Diskless server? Thats inconceivable!



  • So, am specing out some new hardware for the office here, it being 2013 and all, this means something suitable for a VM platform. Which means shared storage. (for the record, a EMC VNXe 3150; for not significantly more cost than a server with that many drive slots, you get the general awesomeness of a tuned appliance.). Which means diskless servers.

    But not from Dell.

    Time	 Details
    08/09/2013 02:06:26PM	Session Started with Agent (Marie T)
    08/09/2013 02:06:26PM	Jeff Warnica: "."
    08/09/2013 02:06:33PM	Agent (Marie T): "Welcome to Dell US Small Office Chat. My name is Marie and I will be your Dell.com Sales Chat Expert."
    08/09/2013 02:06:33PM	Agent (Marie T): "I can be reached at talkingkeyboard17@dell.com. You can now chat with us 24 hours a day, 7 days a week! :)"
    08/09/2013 02:06:35PM	Agent (Marie T): "Hi Jeff"
    08/09/2013 02:06:37PM	Agent (Marie T): "How can I help you today?"
    08/09/2013 02:06:41PM	Jeff Warnica: "Hello."
    08/09/2013 02:07:10PM	Jeff Warnica: "I'm trying to spec out an R420, and the configuration tool is rejecting my requirement of not having any drives."
    08/09/2013 02:07:37PM	Agent (Marie T): "Thank you for chatting with us, Sorry about that, let me see what I can do to help"
    08/09/2013 02:07:51PM	Agent (Marie T): "so you would like to get a server with no hard drive right?"
    08/09/2013 02:08:06PM	Jeff Warnica: "Correct."
    08/09/2013 02:08:25PM	Agent (Marie T): "Please give me 2 to 3 minutes to check on that for you"
    08/09/2013 02:09:12PM	Jeff Warnica: "Sure."
    08/09/2013 02:09:53PM	Agent (Marie T): "Thank you for patiently waiting"
    08/09/2013 02:10:57PM	Agent (Marie T): "checked on the server options here and I sorry to say but we cannot ship out a system with no hard drive, this is why you keep getting an error. the minimum required is at least 1 drive no raid. That is the least drive option we can get"
    08/09/2013 02:11:10PM	Agent (Marie T): "*I am"
    08/09/2013 02:11:46PM	Jeff Warnica: "Who can affect an override on that configuration?"
    08/09/2013 02:12:22PM	Agent (Marie T): "I actually checked that with our enterprise specialist however this is not an option even using their configuration tool"
    08/09/2013 02:13:36PM	Jeff Warnica: "So, despite also selling SAN systems - a major reason to run such products is to have diskless servers - you are unable to sell servers which don't come without disks?"
    08/09/2013 02:15:31PM	Agent (Marie T): "if you need a SAN solution, we have specialist that can discuss that option for you and the best systems for that, however to answer your question on regular rack servers it is not an option"
    08/09/2013 02:16:47PM	Jeff Warnica: "I'm not asking to configure a SAN, I'm asking to configure a rack server which will happen to be used where the is a SAN, e.g. a diskless server."
    08/09/2013 02:17:30PM	Agent (Marie T): "no problem, checking if we have another option here"
    08/09/2013 02:17:38PM	Agent (Marie T): "just a moment"
    08/09/2013 02:20:00PM	Agent (Marie T): "I'm sorry for the wait. I am still working on it"
    08/09/2013 02:20:26PM	Jeff Warnica: "By all means."
    08/09/2013 02:21:46PM	Agent (Marie T): "Thanks Jeff"
    08/09/2013 02:24:31PM	Agent (Marie T): "Thank you for patiently waiting"
    08/09/2013 02:26:01PM	Jeff Warnica: "no problem"
    08/09/2013 02:26:33PM	Agent (Marie T): "had our account manager try to quote the server using her own configurator and it really requires at least 1 drive as well"
    08/09/2013 02:28:26PM	Jeff Warnica: "Ok. I'm about to head home for the day. If I talk to an account rep on Monday, do they have access to different tools?"
    08/09/2013 02:29:34PM	Agent (Marie T): "enterprise specialist like our account managers have the same access/tools so just to set your expectations they would probably give the same info but it's always okay to try"
    08/09/2013 02:30:08PM	Jeff Warnica: "OK."
    08/09/2013 02:31:04PM	Agent (Marie T): "thanks Jeff, can I help you with anything else though?"
    

  • Winner of the 2016 Presidential Election

    I see the WTF. The date is labelled time and the time is labelled details.



  •  Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.  (OK, I'll grant you that on some Datawarehouse Servers ... this was highish volume OLTP).  In a meeting where the fact a server R5, was running out of space, he smugly came out with "I would have thought Scumspawn would have been right in there, but as he's not paying attention I'll point out we just need to move to RAID 10".  Err ... why.  "Well, because then we'll have 10 discs, not 5".

    So all you really needed to do was ask them for a RAID 0 server

    (I did actually send him Art Kagel's NO RAID 5, and the idiot's guide to RAID configuration, he was far too good to have to read stuff like that though)

     



  • I've tried boot from SAN before and I'd rather have a local disk for the OS. It keeps the SAN from having tons of tiny LUNs and makes disaster recovery 100 times easier. HP sells flash media for their servers, along with a custom build of VMware. I wonder if Dell has something similar?



