Announcement

Collapse
No announcement yet.

RAID performance

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • RAID performance

    first, apologies for probably asking questions that have been answered before, but i have spent a long time reading about this and am confused. i am creating a new thread because i have used google, rather than not


    i have a few machines to test but am limited to what i can do with them, but would appreciate any suggestions. i will have to make do with what i have as i cannot afford to be spending several hundred pounds on hardware controllers. so what i have read is:


    1. windows software RAID5 is so poor at writes it is not worth using

    ..other than for situations where economy is preferred over performance. i have seen many people debate fiercely about write speeds as low as 5MB/s, i am not sure how to test this myself but well, i have just coped 250GB from one windows RAID5 box to another over a gigabit LAN, in around 14 hours. i was under the impression it was due to it being an older system and limited by the PCI bus, but now i am not so sure.

    does windows RAID5 also suffer this much at reads? ie, once the data is finally on the volume it will be the expected speed of the multiple of the number of disks, minus one?


    2. FAKERAID 'IS' windows/software RAID

    so a FAKERAID 5's performance will be equal to windows' RAID5 performance?


    3. RAID10 is better than RAID01

    i have some systems with FAKERAID 4+ SATA ports, if i create a 'RAID10' set is it definitely '10' and not '01'?

    also at least one system has 2 or more 2-port SATA controllers. would i be best to create RAID1 sets in FAKERAID, then stripe them in windows for RAID0? is this even a good idea, or at least better than software RAID5?


    4. linux RAID5 doesn't suffer the write performance of windows RAID5

    i either misunderstood or was the question that attracted so many flames on other forums i couldn't make out an answer..




    most machines i have atm are either FAKERAID5, or a single boot drive plus a separate set of disks for windows RAID5. i am considering making as many as possible FAKERAID10, i can certainly use one for now to test this.


    thanks in advance, i appreciate any help. i am so lost now on this subject i'm not sure if i'm even asking the right questions!


    cheers

  • #2
    Re: RAID performance

    Dude, what a mess you have here...
    First of all, take a look at this site for a (cute, I would say) explanation of the different terms in RAID.
    Now: RAID (Redundant Array of Inexpensive Disks) can be achieved in two major ways: software-based or hardware-based.
    Software RAID (not only Windows, but Linux also) is the RAID that is managed by the Operating System itself.
    Wait, I should I rewrite the same info others wrote already? Take a look at what Wikipedia has to say about it.
    If, after all this reading, you'll still have questions, we're here...

    Sorin Solomon


    In order to succeed, your desire for success should be greater than your fear of failure.
    -

    Comment


    • #3
      Re: RAID performance

      Originally posted by sentofuno View Post
      1. windows software RAID5 is so poor at writes it is not worth using
      software raid 5 is not only slow, but very unsafe. it?s like a russian roulette, where the gun points to your data (and possibly your job)

      ..other than for situations where economy is preferred over performance. i have seen many people debate fiercely about write speeds as low as 5MB/s, i am not sure how to test this myself but well, i have just coped 250GB from one windows RAID5 box to another over a gigabit LAN, in around 14 hours. i was under the impression it was due to it being an older system and limited by the PCI bus, but now i am not so sure.
      IOmeter is a good software for io testing
      anyhow, might be quite a few bottlenecks on the way between two machines

      does windows RAID5 also suffer this much at reads? ie, once the data is finally on the volume it will be the expected speed of the multiple of the number of disks, minus one?
      it depends on how many disks are in use. raid5 loses the speed edge at about 5-6 disks. plus the software based overhead.


      2. FAKERAID 'IS' windows/software RAID
      fakeraid is raid based on a hardware raid controller, that actually only has a nice bios setup interface, and will not work without a driver. through that driver, the card presents the OS with the raid arrays, but the card has no processing capabilities, so it stresses the CPU for parity calculations. a normal raid controller will calculate parity, consistency checks etc, on it? own. that is why the cards that cannot calculate on their own, are called fakeraid - they seem like the real thing, but behind the curtains, they are as bad as software.

      so a FAKERAID 5's performance will be equal to windows' RAID5 performance?
      might be faster or slower, it all depends. but they both lose to a real raid solution.


      3. RAID10 is better than RAID01
      BS

      i have some systems with FAKERAID 4+ SATA ports, if i create a 'RAID10' set is it definitely '10' and not '01'?
      depends on whether you stripe over two mirrors or vice versa.

      also at least one system has 2 or more 2-port SATA controllers. would i be best to create RAID1 sets in FAKERAID, then stripe them in windows for RAID0? is this even a good idea, or at least better than software RAID5?
      bad idea no matter how you look at it. will work, but remember the russian roulette?

