solaris x86 versus linux on the same hardware - Solaris

This is a discussion on solaris x86 versus linux on the same hardware - Solaris ; How much of the speed difference between a, for example, 500 Mhz SPARCstation and a 2.x Ghz Dell desktop pc is due to the cpu speed versus the operating system? I've always believed the mantra that RISC is so much ...

+ Reply to Thread
Page 1 of 2 1 2 LastLast
Results 1 to 20 of 22

Thread: solaris x86 versus linux on the same hardware

  1. solaris x86 versus linux on the same hardware

    How much of the speed difference between a, for example, 500 Mhz
    SPARCstation and a 2.x Ghz Dell desktop pc is due to the cpu speed
    versus the operating system? I've always believed the mantra that
    RISC is so much faster than CISC but was amazed at how much faster
    things are on the Dell running linux. Should I get the same speed
    increase if I replace linux with Solaris?

    Thanks

  2. Re: solaris x86 versus linux on the same hardware

    rusty@grunt.berkeley.edu wrote:
    > How much of the speed difference between a, for example, 500 Mhz
    > SPARCstation and a 2.x Ghz Dell desktop pc is due to the cpu speed
    > versus the operating system? I've always believed the mantra that
    > RISC is so much faster than CISC but was amazed at how much faster
    > things are on the Dell running linux. Should I get the same speed
    > increase if I replace linux with Solaris?
    >
    > Thanks


    You are likely to get as many different views as responses on this one,
    but here is my 2c.

    You are picking two machines of quite different ages I suspect. 500 MHz
    is certainly less than half (and probably only about a third) of the
    fastest SPARC chips. I don't know what the latest SPARC chips runs at,
    but I believe they are well over 1 GHz. Yet your 2.x GHz Dell is well
    over half the fastest possible speed of an Intel chip.

    Whilst I think your particular comparison is a bit unfair, I think the
    RISC/CISC advantage has gone now. CISC chips have some instructions that
    execute in very few cycles now, so the boundary between RISC and CISC
    has blurred to the point of almost non-existence.

    Where I used to work, we used Suns for numerical work as PCs were
    totally inadequate. But that was years ago and now for CPU intensive
    simulations, an x86 based system running Linux or Solaris will beat a
    SPARC system costing the same. (I'm talking about machines with 4 GB of
    RAM or less, I have no experience with the sort of simulations the Met
    office might perform)

    I don't think the operating system will have a significant effect on
    speed on the size of machines you are talking about. Stability yes, but
    not speed. So you will not gain any speed advantage switching from
    Solaris to Linux, or the other way around.

    SPARC is said to scale better than Linux, but are not going to see that
    on a single processor machine, or a machine with very few processors.

    The main advantage (as I see it) for SPARC hardware now in low-end
    machines like has tended to be stability. Cheap PCs have a very low mean
    time before failure (MTBF) in my experience. I'd like to see some hard
    data about some of the better PCs, such as Dell servers. The modern
    cheap Suns don't have the build quality of the older ones and some of
    the better PCs are pretty well built now. (I asked on one of the
    newsgroups the other day for facts about hardware relieabiltiy of
    Suns/PCs and never got a single response).

    If you go to big hardware, then I think then there is no contest and Sun
    hardware and software will win hards down, but on the lower-spec
    hardware, which does not need a machine room to work in, it is not clear
    to me Sun (SPARC) has any edge over x86. I suspect that is why Sun now
    ships a lot of boxes based on x86.


    --
    Dave K

    http://www.southminster-branch-line.org.uk/

    Please note my email address changes periodically to avoid spam.
    It is always of the form: month-year@domain. Hitting reply will work
    for a couple of months only. Later set it manually. The month is
    always written in 3 letters (e.g. Jan, not January etc)

  3. Re: solaris x86 versus linux on the same hardware

    rusty@grunt.berkeley.edu schrieb:

    > How much of the speed difference between a, for example, 500 Mhz
    > SPARCstation and a 2.x Ghz Dell desktop pc is due to the cpu speed
    > versus the operating system? I've always believed the mantra that
    > RISC is so much faster than CISC but was amazed at how much faster
    > things are on the Dell running linux. Should I get the same speed
    > increase if I replace linux with Solaris?


    Presumeably...

    SPARC was never known for very high performance. The 500MHz UltraSPARC
    II is around as fast as a P3 500/550MHz. Of course your Dell runs
    circles around it not only because it has a much faster cpu but also
    because it has much better I/O performance (i.e memory)...

    If you want Solaris going really fast and if you don't require SPARC
    binary compatibility you'll get the best performance out of Solaris on a
    PC...