  • @RangerNS said:

    So, am specing out some new hardware for the office here, it being 2013 and all, this means something suitable for a VM platform. Which means shared storage. (for the record, a EMC VNXe 3150; for not significantly more cost than a server with that many drive slots, you get the general awesomeness of a tuned appliance.). Which means diskless servers.

    But not from Dell.

    Time	 Details
    08/09/2013 02:06:26PM	Session Started with Agent (Marie T)
    08/09/2013 02:06:26PM	Jeff Warnica: "."
    08/09/2013 02:06:33PM	Agent (Marie T): "Welcome to Dell US Small Office Chat. My name is Marie and I will be your Dell.com Sales Chat Expert."
    08/09/2013 02:06:33PM	Agent (Marie T): "I can be reached at talkingkeyboard17@dell.com. You can now chat with us 24 hours a day, 7 days a week! :)"
    08/09/2013 02:06:35PM	Agent (Marie T): "Hi Jeff"
    08/09/2013 02:06:37PM	Agent (Marie T): "How can I help you today?"
    08/09/2013 02:06:41PM	Jeff Warnica: "Hello."
    08/09/2013 02:07:10PM	Jeff Warnica: "I'm trying to spec out an R420, and the configuration tool is rejecting my requirement of not having any drives."
    08/09/2013 02:07:37PM	Agent (Marie T): "Thank you for chatting with us, Sorry about that, let me see what I can do to help"
    08/09/2013 02:07:51PM	Agent (Marie T): "so you would like to get a server with no hard drive right?"
    08/09/2013 02:08:06PM	Jeff Warnica: "Correct."
    08/09/2013 02:08:25PM	Agent (Marie T): "Please give me 2 to 3 minutes to check on that for you"
    08/09/2013 02:09:12PM	Jeff Warnica: "Sure."
    08/09/2013 02:09:53PM	Agent (Marie T): "Thank you for patiently waiting"
    08/09/2013 02:10:57PM	Agent (Marie T): "checked on the server options here and I sorry to say but we cannot ship out a system with no hard drive, this is why you keep getting an error. the minimum required is at least 1 drive no raid. That is the least drive option we can get"
    08/09/2013 02:11:10PM	Agent (Marie T): "*I am"
    08/09/2013 02:11:46PM	Jeff Warnica: "Who can affect an override on that configuration?"
    08/09/2013 02:12:22PM	Agent (Marie T): "I actually checked that with our enterprise specialist however this is not an option even using their configuration tool"
    08/09/2013 02:13:36PM	Jeff Warnica: "So, despite also selling SAN systems - a major reason to run such products is to have diskless servers - you are unable to sell servers which don't come without disks?"
    08/09/2013 02:15:31PM	Agent (Marie T): "if you need a SAN solution, we have specialist that can discuss that option for you and the best systems for that, however to answer your question on regular rack servers it is not an option"
    08/09/2013 02:16:47PM	Jeff Warnica: "I'm not asking to configure a SAN, I'm asking to configure a rack server which will happen to be used where the is a SAN, e.g. a diskless server."
    08/09/2013 02:17:30PM	Agent (Marie T): "no problem, checking if we have another option here"
    08/09/2013 02:17:38PM	Agent (Marie T): "just a moment"
    08/09/2013 02:20:00PM	Agent (Marie T): "I'm sorry for the wait. I am still working on it"
    08/09/2013 02:20:26PM	Jeff Warnica: "By all means."
    08/09/2013 02:21:46PM	Agent (Marie T): "Thanks Jeff"
    08/09/2013 02:24:31PM	Agent (Marie T): "Thank you for patiently waiting"
    08/09/2013 02:26:01PM	Jeff Warnica: "no problem"
    08/09/2013 02:26:33PM	Agent (Marie T): "had our account manager try to quote the server using her own configurator and it really requires at least 1 drive as well"
    08/09/2013 02:28:26PM	Jeff Warnica: "Ok. I'm about to head home for the day. If I talk to an account rep on Monday, do they have access to different tools?"
    08/09/2013 02:29:34PM	Agent (Marie T): "enterprise specialist like our account managers have the same access/tools so just to set your expectations they would probably give the same info but it's always okay to try"
    08/09/2013 02:30:08PM	Jeff Warnica: "OK."
    08/09/2013 02:31:04PM	Agent (Marie T): "thanks Jeff, can I help you with anything else though?"
    

    You stayed on the line for 30 minutes trying to shave a $100 disk off the purchase of a $1000 server. Maybe you think saving $100 in one hour is great if you earn less than that, but it's not. The $100 expense will go in the Capex column; it will be spread over 12 fiscal quarters and with depreciation and various other accounting factors it will end up eating maybe $8 of the annual budget. While your time comes out straight from this year's budget.

    That's like going to McDonalds and arguing with the manager to get a discount when you buy a medium Coke on $1-for-any-size-drink days.


  • Discourse touched me in a no-no place

    @Scumspawn said:

     Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.

    Grrr. I know you had more to the story than that but I had to stop and let my blood pressure drop a bit here.

    I worked at a place once, years ago, where they were concerned for the database performance. They got the DB vendor (Progress) to come out and try to help them tune it. The consultant got there, took one look at the RAID 5 array, and said "call us back when you're off RAID 5."


