help, which Linux OS for clusters?

Discussion in 'Other Operating Systems' started by Mjollnir, Oct 1, 2007.

  1. Mjollnir

    Mjollnir Member

    Joined:
    Apr 20, 2002
    Messages:
    6,089
    Location:
    Melbourne, VIC3104
    Hi,

    Can anyone help me choosing a OS?

    I just joined a company, they asked me to choose a linux distro for a project.

    we are going to buy 6-8 rack mount dual core servers each with 8gb ram and 1.5TB HDD, that running either MySQL or Oracle Server that keeps 10 millions+ records of our clients.

    rack mount servers are to be storing the same database and to be queried and updated constantly, (normally expect to have 1000 online users). There are gonna be 2 load balancing server for this as well.

    I need to prepare a short presentation to everyone telling them which Linux distro is the best choice that can give stable 7/24 uptime and reasonable licensing price.

    Is there any people have some experience on this kind of work?? plz give some advice. thanks in advance!
     
  2. Snoops

    Snoops Member

    Joined:
    Jan 17, 2004
    Messages:
    1,458
    Location:
    Brisbane
    MySQL can run in cluster modes, so there is really no reason to worry about clustering linux.

    I would run Red Hat Enterprise or CentOS.

    The network behind them will be more important, you should have at least 3 networks -

    1) for all the mysql servers to talk

    2&3) failover/loadbalance access to the DB.
     
  3. SpecGen

    SpecGen Member

    Joined:
    Jun 28, 2001
    Messages:
    513
    Location:
    Brisbane
    I'd probably look first at which distributions are supported by the software vendor, and then see if the hardware vendor supports any of those.

    It'd probably be much different from what i'd be running if my decision was final, but 6-8 servers of that spec will be a significant cost, if you're running oracle it will probably double it.

    I'd want to be sure I could get good support quickly and efficiently to make my investment the most productive.

    My guess is that Redhat will be the only one on both the software and hardware supported list.
     
  4. Snoops

    Snoops Member

    Joined:
    Jan 17, 2004
    Messages:
    1,458
    Location:
    Brisbane
    BTW that is a hell of alot of capacity for such a database.

    10M records / 1000 users.... its above the normal, but wouldnt be any different to ocau on a "lets break a record night!"

    My main db runs 8 cpus (2 x quad core 2.66 xeons) has a few binary tables(about 45gig of binary) and around 18M records.

    We wouldnt ever have 1000 users logged in at once, but a search result (which uses thumbnails; results in >100 db selects) and about 10 db updates.

    Typically we have around 400 users logged during peak times...

    Or around 5000 db hits per minute. Server rarely pushes 1.00.
    (keep in mind, webserver is seperate)
     
  5. OP
    OP
    Mjollnir

    Mjollnir Member

    Joined:
    Apr 20, 2002
    Messages:
    6,089
    Location:
    Melbourne, VIC3104
    ok, i got it, thanks.

    this is for a medical record systems. doctors are constantly looking for test results.. therefore 1k online user is normal.

    can you tell me some more why Red Hat Enterprise is better than other Linux OS? Thanks.
     
  6. Snoops

    Snoops Member

    Joined:
    Jan 17, 2004
    Messages:
    1,458
    Location:
    Brisbane
    ...

    1 reason

    Vendor Support
     
  7. Doc-of-FC

    Doc-of-FC Member

    Joined:
    Aug 30, 2001
    Messages:
    3,284
    Location:
    Canberra
    I'm sorry to sound bias, but anything to do with enterprise level DB esp including medical records of people considering we're talking about test results which may have a life and death decision.
    For a server cluster i wouldn't go past anything but Oracle Linux running Oracle 10G back ended to something like a NetApp SAN on top of that i would go with Oracle Unbreakable Linux for package management.

    whats cheaper, the cost of implementing a 5 9's system 99.999% uptime or paying the compensation because of a db bug.

    for a contact regarding Oracle DB Specialists PM me and I'll drop you a reliable company.
     
  8. elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    35,906
    Location:
    Brisbane
    I would first evaluate the vendor support.

    So many people tell me to use RHEL or SLED because of "vendor support". Personal experience with both has shown me that neither are always worth the money, especially if you are going to be doing anything outside of a stock-standard install (or god forbid, need to do something like patch third-party code into their default provided kernel - eg: IPSec/OpenSWAN, DRBD for use with LinuxHA/Heartbeat/MySQL HA, etc).

    Doing anything custom typically yields vendors snubbing you unless you fall back to defaults, which typically makes your system useless (eg: I use DRBD extensively for MySQL high availability, and just can't live without it).