      4. linux RAID5 doesn't suffer the write performance of windows RAID5
      it?s kernel based, so it operates on a lower level. it is also much more flexible. but as for safety... same roulette.
      ________
      Extreme Q Vaporizer
      Last edited by DYasny; 6th March 2011, 19:15.
      Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

      BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

      Comment


      • #4
        Re: RAID performance

        thanks very much for the replies,

        sorinso, i appreciate the link. but it is as you say 'cute' to me too, i am quite comfortable with the theory of each RAID level, its the issues in real life performance that has me confused

        DYasny, your info answers a lot of questions, thank you. could you elaborate on a few things so i am clear?



        software raid 5 is not only slow, but very unsafe. its like a russian roulette, where the gun points to your data (and possibly your job)
        i'm not sure how hardware, software or FAKERAID differ in terms of fault tolerance. i assumed in any setup the idea is if one drive dies, and is replaced and resynced before another dies, the data is safe? is that not the case?

        Quote:
        does windows RAID5 also suffer this much at reads? ie, once the data is finally on the volume it will be the expected speed of the multiple of the number of disks, minus one?

        it depends on how many disks are in use. raid5 loses the speed edge at about 5-6 disks. plus the software based overhead.
        i have just built another machine with 4 drives in FAKERAID5 intended for a home media server. it can stream video happily to two other machines on a gigabit network fine. so i guess its proven to be acceptable in that regard - but atm its not too late to make changes, are there any other issues that might become a nasty surprise later on with this setup?


        Quote:
        3. RAID10 is better than RAID01

        BS
        BS - they are equal or BS - the opposite is true?


        Quote:
        also at least one system has 2 or more 2-port SATA controllers. would i be best to create RAID1 sets in FAKERAID, then stripe them in windows for RAID0? is this even a good idea, or at least better than software RAID5?

        bad idea no matter how you look at it. will work, but remember the russian roulette?
        in this scenario, if there are 6 disks i would have thought up to 3 drives could die, assuming only one each in the 3 RAID1 arrays died. i'm not sure what you mean about russian roulette, is there something fundamentally unsafe about software RAID or FAKERAID?


        Quote:
        4. linux RAID5 doesn't suffer the write performance of windows RAID5

        its kernel based, so it operates on a lower level. it is also much more flexible. but as for safety... same roulette.
        sorry just want to clarify - so does linux not suffer the write performance of windows, at all? or much less so? is it better, but still bad?


        thanks again to both for taking the time to reply, i appreciate the help

        Comment


        • #5
          Re: RAID performance

          Originally posted by sentofuno View Post
          i'm not sure how hardware, software or FAKERAID differ in terms of fault tolerance. i assumed in any setup the idea is if one drive dies, and is replaced and resynced before another dies, the data is safe? is that not the case?
          man, you have to read up on what a raid controller does. it is far more than a simple parity check algorithm.

          i have just built another machine with 4 drives in FAKERAID5 intended for a home media server. it can stream video happily to two other machines on a gigabit network fine. so i guess its proven to be acceptable in that regard - but atm its not too late to make changes, are there any other issues that might become a nasty surprise later on with this setup?
          yes, no caching - writeback or readahead possible. and what happens if the server loses power during heavy i/o?

          BS - they are equal or BS - the opposite is true?
          every particular installation is different

          in this scenario, if there are 6 disks i would have thought up to 3 drives could die, assuming only one each in the 3 RAID1 arrays died. i'm not sure what you mean about russian roulette, is there something fundamentally unsafe about software RAID or FAKERAID?
          yes, as I said, just imagine what happens if the server dies during a heavy i/o load

          sorry just want to clarify - so does linux not suffer the write performance of windows, at all? or much less so? is it better, but still bad?
          it suffers less, due to a more effective implementation. but it is still unsafe
          ________
          Xanax Rehab Forum
          Last edited by DYasny; 6th March 2011, 19:15.
          Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

          BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

          Comment


          • #6
            Re: RAID performance

            ok, i have read up a little more on the dangers of write caching. unfortunately its nothing i can prevent, as i say there is no way i can afford to replace all my RAID controllers with proper hardware solutions.

            the option to disable caching in windows appears to be greyed out for me. i'll look into buying a UPS when i can.

            Comment


            • #7
              Re: RAID performance

              UPS, redundant power supplies, hotspares and ulitmately raid 10/50 or at least 6.
              ________
              Omaha motor company assembly plant
              Last edited by DYasny; 6th March 2011, 19:16.
              Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

              BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

              Comment


              • #8
                Re: RAID performance

                Notice that anything above RAID6 (including) will need a more powerful RAID controller, with 256MB RAM on it (at least), due to the huge quantity of calculations the controller needs to do. Anything less and you will face performance degradation. A battery-backed cache is also a need.