  • Winner of the 2016 Presidential Election

    @Ronald said:

    You stayed on the line for 30 minutes trying to shave a $100 disk off the purchase of a $1000 server. Maybe you think saving $100 in one hour is great if you earn less than that, but it's not. The $100 expense will go in the Capex column; it will be spread over 12 fiscal quarters and with depreciation and various other accounting factors it will end up eating maybe $8 of the annual budget. While your time comes out straight from this year's budget.

    That's like going to McDonalds and arguing with the manager to get a discount when you buy a medium Coke on $1-for-any-size-drink days.

    I don't really see how fiddling around with some accounting numbers in a spreadsheet somewhere makes a smaller number more than a bigger number. I mean at some point there are physical dollars, and no amount of budgetary shuffling makes you spend less on a higher dollar figure.


  • @FrostCat said:

    @Scumspawn said:

     Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.

    Grrr. I know you had more to the story than that but I had to stop and let my blood pressure drop a bit here.

    I worked at a place once, years ago, where they were concerned for the database performance. They got the DB vendor (Progress) to come out and try to help them tune it. The consultant got there, took one look at the RAID 5 array, and said "call us back when you're off RAID 5."

    Had this myself two weeks ago, vendor insisted that their software needed a SSD for the database.  Inserting new items in the products table was going at about 3 rows per second.  Same vendor asked why the company was running a domain, and was "wasting money" on CALs.  This is at a outdoor powersports (golfcarts and ATVs) dealer with their own electric & gas service shop, upholstery shop, paint & vinyl, 40 or so employees, 20 computers, each requiring varying levels of access to company resources, group policies, shared computers, roaming profiles, etc.  Back to the reason he was there, they spent $15k for a industry specific end-to-end business management program.  Think Quickbooks Enterprise, but with automatic import of products and prices from vendors, VIN tracking (even across customers) and much lower initial license cost. The software has been in continuous development since 1985.  Yet it requires an SSD, i7 2.9GHz, 16GB+ of ram, and 10GE to function correctly.

     



  •  @Jaime said:

    I've tried boot from SAN before and I'd rather have a local disk for the OS. It keeps the SAN from having tons of tiny LUNs and makes disaster recovery 100 times easier. HP sells flash media for their servers, along with a custom build of VMware. I wonder if Dell has something similar?

    +1 

     

    I was also going to say something about being able to remove the shipped local drive for reuse elsewhere, but then realized that it'd void the warranty. And with an engineer's salary, unattaching the drive would cost the company more than the drive itself did.



  • It happens that Dell hardware (at least, this model) has allowances for SD storage locally. This requires the highest level of out-of-band management option, iDRAC enterprise. The build I wasn't able to put in the shopping cart did have the iDRAC Enterprise option, with one of their Dell branded SDHC cards. (not that the rep exerted any effort to indicate that they knew what I was trying to build; I'm confident that they have no access to the design I was building with their tool). "I don't think that that is a good idea" is a reasonable, if absurd, response. "our stupid tool doesn't allow this entirely reasonable option" is the WTF?.

    Except for infrequent configuration changes, and maybe logging, I can't see the VM host itself being responsible for any serious disk IO. I'd be entirely willing to accept some of the "compute host" memory to cache such that it never hits disk, e.g. the hardware has 64GB of memory, yet a 4 GB OS root, and 0 swappyness. The OS never has to hit "disk" (for some very expansive definition of "disk") past boot up.

    Yeah, HP has allowed SCHD cards for at least 1 "generation" more worth of servers (or, at least, in-the-case, can't be knocked out, USB ports on the MOBO). If I thought I had a 0.0001% chance of purchasing non-Dell servers, I'd just go ahead and point dell.ca to 127.0.0.1. But, well, mumble*golf*financing*relationships*



  • @Scumspawn said:

     Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.  (OK, I'll grant you that on some Datawarehouse Servers ... this was highish volume OLTP).  In a meeting where the fact a server R5, was running out of space, he smugly came out with "I would have thought Scumspawn would have been right in there, but as he's not paying attention I'll point out we just need to move to RAID 10".  Err ... why.  "Well, because then we'll have 10 discs, not 5".

    So all you really needed to do was ask them for a RAID 0 server

    (I did actually send him Art Kagel's NO RAID 5, and the idiot's guide to RAID configuration, he was far too good to have to read stuff like that though)

     

    I had to sell my boss on using any form of RAID at all once. His idea was to just mount all 24 drives to separate directories and have our software decide where to write files (videos saved as a folder with single image files). Took 3 weeks of convincing. He vaguely remembered reading something about disk head movements being synchronized and was worried about real-time performance. After a lot of discussion we ended up with RAID 6 (The hardware budget for the project wouldn't pay for enough disks to get the promised capacity at RAID 10 because the contract was signed without first checking this).



  • @joe.edwards said:

    @Ronald said:

    You stayed on the line for 30 minutes trying to shave a $100 disk off the purchase of a $1000 server. Maybe you think saving $100 in one hour is great if you earn less than that, but it's not. The $100 expense will go in the Capex column; it will be spread over 12 fiscal quarters and with depreciation and various other accounting factors it will end up eating maybe $8 of the annual budget. While your time comes out straight from this year's budget.