    Similarly, I've seen the same sorts of failures with hardware vendors. Sun Data (IBM reseller) have failed me time and time again supporting IBM xSeries hardware with Linux installed. A recent set of servers bought had RAID cards that didn't allow us to monitor the RAID status in realtime. "Senior engineers" were sent onsite to flash the RAID card firmware to solve the issue. However, the engineers believed that flashing LSI controllers with Adaptec firmware would make them work with Adaptec tools! What planet these guys live on, I don't know.

    Moral of the story: don't buy software just for vendor support if the vendor support sucks. I now use Debian exclusively on the above servers, and have saved thousands across 40 sites running multiple servers in each site, and have a system that is more configurable, more customisable, and more easily upgraded when the need arises.

    And before anyone cracks the shits: no, not all vendor support sucks. But likewise, not all of it is golden either. Do your research first.
     
  9. Heist

    Heist Member

    Joined:
    Jul 23, 2001
    Messages:
    1,192
    Location:
    Sydney
    No offence, but this sounds like homework! I find it a little hard to believe such a choice would be left to a situation like this. If I were in this situation and didn't fully understand the repercussions of the decision I'd be making, I'd push to engage someone for professional services.

    That doesn't mean choosing a vendor yet, but someone with more experience can make a better recommendation on a vendor and for what reasons, along with help you along the process.
     
  10. predder

    predder Member

    Joined:
    Jun 24, 2003
    Messages:
    1,364
    Location:
    Gold Coast, Qld.
    I look after a MySQL replication pair at work, we use Debian exclusively and have had no major issues with it distribution wise however there is no official support as it is a community project. If you choose MySQL over Oracle - I would suggest looking at the MySQL enterprise version and support package available from MySQL AB. They can help you with performance tuning and HA. Also, before jumping into MySQL cluster I would also firmly suggest reading the known faults and limitations listing. The two real major ones are that tables must be all in memory as of version 5.0 (current stable afaik) and you are stuck with the NDB "engine" (Rather than myisam or innodb). Also, if you need low latency performance you'll most likely need to interconnect boxes with dolphin SCI cards.

    I find it very interesting that you've selected the hardware before the OS and DBMS software. Surely a SAN might be suitable for your needs and might be worth considering? What are the hardware requirements of an Oracle HA solution if you decide to go for that over MySQL?

    On a side note: I want your job.

    Edit: Heist has said what I was thinking.. in a nutshell "if you're asking these sort of questions I don't see how you could be doing your job effectively without a _lot_ of research. There are professional consulting services that could make suggestions based on your situation, why not engage someone like that? If you have 1000 concurrent users surely you can afford it"
     
    Last edited: Oct 1, 2007
  11. lavi

    lavi Member

    Joined:
    Dec 20, 2002
    Messages:
    4,004
    Location:
    Brisbane
    i would get less servers and buy an EMC Sym for storage with 4GbE HBA's

    but that is just my oppinion as as soon as you "swap" your db performance goes into the ground, the sym should give you very fast storage providing you use it accordingly, the more spindless the better (most of the time)

    if every transaction could be 100.000$ i'd personally go for a GS320 with 16x proc and 32gb ram hooked up to a EMC sym or eva 8xxx, if you can't afford a GS320 (and trust me, they simply have astonishing power and reliablilty) the Integrity is not bad.

    horses for courses but it all comes down to how much a lost transaction can cost you :) if it's 100.000$ then do the math

    as for linux, i would not run linux. i know me bad but if you want stability why do you look at the OS first? look at your db engine, if you use oracle or mysql or whaterver, use the OS that best matches the db.

    i think this is homework or some shit as no one would buy the hardware then think of the rest
     
    Last edited: Oct 1, 2007
  12. GumbyNoTalent

    GumbyNoTalent Member

    Joined:
    Jan 8, 2003
    Messages:
    6,980
    Location:
    Briz Vegas
    +1 to storage speed as 1000 users doing Queries does not require a cluster of 6-8 machines, and by simultaneous we mean all at the same time not just logged in. When I worked at Royal Perth Hospital there where 1000+ simultaneous sessions daily on iSoft, so if you use iSoft then you will be forced to use Oracle.

    Now for storage, the current system I'm working on was disk bound with average response times of .1 - .25 per query on an old legacy system with a middleware web interface gateway, the new system was considered over kill and couldn't be cost justified by those opposing it. Now our response times are .003 of a second 30 - 80 times faster, monthly reporting and data mining has been reduced from 4.5 hrs to 40mins, weekly processing 40mins to 3-4mins and daily start and finish processing 20mins to 3mins just the productivity gains in cost of wages for out of hours processing will pay for itself in 9 months. :)

    Remember having all the processing power in the world will not help is the system is disk bound.