                Sorin Solomon


                In order to succeed, your desire for success should be greater than your fear of failure.
                -

                Comment


                • #9
                  Re: RAID performance

                  Originally posted by sorinso View Post
                  Notice that anything above RAID6 (including) will need a more powerful RAID controller, with 256MB RAM on it (at least), due to the huge quantity of calculations the controller needs to do. Anything less and you will face performance degradation. A battery-backed cache is also a need.
                  memory is there for caching, not calculations
                  what you need is a good calculating CPU on the controller. I?d go for LSI or 3ware. Dell PERC6 is based on LSI, and is pretty good as well
                  ________
                  Mazda Cosmo History
                  Last edited by DYasny; 6th March 2011, 19:16.
                  Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

                  BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

                  Comment


                  • #10
                    Re: RAID performance

                    thanks for the advice, but as i said i will have to make do with what i have, and optimize it as best i can. these servers are all ones i use at home, i don't have the kind of money to spend on this level of tolerance.

                    Comment


                    • #11
                      Re: RAID performance

                      in that case, I?d invest in a backup solution - a NAS box, or some external drives
                      ________
                      Vaporizer affiliates
                      Last edited by DYasny; 6th March 2011, 19:16.
                      Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

                      BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

                      Comment


                      • #12
                        Re: RAID performance

                        these servers are the backups, the important data exists on several other machines i use, which backup to the servers i have mentioned. much of the data on the servers are acronis images of other machines.

                        i was considering FreeNAS in the future and reuse one of my windows boxes - one of the reasons i asked about whether the poor write performance was specific to windows, or to any software or FAKERAID RAID5. i assume linux's performance at software RAID is similar to BSD?


                        sorry if i've not been clear about why i'm asking these questions. what i wanted to know is, assuming i only reuse what i have (motherboards with FAKERAID, and some 2-port PCI FAKERAID), and assuming i value performance over loss in storage space, is it worth making the change to RAID10?

                        if the performance of FAKERAID RAID10 is what i expect it to be, ie the speed of a single drive multiplied by half the number of total drives, minus some negligible overhead, i think my question has been answered.


                        i will revisit the question of linux/BSD software RAID5 performance when i am ready to run my own tests, i think

                        Comment


                        • #13
                          Re: RAID performance

                          linux and BSD are different systems. to test, simply try to install a samba server on linux and on freeBSD, and test with iometer to see which is better suited for your environment

                          make sure you tweak samba to fit the network, since it is highly configurable, and is known to provide much higher performance when configured correctly

                          sample samba configs are easily found on google
                          ________
                          Lincoln-zephyr v12 engine specifications
                          Last edited by DYasny; 6th March 2011, 19:17.
                          Real stupidity always beats Artificial Intelligence (c) Terry Pratchett

                          BA (BM), RHCE, MCSE, DCSE, Linux+, Network+

                          Comment


                          • #14
                            Re: RAID performance

                            Originally posted by DYasny View Post
                            memory is there for caching, not calculations
                            Yeah, I wrote it too fast and did not check...
                            What I meant is that the controller has a lot of calculations to do, so the data to be written on the disk should be stored in the cache. To avoid data loss in case of power outage (or such) while there is data in the cache, it should be backed up by battery.

                            Sorin Solomon


                            In order to succeed, your desire for success should be greater than your fear of failure.
                            -

                            Comment


                            • #15
                              Re: RAID performance

                              sentofuno:

                              Sounds like you have a fixed set of hardware with limited capability and are more or less leading us on a "wild goose chase".

                              Points to appreciate (and I've dealt with RAID when "inexpensive" held a pretty price tag):

                              1) Nobody rides for free; you get what you pay for; every other cliche in the book...

                              A good RAID controller is still not inexpensive by consumer standards but well worth it.

                              2) FAKERAID (I really dislike that term) is still RAID -- just somewhere between software and hardware RAID. RAID is RAID (except RAID 0 as it is not REDUNDANT!!!!). I happily run ICHRn RAID and SATA RAID that typically comes on m'boards all the time. ICHRn is nice in a RAID 1 configuration as it reads at almost RAID 0 speeds.

                              3) RAID 10 (or RAID 01) is overrated in my experience from a performance perspective. Perf-wise both should be equivalent. Failure-wise, I believe RAID 10 can take a few more hits (as long as not on the same mirrored pair). My experience with RAID 10 has been exclusively on 3ware Escalade products and I have not been impressed.

                              4) Use real world benchmarks. Nothing else compares.

                              5) Consider several smaller RAID volumes over one big monster. Depending on your controller(s) and usage, it could make a huge difference.

                              6) I've found over the years that larger, faster, better cached drives have more of a performance impact than RAID. Bytes per revolution seems to be the biggest determinant.
                              Cheers,

                              Rick

                              ** Remember to give credit where credit is due and leave reputation points sigpic where appropriate **

                              2006-2099 R Valstar. This post is offered "as is" for discussion purposes only with no express or implied warranty of any kind including, but not limited to, correctness or fitness for use. Nothing herein shall be construed as advice. Attempting any activity based on information in this post is done at your own risk.

                              Comment

                              Working...
                              X