    That's like going to McDonalds and arguing with the manager to get a discount when you buy a medium Coke on $1-for-any-size-drink days.

    I don't really see how fiddling around with some accounting numbers in a spreadsheet somewhere makes a smaller number more than a bigger number. I mean at some point there are physical dollars, and no amount of budgetary shuffling makes you spend less on a higher dollar figure.

    Physical dollars? Do you think that when you go to the bank and make a deposit they physically put bills in a small box with your name on it? This is hilarious.

    People like you are what make companies like Wells Fargo or American Express so successful.

    Since money is a big mystery to you, here is a puzzling riddle that will keep you up at night if you don't google for the answer:

    @Missing dollar riddle said:

    Three guests check into a hotel room. The clerk says the bill is $30, so each guest pays $10. Later the clerk realizes the bill should only be $25. To rectify this, he gives the bellhop $5 to return to the guests. On the way to the room, the bellhop realizes that he cannot divide the money equally. As the guests didn't know the total of the revised bill, the bellhop decides to just give each guest $1 and keep $2 for himself. Each guest got $1 back: so now each guest only paid $9; bringing the total paid to $27. The bellhop has $2. And $27 + $2 = $29 so, if the guests originally handed over $30, what happened to the remaining $1?



  • @Ronald said:

    And $27 + $2 = $29
    I understand the whole thing's supposed to be a kind of joke, but I've never really got it. Why is it funny to randomly add an extra two dollars to a number and then say it's not the right number? Of course it's not the right number, it's two too big because you just added two to it.


    Is there some reason I'm missing as to why you'd add the bellhop's share a second time?



  • @FrostCat said:

    @Scumspawn said:

     Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.

    Grrr. I know you had more to the story than that but I had to stop and let my blood pressure drop a bit here.

    I worked at a place once, years ago, where they were concerned for the database performance. They got the DB vendor (Progress) to come out and try to help them tune it. The consultant got there, took one look at the RAID 5 array, and said "call us back when you're off RAID 5."

    RAID-5 is a piece of shit, especially with big disks. Not only is it slow as hell on write (because it has to compute the parity) but for the same reason it is also very very slow on rebuild. So if you have an array of 1TB disks and one of them fails, while the rebuild is running (which takes forever) not only is the array under significantly higher load (with one less spindle to fan out the read I/O) but it is also straining to recalculate the parity for an entire disk while trying to keep up with the extra workload. Which often leads to the next weakest spindle to go down in flames and it's over - you need new disks AND you need to call Iron Mountain to checkout a shitload of tape backups (because of course if you are cheap to the point of using RAID-5 you are not paying for an online backup solution like a Datadomain or Protectier).

    Meanwhile RAID-10 is a lot faster on both read and write, and will rebuild drives in a jiffy since the controller only has to clone the disk, no parity calculation is needed. RAID-10 is way faster on all kinds of workload but especially on big sequential writes (typical for databases because of the transactions logs). With the price of hard disks, even years ago, it is not a cost-effective solution to strangle the storage infrastructure to save a few dollars. It's better to streamline disk usage or build tiers so the database and other I/O intensive workloads get the VIP treatment while the static, useless stuff (like archived data or software library ISOs) are downgraded to commodity storage.

    A lousy RAID controllers with a cheapskate configuration like RAID-5 is often the root cause of mysterious performance problems. The consultant was right; it's like calling a cable guy to fix a bad signal problem and when he walks in he sees that you have hooked three cheap splitters that you got for $0.25 on ebay (shipping included), it is just not worth looking further.



  • @TDWTF123 said:

    @Ronald said:
    And $27 + $2 = $29
    I understand the whole thing's supposed to be a kind of joke, but I've never really got it. Why is it funny to randomly add an extra two dollars to a number and then say it's not the right number? Of course it's not the right number, it's two too big because you just added two to it.


    Is there some reason I'm missing as to why you'd add the bellhop's share a second time?

    Well that story is better told when it happens in a restaurant, but I was lazy and copy-pasted the version from Wikipedia. Try this one:

    Three friends go to a restaurant. At the end of the meal the bill is $30 so they pay $10 each. Then a few minutes later the waiter comes back and 
    says there has been a mistake, the bill should be $25 so he has to give them back $5. This amount can't be divided in 3 easily so one of the 
    guest says: well you've been kind to let us know, so give us back a dollar each and keep the remaining $2. 
    
    So the guests paid $27 (3 x $9) for food and $2 as a tip. This means $29 - where is the missing $1?
    

    There is no "missing" money, it's just like a used car salesman putting numbers together so it almost look like he is paying you to get the car (like: "this car is worth $10,000 but I'm selling it to you for $5,000 plus I give you this $1,000 Best Buy gift card - so do the math: with a discount of $5,000 on a car that you pay only $5,000 and with the added $1,000 gift card, you are making money AND basically getting a beautiful Ford Escort for free, but hurry up another couple was here two hours ago and they said they would come back this afternoon to buy it!").



  • @Ronald said:

    So the guests paid $27 (3 x $9) for food and $2 as a tip.
    Well, at least that makes the nonsense obvious. But why would anyone say the guests paid $27 for food, rather than food and tip, in the first place? And why would you then add the tip a second time?