    EDIT :: I think I got it wrong, iSoft runs on a MSSQL or Oracle platform, can't remember exactly as we had both DB's and a multitude of applications from just as many vendors.
     
    Last edited: Oct 2, 2007
  13. Doc-of-FC

    Doc-of-FC Member

    Joined:
    Aug 30, 2001
    Messages:
    3,284
    Location:
    Canberra
    /me loves oracle 10G ADDM, as efficent queries and index's will yield amazing speed improvements on their own, let alone back ended to NetAPP SAN with 20 spindles.
     
  14. Bangers

    Bangers Member

    Joined:
    Dec 25, 2001
    Messages:
    7,254
    Location:
    Silicon Valley
    Solaris + VCS.
    Solaris + RAC.

    I'd go RAC, if you can afford it. VCS Clusters running Oracle are a bitch, specially with 10TB+. Lots of un though.
     
  15. elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    35,906
    Location:
    Brisbane
    Call me old and cynical, but it sounds to me like you lot haven't worked for a wide enough array of businesses to build up a healthy dose of cynicism.

    There are people on these forums who have worked on much bigger systems than me. There are people on these forums who have worked on much more expensive hardware than me. But I'm fairly confident I trump the lot of you when it comes to doing contracts for bonehead IT managers and CIOs who do very stupid shit - like choosing a hardware platform first, and then trying retroactively to hammer in a software/os/db solution to fit it, usually at great cost and delay to the parent company in question.

    The longer I work in IT, the more I wish I was a carpenter.
     
  16. joe_sixpack

    joe_sixpack Member

    Joined:
    Jan 21, 2002
    Messages:
    2,850
    Location:
    Brisbane
    If you go with Oracle, you need to be aware that the installation doesn't support a lot of Linux distro's natively. Even RHEL 5 requires a small change to the redhat-release file to install it (with out switching on the ignore params option).

    If you can afford to fit out a full blown Oracle cluster, the cost of a vendor supported Linux distro (RHEL, SUSE) is a sidenote.

    Further more, I'd recommend running a 64bit OS for Oracle servers with more then 6GB RAM. Memory / kernel tuning becomes a pain on 32bit OS's, still can be relatively easy if you're knowlegable in memory/buffer params.

    6 - 8 Intel dual core servers (2 of them as load balancers..) :lol: You're hardware choice seems rather silly, when you could probably just use a couple of IBM POWER or SUN SPARC based server to do the same job.
     
  17. Doc-of-FC

    Doc-of-FC Member

    Joined:
    Aug 30, 2001
    Messages:
    3,284
    Location:
    Canberra
    hardware wise for Oracle DB, I would have gone with 2 x 2950's running RHEL5, Oracle in a RAC configuration and a decent amount of memory depending on overall DB transactions and back ending the DB to a san over fibrechannel.

    we have our current oracle 10G DB running within ESX 3i on a cluster of decked out 6850's (4 x dual core processors, 64GB ram), vmotion, HA, automatic failover, all back ended to a NetApp SAN/NAS.
     
  18. Bangers

    Bangers Member

    Joined:
    Dec 25, 2001
    Messages:
    7,254
    Location:
    Silicon Valley
    Rrrright. Not only did I quiver from you recommending x86 with Oracle, but then you make an assumption on hardware without seeing IO or transactions.

    You suggest running RAC over two 2950's? I'm honestly curious to see your reasons. When I think Enterprise Oracle DB, I think E25K's, RAC'd E6900's with partitioning etc etc. I don't think x86 Oracle inside ESX. It is interesting to see how others have chosen to implement Oracle, but I do appreciate the money and business reasons behind other possible implementations.
     
  19. joe_sixpack

    joe_sixpack Member

    Joined:
    Jan 21, 2002
    Messages:
    2,850
    Location:
    Brisbane
    Oracle inside ESX server? strange..

    That's kinda like having someone heavy stand on your sprinkler hose; water still flows but it just isn't the same pressure. :confused:
     
  20. Crinos

    Crinos Member

    Joined:
    Jul 1, 2002
    Messages:
    4,027
    Location:
    Tasmania
    Not really, if the VMs have direct connections to the SAN and ample resources available to it, it's not going to perform any worse than physical hardware(or negligibly worse at most). Hardly a fat person on a hose ;)
     

Share This Page