
| G | ahhh yeah, I've got one of those WD Red 3TBs in my Synology box | 23:24 |
| lifeless | so yeah, I think it will work well, I'm going to piece together a plan over a few months, not rush it | 23:24 |
| lifeless | just balking a little at buying a 24port 10Gb switch :( | 23:25 |
| lifeless | G: of course, http://en.wikipedia.org/wiki/100_gigabit_Ethernet would be even more future proof :) | 23:30 |
| lifeless | G: http://www.attotech.com/products/product.php?scat=32&prod=104&sku=TLNS-1101-D00 | 23:35 |
| lifeless | fugly tho :) | 23:36 |
| lifeless | G: NUC's have thunderbolt right ? | 23:36 |
| lifeless | G: 1 or two ? | 23:36 |
| G | lifeless: yeah, the non-IYE model | 23:36 |
| lifeless | G: how much was the unit ? | 23:37 |
| lifeless | [ballpark is fine] | 23:37 |
| G | mine (the -IYE/2xHDMI model) was ~510 from PB Tech, + ~180 NZD for an 128GB mSATA from Crucial + RAM | 23:38 |
| G | lifeless: http://www.intel.co.nz/content/www/us/en/motherboards/desktop-motherboards/next-unit-computing-introduction.html has all the juice on the NUCs | 23:41 |
| G | lifeless: and if you throw the model numbers into http://ark.intel.com you'll get extra info | 23:42 |
| lifeless | G: yeah. Done that :) I don't see any external openinings for the minipci-E ? | 23:42 |
| G | nope, it's on the board, the idea is that the mini-size = WiFi/BT, Full-size = mSATA | 23:43 |
| lifeless | ah | 23:43 |
| lifeless | stab | 23:43 |
| G | also to note that the full size, is directly on top of the half-size/wifi one | 23:44 |
| G | lifeless: also looks like only 1x Thunderbolt on the TB one | 23:44 |
| lifeless | yeah | 23:45 |
| lifeless | so NUC not suitable | 23:45 |
| lifeless | thats ok | 23:45 |
| lifeless | the other thing to note is that by making the compute aspect of a storage node low-end, you can replace that without mass migration of everything | 23:46 |
| lifeless | wheee http://www.attotech.com/howtobuy/index.php?cat=&scat=32&prod=103 | 23:49 |
| lifeless | $price | 23:49 |
| G | oooh, I like how it takes a while to load the price, "Lets keep them in suspense for a moment" | 23:50 |
| lifeless | G: have you seen the shouting in the datacentre video ? | 23:50 |
| G | lifeless: nope | 23:50 |
| lifeless | apropos of storage geeking-out | 23:50 |
| lifeless | G: http://www.youtube.com/watch?v=tDacjrSCeq4 | 23:51 |
| lifeless | warning, loud. | 23:51 |
| G | well there goes my idea of putting a siren in my PC case ;) | 23:54 |
| lifeless | lol | 23:54 |
| * lifeless presumes you were kidding | 23:54 | |
| G | of course | 23:54 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!