    It's no more a riddle than asking 'why is 2+2 not equal to seventeen'. Who would ever think it was in the first place?



  • @TDWTF123 said:

    It's no more a riddle than asking 'why is 2+2 not equal to seventeen'. Who would ever think it was in the first place?

    People who think that the bank is putting physical dollars in small boxes with their name on it.



  • @Ronald said:

    You stayed on the line for 30 minutes trying to shave a $100 disk off the purchase of a $1000 server. Maybe you think saving $100 in one hour is great if you earn less than that, but it's not. The $100 expense will go in the Capex column; it will be spread over 12 fiscal quarters and with depreciation and various other accounting factors it will end up eating maybe $8 of the annual budget. While your time comes out straight from this year's budget.

    That's like going to McDonalds and arguing with the manager to get a discount when you buy a medium Coke on $1-for-any-size-drink days.

    It may be you that is missing something financially. Trading cap for expense will look good on paper the first year, but it will continue to hit your budget in years two and three. That's why the bean counters don't let people just mark anything as "Cap". It's not a magic accounting wand. At the end of the day, $100 is $100 whether it's amortized or not.


  • @Jaime said:

    @Ronald said:

    You stayed on the line for 30 minutes trying to shave a $100 disk off the purchase of a $1000 server. Maybe you think saving $100 in one hour is great if you earn less than that, but it's not. The $100 expense will go in the Capex column; it will be spread over 12 fiscal quarters and with depreciation and various other accounting factors it will end up eating maybe $8 of the annual budget. While your time comes out straight from this year's budget.

    That's like going to McDonalds and arguing with the manager to get a discount when you buy a medium Coke on $1-for-any-size-drink days.

    It may be you that is missing something financially. Trading cap for expense will look good on paper the first year, but it will continue to hit your budget in years two and three. That's why the bean counters don't let people just mark anything as "Cap". It's not a magic accounting wand. At the end of the day, $100 is $100 whether it's amortized or not.

    No it's not. And when you say "at the end of the day", what day is this? Because you see, time and money have a complex relationship that appears to elude you. $100 spent today and $100 spent over three years is not the same thing at all. And $100 spent on labor is also different from $100 spent on assets.

    Do you have actual experience handling real life budgets (and the social committee budget for birthday cakes does not count)? Because the fact that you think that "bean counters" decide what people can "mark as Cap" or not is a clear indicator that you don't. Actually, just by saying "bean counters" you show that you have no experience in these matters; it's like a news reporter, until they do a fluff piece about "hackers" and get schooled by actual IT people you can tell they just use words they don't really understand.



  • In the few dealings I have had with Dell, I am amazed that they stay in business.  They seem to do everything they can to NOT sell their products.



  • @Ronald said:

    RAID-5 is a piece of shit, especially with big disks.

    The past couple of weeks, I've been watching what must be some big joke as my co-workers try to restore replication for our nearly 2 TB database. We've actually been running off the secondary server this whole time because the main server got corrupted somehow when they tried to expand the capacity of the RAID 10 and found out our servers didn't allow that. So they tried to reconfigure it as a RAID 6, for extra capacity, and lo-and-behold, replication just couldn't keep up. I told them RAID 6 has a 6x write penalty whereas RAID 10 only has a 2x penalty. They didn't think it was a big deal since we're using SSDs. 😞 So they put the server back to a RAID 10 and finally got another successful backup of the database today (another joke) and should have replication running again sometime tonight. I really really hope it's able to keep up. Did I mention the secondary server runs on a RAID 0? Yeah, sleep has been a little difficult while all this has been going on.


  • Winner of the 2016 Presidential Election

    I don't claim to be an accountant, and no I don't think the bank keeps separate drawers of money for each customer. But I do believe the numbers on a spreadsheet are an abstract representation of an actual tangible resource, and no amount of mathemagical manipulation can change the fact that you're dealing with a quantifiable resource. If I have five apples and Susy has eight apples, she has more apples than me even if her budget won't let her eat more than two this year.



  • @Ronald said:

    @TDWTF123 said:
    It's no more a riddle than asking 'why is 2+2 not equal to seventeen'. Who would ever think it was in the first place?

    People who think that the bank is putting physical dollars in small boxes with their name on it.

    Even they tend not to fail at basic arithmetic.


    I think I'm complaining that it's called a riddle. It's not. It's just a trick question, where if you don't listen carefully someone can slip in a wrong number. It's, in short form: 'you buy something for $27, with a $2 tip, and receive no change from $30. Why?' And the answer is 'because I lied about the $27'.


    It reminds me of the old joke: What's green, hangs on a wall, and whistles? "A Herring" — because you can paint it green, nail it to the wall — and the whistling part is added just to make the riddle hard.



  • @joe.edwards said:

    If I have five apples and Susy has eight apples, she has more apples than me even if her budget won't let her eat more than two this year.
    If you have five apples, and the government says they'll give you an extra one for each you eat this year, but not for those you turn into juice, how many apples do you have? The answer now depends on how many you eat and how many you juice.


    It's not mathematics that changes the amount of money, but tax laws and accountancy. Your sniff-sense is right to detect that there's something odd going on, but it's not the accountancy so much as the ridiculously complex corporate tax system(s) Europeans and Americans unfortunately have, and which results in stuff like this actually working.