    Benjamin

  4. Re: solaris x86 versus linux on the same hardware

    > SPARC was never known for very high performance. The 500MHz UltraSPARC
    > II is around as fast as a P3 500/550MHz. Of course your Dell runs
    > circles around it not only because it has a much faster cpu but also
    >because it has much better I/O performance (i.e memory)...


    Benjamin, can you show us some figures supporting that? I belive a
    reasonably new
    SPARC can compete with a x86 in some cases. Regarding memory I think
    that's somewhere where the SPARC based systems really is better. Also
    disk I/O is mostly
    an order of magnitude better as they run SCSI (or SAS) disks.

    So, please show us som figures!!!


  5. Re: solaris x86 versus linux on the same hardware

    Thommy M. wrote:
    > Also
    > disk I/O is mostly
    > an order of magnitude better as they run SCSI (or SAS) disks.
    >
    > So, please show us som figures!!!
    >


    That depends a lot on what you are doing. For a single user workstation,
    it really makes no difference. And even on servers, I think an order of
    magnitude is a gross exaggeration.

    I think now the advantage of SCSI is much less, and the price
    differential much higher, which makes it a lot less attractive.

    I'm not anti-SCSI. I have 7 Sun workstations here and all of them can
    only take SCSI. Even my PC, whilst having an IDE capability, is actually
    SCSI. But if I bought a new machine now for workstation use, I'd find it
    difficult to justify buying SCSI.


    --
    Dave K

    http://www.southminster-branch-line.org.uk/

    Please note my email address changes periodically to avoid spam.
    It is always of the form: month-year@domain. Hitting reply will work
    for a couple of months only. Later set it manually. The month is
    always written in 3 letters (e.g. Jan, not January etc)

  6. Re: solaris x86 versus linux on the same hardware

    Thommy M. schrieb:

    >>SPARC was never known for very high performance. The 500MHz UltraSPARC
    >>II is around as fast as a P3 500/550MHz. Of course your Dell runs
    >>circles around it not only because it has a much faster cpu but also
    >>because it has much better I/O performance (i.e memory)...

    >
    >
    > Benjamin, can you show us some figures supporting that?


    Sure, for example the figures on www.spec.org...

    > I belive a
    > reasonably new
    > SPARC can compete with a x86 in some cases.


    Well, a 500MHz UltraSPARC is hardly "reasonably new". In 2002 I got a
    new Blade 1000 with two 900MHZ USIII processors. I was a bit
    disappointed that my cheap Dell P4 1.5GHz ran circles around the Sun in
    terms of fp performance. Especially when my Dell was around 1200EUR
    (with 15" TFT) while the Bladed 1000 was over 20kEUR...

    > Regarding memory I think
    > that's somewhere where the SPARC based systems really is better.


    Nope. The 500MHz USII uses slow EDORAM, and even the more modern 550MHz
    USIIe with standard SDRAM is way behind every standard PC today which
    uses fast DDR- or DDR2-SDRAM in dual channel configuration...

    > Also
    > disk I/O is mostly
    > an order of magnitude better as they run SCSI (or SAS) disks.


    Well, for destop use SCSI isn't really faster (just magnitudes more
    expensive). That's why most high end desktop workstations today come
    with fast SATA drives...

    Of course it's different with huge servers that have multiple PCI-X
    busses and multiple processors. But even there Opteron and also Itanium
    simply kick asses. And Sun probably already is aware of that, look at
    the work they put in their Opteron products...

    Of course this all is quite technical. In the end it simply depends on
    the application and how it's optimized for a certain platform...

    Benjamin

  7. Re: solaris x86 versus linux on the same hardware

    Thanks for all of the great feedback. I have several Sun 1u Sparc
    servers that are 500 Mhz or thereabouts that I'll eventually replace
    and I wanted to be sure that getting Sun's AMD based boxes would give
    me a nice speed boost. The machines are lightly used, running apache
    with php, mysql, and tomcat.

    I've tried Linux but I've been doing Solaris for so long that it's
    hard to adjust to the differences. It's nice finally having an x86
    option.

    Thanks again.

  8. Re: solaris x86 versus linux on the same hardware

    rusty@grunt.berkeley.edu wrote:

    > Thanks for all of the great feedback. I have several Sun 1u Sparc
    > servers that are 500 Mhz or thereabouts that I'll eventually replace
    > and I wanted to be sure that getting Sun's AMD based boxes would give
    > me a nice speed boost. The machines are lightly used, running apache
    > with php, mysql, and tomcat.
    >
    > I've tried Linux but I've been doing Solaris for so long that it's
    > hard to adjust to the differences. It's nice finally having an x86
    > option.
    >
    > Thanks again.


    Solaris has been available for X86 for years.