  • Discourse touched me in a no-no place

    @GrizzlyAdams said:

    @FrostCat said:

    @Scumspawn said:

     Well, at a previous place of employment, a spectacularly arrogant and useless colleague used to argue like hell there was no point in specifying RAID10 for database servers rather than RAID 5.

    Grrr. I know you had more to the story than that but I had to stop and let my blood pressure drop a bit here.

    I worked at a place once, years ago, where they were concerned for the database performance. They got the DB vendor (Progress) to come out and try to help them tune it. The consultant got there, took one look at the RAID 5 array, and said "call us back when you're off RAID 5."

    Had this myself two weeks ago, vendor insisted that their software needed a SSD for the database.  Inserting new items in the products table was going at about 3 rows per second.  Same vendor asked why the company was running a domain, and was "wasting money" on CALs.  This is at a outdoor powersports (golfcarts and ATVs) dealer with their own electric & gas service shop, upholstery shop, paint & vinyl, 40 or so employees, 20 computers, each requiring varying levels of access to company resources, group policies, shared computers, roaming profiles, etc.  Back to the reason he was there, they spent $15k for a industry specific end-to-end business management program.  Think Quickbooks Enterprise, but with automatic import of products and prices from vendors, VIN tracking (even across customers) and much lower initial license cost. The software has been in continuous development since 1985.  Yet it requires an SSD, i7 2.9GHz, 16GB+ of ram, and 10GE to function correctly.

     

    Nice WTF, but not what I meant. The database vendor's point was "nothing else you can do to tune the system until you use almost anything but RAID 5." Another data point, and I've probably told this story here before: at another client, years later, I moved the database from a RAID 5 SAN to a RAID 0 or 1 (I forget which; it was probably mirrored) pair of disks built into the server, without the users ever noticing. That means that those two disks were at least as fast as the 30-spindle network array.



  • So why does raid 5 even exist?

    Do I know anything about raid?

    No, I do not. But clearly that didn't stop me from asking this question.



  • @dhromed said:

    So why does raid 5 even exist?

    Do I know anything about raid?

    No, I do not. But clearly that didn't stop me from asking this question.

    No worries. Most people who *decide* what RAID should be used in a storage unit don't know shit about RAID.

    Here is a simplified version. Suppose you have 3 disks of 1TB each. If you connect them to a RAID controller and choose the RAID-5 configuration, the 3 disks will now behave like a big virtual disk, and under the hood this virtual disk will be divided in stripes (1 block on each disk = 1 stripe). If you need to write a block and it turns out that it is physically located on the first disk, then the controller will check what is in the equivalent block on the 2nd disk and possibly update the 3rd disk to make it a parity block. That way, if any of the disks fail, you can guess what was was on the block for each stripe by looking at the two other disks.

    The performance is terrible because for every single write the parity must be calculated then written, and for each read 1/3 of the disks in the stripe must be skipped over. The parity blocks are not all on the same disk (by design) so the controller always needs to figure out where it stands.

    Then suppose one of the disks fails. Good news: you can plug in a new disk and rebuild it by looking at the two other disks. However this takes forever on a 1TB disk because it has to be done stripe by stripe (a stripe can be as small as 32K)... meanwhile the controller is still up and running so the rebuild work is done in addition to the regular workload. This puts a lot of stress on the two remaining disks and that is often a time where they decide to call it quit. And as soon as you lose a second drive the data is gone (all of it).

    Now compare with a RAID-10. There are actually two kinds of RAID-10 (1+0 or 0+1) but basically it's the same idea: either you have a stripe that sits on two pairs of mirrored disks, or you have two pairs of disks with a mirror of the same stripe. Since you work with pairs of disks, there is no parity calculation and no need to skip over disks; actually during read operations controllers can even poll disks in round-robin to spread the load, which is a major performance boost (the bottleneck in storage is always the spindle). If you lose a disk, you plug another one and it is simple cloned, which is very fast compared to reading, calculating and writing a parity block.

    The reason why so many people like RAID-5 is because the redundancy comes at a lower expense. With RAID-5 you lose 1/3 of the available storage because of the parity block; with RAID-10 you lose half since every disk has a mirror. On paper this looks like a good idea, but in fact when you need the RAID-5 to perform its fault-tolerance duties, it usually shits itself and/or offers such a degraded performance that it's basically not worth it. And during day-to-day operations the performance is horrible. So to save the difference between 1/3 and 1/2 of the capacity you cripple the infrastructure.

    Bottom line: fuck RAID-5.



  • @twilsonxpert said:

    Did I mention the secondary server runs on a RAID 0?

    RAID-0 is funny because not only is it not redundant, it also makes the array less reliable than individual disks since a single failed disk will cause the loss of all the data in the array. It's like a chain-gang walking on thin ice*.

    * suggestion: have a look at the picture of the girl who posted that clip.



  • I propose RAID--1, which stores two copies of the data on the same drive.



  • @Jaime said:

    I've tried boot from SAN before and I'd rather have a local disk for the OS. It keeps the SAN from having tons of tiny LUNs and makes disaster recovery 100 times easier. HP sells flash media for their servers, along with a custom build of VMware. I wonder if Dell has something similar?

    That's not how you are supposed to do scalable diskless (at least for web farms, and other setups where you have bunches of identical machines). You create a master image, and you TFTP-boot your servers. That way, when you want to upgrade any of them, you upgrade the master image, and just reboot everything. The only LUNs you create are for the data partitions (if any), which you do want on the SAN.

    Even if you have 1000 custom servers (each of which has a different OS and software configuration), it's still better to create 1000 images and TFTP-boot them, than create 1000 LUNs for the OS partitions.


  • Winner of the 2016 Presidential Election

    @TDWTF123 said:

    @joe.edwards said:
    If I have five apples and Susy has eight apples, she has more apples than me even if her budget won't let her eat more than two this year.
    If you have five apples, and the government says they'll give you an extra one for each you eat this year, but not for those you turn into juice, how many apples do you have? The answer now depends on how many you eat and how many you juice.


    It's not mathematics that changes the amount of money, but tax laws and accountancy. Your sniff-sense is right to detect that there's something odd going on, but it's not the accountancy so much as the ridiculously complex corporate tax system(s) Europeans and Americans unfortunately have, and which results in stuff like this actually working.

    I wasn't expecting an answer that made sense, but you've provided one. Thank you.



  • @suid said:

    @Jaime said:
    I've tried boot from SAN before and I'd rather have a local disk for the OS. It keeps the SAN from having tons of tiny LUNs and makes disaster recovery 100 times easier. HP sells flash media for their servers, along with a custom build of VMware. I wonder if Dell has something similar?

    That's not how you are supposed to do scalable diskless (at least for web farms, and other setups where you have bunches of identical machines). You create a master image, and you TFTP-boot your servers. That way, when you want to upgrade any of them, you upgrade the master image, and just reboot everything. The only LUNs you create are for the data partitions (if any), which you do want on the SAN.

    Even if you have 1000 custom servers (each of which has a different OS and software configuration), it's still better to create 1000 images and TFTP-boot them, than create 1000 LUNs for the OS partitions.

    Remind me again how you TFTP boot Windows...

    As for VMware, I did this in the VMware 3 and 4 days and we didn't run ESXi, so network boot wasn't an option.



  • @Ben L. said:

    I propose RAID--1, which stores two copies of the data on the same drive.

    Deploy this on a single SSD and you basically contradict every single letter in the RAID acronym.

    Technically if you have a large number of disks and you can shard your data it's a highly reliable architecture from a system perspective. That's how terrorist cells work.



  • @Ronald said:

    Technically if you have a large number of disks and you can shard your data it's a highly reliable architecture from a system perspective. That's how terrorist cells work.

    That would be RAID-1. RAID--1 requires that you only have one disk accessible to any machine.



  • @Ben L. said:

    @Ronald said:
    Technically if you have a large number of disks and you can shard your data it's a highly reliable architecture from a system perspective. That's how terrorist cells work.

    That would be RAID-1. RAID--1 requires that you only have one disk accessible to any machine.

    Are you not following? RAID-1 is a mirror of two disks. Terrorist cells don't have a mirror cell. At least not the ones on tv.



  • @Ronald said:

    you basically contradict every single letter in the RAID acronym.
    The kid's got talent.



  • @Ronald said:

    No it's not. And when you say "at the end of the day", what day is this? Because you see, time and money have a complex relationship that appears to elude you. $100 spent today and $100 spent over three years is not the same thing at all. And $100 spent on labor is also different from $100 spent on assets.

    Do you have actual experience handling real life budgets (and the social committee budget for birthday cakes does not count)? Because the fact that you think that "bean counters" decide what people can "mark as Cap" or not is a clear indicator that you don't. Actually, just by saying "bean counters" you show that you have no experience in these matters; it's like a news reporter, until they do a fluff piece about "hackers" and get schooled by actual IT people you can tell they just use words they don't really understand.

    Yes, $100 spent on labor is different from $100 spent on assets. However, your previous explanation gave the impression that the first year cost is the only thing that matters, artificially making labor look much more expensive than capital. Of course interest rates and takes make the comparisons more complicated than can be explained in a simple paragraph, but the fact remains that capital will hit the budget for many years and expense will hit entirely in the budgetary period in which it occurs.

    Which actually costs more will depend on a lot of things, but the difference will not be any where near as significant as you made it out to be ($100 vs. $8).

    BTW, your personal attacks on me are just a signal that your position is so weak that you need to attack the attacker instead of the message.

    @Ronald said:

    ...time and money have a complex relationship that appears to elude you. $100 spent today and $100 spent over three years is not the same thing at all.

    You seem to have been living in a box for quite a long time. Interest rates have been close to zero for a very long time now. It's almost to the point where they are the same thing.


  • @RangerNS said:

    08/09/2013 02:11:46PM	Jeff Warnica: "Who can affect an override on that configuration?"


    Ah, there's the WTF!  :-p



  • Discourse touched me in a no-no place

    @Ronald said:

    Deploy this on a single SSD and you basically contradict every single letter in the RAID acronym.
    Connect it over 300 baud traditional telephone lines (with acoustic coupler, of course) for a perfect 0%! Or some kind of exotic variant of infiniband if you're going to focus on boosting cost to insane levels instead of dropping performance through the basement (heck, through the whole crust).