    --
    The e-mail address in our reply-to line is reversed in an attempt to
    minimize spam. Our true address is of the form che...@prodigy.net.

  9. Re: solaris x86 versus linux on the same hardware

    http://www.spec.org/web2005/results/web2005.html
    Who's running circles around whom? (Is that correct English?)

    Wonder why Andy Bechtolsheim put SAS disks in the T2000 if SATA is that
    brilliant?

    Sun is putting their main efforts on Niagara and the upcoming Rock
    chips. That's why
    Andy is back in the company, to build systems around those CPU's...


  10. Re: solaris x86 versus linux on the same hardware

    In article <1136286748.982730.219680@g49g2000cwa.googlegroups. com>,
    Thommy M. wrote:
    >Wonder why Andy Bechtolsheim put SAS disks in the T2000 if SATA is that
    >brilliant?


    Ashlee Vance reports that Thumper uses SATA drives.

    | A system code-named Thumper should ship in the first half of 2006. The
    | 4U high system will hold two dual-core Opterons and support up to 16GB
    | of memory. A more unique part of the server will be Sun's use of 48
    | SATA drives. In addition, Thumper will have room for four Gigabit
    | Ethernet ports and make use of all four Hypertransport I/O links that
    | are unique to Opteron, according to information obtained by The
    | Register.

    He doesn't mention what kind of disk controllers it uses.
    According to LSI Logic, you can attach SATA disks to SAS controllers.


    John
    groenveld@acm.org

  11. Re: solaris x86 versus linux on the same hardware

    Thommy M. schrieb:

    > http://www.spec.org/web2005/results/web2005.html
    > Who's running circles around whom? (Is that correct English?)


    Yeah, right. Different systems running a different web server software
    on a different number of cpus. ThatÄs probably benchmark Sun likes ;-)

    Notice that the Sun needs 8 CPU-cores while the other systems have only
    1,2 or 4 cores. The performance per core is much better on x86
    (especially on Opterons which are missing here). The advantage of the
    T1000/T2000 is that they have a lots of cores on a single processor
    which takes less room and produces less heat and so is good for web
    servers...

    > Wonder why Andy Bechtolsheim put SAS disks in the T2000 if SATA is that
    > brilliant?


    Probably because SAS is more suitable for connecting a large number of
    drives which even can be located externally than SATA links. BTW: SAS
    can also be used for SATA disk drives...

    > Sun is putting their main efforts on Niagara and the upcoming Rock
    > chips. That's why
    > Andy is back in the company, to build systems around those CPU's...


    And that exactly has what relevance to the OPs question?

    Benjamin

  12. Re: solaris x86 versus linux on the same hardware

    >> Sun is putting their main efforts on Niagara and the upcoming Rock
    >> chips. That's why
    >> Andy is back in the company, to build systems around those CPU's...

    >
    >And that exactly has what relevance to the OPs question?


    It was you who said that Sun was putting much work at the Opteron. I
    say they are not,
    but still focusing on the SPARC.


  13. Re: solaris x86 versus linux on the same hardware

    On 4/1/06 12:47, in article
    1136378860.524903.249730@o13g2000cwo.googlegroups. com, "Thommy M."
    wrote:

    >>> Sun is putting their main efforts on Niagara and the upcoming Rock
    >>> chips. That's why
    >>> Andy is back in the company, to build systems around those CPU's...

    >>
    >> And that exactly has what relevance to the OPs question?

    >
    > It was you who said that Sun was putting much work at the Opteron. I
    > say they are not,
    > but still focusing on the SPARC.


    Are there any benchmarks comparing Linux performance on SPARC with Solaris
    performance on SPARC?

    Cheers,

    Chris


  14. Re: solaris x86 versus linux on the same hardware

    On Wed, 4 Jan 2006, Thommy M. wrote:

    > It was you who said that Sun was putting much work at the Opteron. I
    > say they are not,
    > but still focusing on the SPARC.


    Sun is focusing on both SPARC and Opteron. That's not surprising
    given that they have different price/performance sweet spots.

    --
    Rich Teer, SCNA, SCSA, OpenSolaris CAB member

    . * * . * .* .
    . * . .*
    President, * . . /\ ( . . *
    Rite Online Inc. . . / .\ . * .
    .*. / * \ . .
    . /* o \ .
    Voice: +1 (250) 979-1638 * '''||''' .
    URL: http://www.rite-group.com/rich ******************

  15. Re: solaris x86 versus linux on the same hardware

    Thommy M. schrieb:

    >>And that exactly has what relevance to the OPs question?

    >
    >
    > It was you who said that Sun was putting much work at the Opteron.