  • @dkf said:

    @Ronald said:
    Deploy this on a single SSD and you basically contradict every single letter in the RAID acronym.
    Connect it over 300 baud traditional telephone lines (with acoustic coupler, of course) for a perfect 0%! Or some kind of exotic variant of infiniband if you're going to focus on boosting cost to insane levels instead of dropping performance through the basement (heck, through the whole crust).

    Infiniband is awesome. The network card can write directly on the local storage (RDMA) without talking to the operating system... I think I'll put together a business plan and create a startup (anything about social networks or 3D printers should do) just so I can get VC and build a data center that runs on Infiniband. Then I'll keep overwriting employees hard disks as I try to build a PingMyDisk utility.



  • @RangerNS said:

    "our stupid tool doesn't allow this entirely reasonable option"

    Yeah, they want you to buy the PowerEdge 6xx-series for that it appears...



  • @Ronald said:

    So why does raid 5 even exist?
     

    @Ronald said:

    Bottom line: fuck RAID-5.

    Now, about why it does exist... There was a time when disks were expensive, so expensive that it was worth it to use specialized processors to calculate the parity, just to have that 1/3 of redundancy, instead of 1/2.



  • @Mcoder said:

    @Ronald said:

    So why does raid 5 even exist?
     

    @Ronald said:

    Bottom line: fuck RAID-5.

    Now, about why it does exist... There was a time when disks were expensive, so expensive that it was worth it to use specialized processors to calculate the parity, just to have that 1/3 of redundancy, instead of 1/2.

    Read the fucking threads this has been discussed extensively already in the parts you don't quote. Also it's not 1/3 for RAID-5, it depends on how many disks you put in the array. See: http://www.servethehome.com/raid-calculator/


  • Discourse touched me in a no-no place

    @Ronald said:

    Infiniband is awesome.
    Yes, but something for people with very well padded checkbooks indeed IIRC. We used to use it for the interconnect for our supercomputers.

    I bet EMC will sell you a RAID-9001 that connects over Infiniband, provided you've got the US national debt to pay for it…



  • Actually, I lied. Or Dells marketing and online catalog departments intentionally left things out.

    The SD card that plugs into the iDRAC card is for... something else... and there is an additional and unadvertised option for a special 2-port SD card holder, that fits inside onto the mobo, and doesn't need Extra Super Delux version of iDRAC; iDRAC takes 8 or 16 GB cards, the special holder 1 or 2 GB cards (and is two ports 'cause it can be mirrored).

    The iDRAC option (and Dell branded SD cards for iDRAC) are only available at dell.ca if you are logged in. The SD card holder thingy requires talking to a person, and... I guess... admitting... to them that the intended use for the server is as a VM host. Only then will Dell let you know about that option.

    Concerning RAID, and the better conversation this thread has turned into... Does it matter ("as much") when the RAID device is some crazy high end storage system. I guess, benchmark. With the SAN appliance in question fully populated at 120 disk, problems... 8 GB cache; writes will be committed when they hit memory, not disk.

    Benchmark, I guess.


  • Discourse touched me in a no-no place

    @RangerNS said:

    Concerning RAID, and the better conversation this thread has turned into... Does it matter ("as much") when the RAID device is some crazy high end storage system. I guess, benchmark. With the SAN appliance in question fully populated at 120 disk, problems... 8 GB cache; writes will be committed when they hit memory, not disk.

    That's great until you have to write more data than can fit in cache (which is where RAID 5 always falls down. Salesweasels will claim that their big cache will prevent performance issues, and it might even be true...until you saturate the cache.)

    @RangerNS said:

    Benchmark, I guess.

    I provided a single-point benchmark upthread: a pair of raid 0 or 1 (probably 1) disks roughly equalling the performance of (some fraction of) a 30-spindle raid 5 SAN, with a heavily-transactional database (plus whatever else was going on on the rest of the SAN, which included a couple of file shares.)



  • @RangerNS said:

    The SD card holder thingy requires talking to a person, and... I guess... admitting... to them that the intended use for the server is as a VM host. Only then will Dell let you know about that option.

    This is typical. It's not misleading, it's because some advanced configuration options often come with a specific firmware or are not compatible with other options. Dell is very good at mass customization but there are limits to what a web wizard can do.

    When you reach a certain amount of business with Dell, typically you don't go to the website anymore, you email the account manager and they come back with very different prices and options. The same applies to business with HP and IBM, but in my experience those companies come with a price premium that does not really mean better equipment. ProLiant or PowerEdge are very similar; actually when you deploy Linux in my experience RHEL and SEL have a better chance to work out of the box on PowerEdge.



  •  For some reason, every time I see this thread title, I think it says "Dickless server."



  • @Snooder said:

     For some reason, every time I see this thread title, I think it says "Dickless server."


    I was literally just about to post this.


  • Winner of the 2016 Presidential Election

    @eViLegion said:

    @Snooder said:

     For some reason, every time I see this thread title, I think it says "Dickless server."


    I was literally just about to post this.

    See tags of first reply.


Log in to reply
 

Looks like your connection to What the Daily WTF? was lost, please wait while we try to reconnect.