    Right, and they definitely are doing that...

    > I
    > say they are not,
    > but still focusing on the SPARC.


    Nope, a short look on Suns current product line and product plans should
    have told you that they indeed put efforts in both Opteron and SPARC.
    But SPARC became less and less competitive more and more areas while
    Opteron delivers usually a better performance at lower costs...

    Don't get me wrong, the SPARC machines are very good systems, but the
    air just gets thinner and thinner for it. The new UltraSPARC T1
    currently has the advantage of being a very small and powersaving
    multicore cpu but it's certainly not a Opteron/XEON killer in terms of
    performance...

    Benjamin

  16. Re: solaris x86 versus linux on the same hardware

    Why bother? I'd think that linux support on SPARC would be rather
    sketchy compared to on the x86. For example, drivers and whatever
    would lag in updates and features compared to x86, similarly for os
    features that take advantage of fancy cpu features.

    Chris Ridd writes:

    > On 4/1/06 12:47, in article
    > 1136378860.524903.249730@o13g2000cwo.googlegroups. com, "Thommy M."
    > wrote:
    >
    > >>> Sun is putting their main efforts on Niagara and the upcoming Rock
    > >>> chips. That's why
    > >>> Andy is back in the company, to build systems around those CPU's...
    > >>
    > >> And that exactly has what relevance to the OPs question?

    > >
    > > It was you who said that Sun was putting much work at the Opteron. I
    > > say they are not,
    > > but still focusing on the SPARC.

    >
    > Are there any benchmarks comparing Linux performance on SPARC with Solaris
    > performance on SPARC?
    >
    > Cheers,
    >
    > Chris


  17. Re: solaris x86 versus linux on the same hardware

    Benjamin Gawert wrote:

    > SPARC was never known for very high performance. The 500MHz UltraSPARC
    > II is around as fast as a P3 500/550MHz. Of course your Dell runs
    > circles around it not only because it has a much faster cpu but also
    > because it has much better I/O performance (i.e memory)...
    >


    Also keep in mind that US-II's are fully 64-bit processors
    whereas Pentium's at that time were 32-bit processors,
    which means that US can shuffle and process twice
    as much data per cycle. Most comparisons are for 32-bit
    applications where this advantage does not come into play.
    As soon as you need to address more than 2G, which we
    needed to do frequently in many simulations, you were
    pretty much dead with any Intel/Linux systems.


  18. Re: solaris x86 versus linux on the same hardware

    > Sun is focusing on both SPARC and Opteron. That's not surprising
    > given that they have different price/performance sweet spots.


    That's not what I hear from Sun insiders... But I think you get a
    different story depending
    on whom you're asking...


  19. Re: solaris x86 versus linux on the same hardware

    On Wed, 4 Jan 2006, Thommy M. wrote:

    > That's not what I hear from Sun insiders... But I think you get a
    > different story depending
    > on whom you're asking...


    I'm also "well connected", but given how much of the spotlight Sun
    has focussed on AMD systems recently, it would be hard to argue that
    SPARC is their only priority.

    Sun made the mistake of ignoring x86 once; they won't do it again.

    --
    Rich Teer, SCNA, SCSA, OpenSolaris CAB member

    . * * . * .* .
    . * . .*
    President, * . . /\ ( . . *
    Rite Online Inc. . . / .\ . * .
    .*. / * \ . .
    . /* o \ .
    Voice: +1 (250) 979-1638 * '''||''' .
    URL: http://www.rite-group.com/rich ******************

  20. Re: solaris x86 versus linux on the same hardware

    Chris Ridd wrote:

    >
    > Are there any benchmarks comparing Linux performance on SPARC with Solaris
    > performance on SPARC?


    Yes, and the results are not very conclusive.
    On 32 bit systems Linux is a tad quicker for most purposes, but support
    for the bigger Sun4D series machines is broken.
    On 64 bit systems Linux 2.4.x was about 15% faster than Solaris 8 for
    most purposes due to better kernel optimisation, but Solaris 10 has
    improved and has overtaken Linux 2.4.x.
    Linux 2.6.x is badly broken on all 32 bit Sun systems, and is a tad
    slower than Solaris 10 in my 64 bit tests due to kernel bloat.

    YMMV, there are other issues such as better SMP support in Solaris, as
    well as better support for FCAL. Linux OTOH supports more non-Sun PCI cards.

    On balance, Solaris 10 has more than overtaken Linux for performance on
    modern 64 bit Sparc systems, especially on the SunFire server range.

    On a more general point, Sun's Sparc systems outperform all X86 hardware
    on big iron, where total throughput is more important than simple CPU
    performance. The legacy X86 architecture is just not capable of
    supporting big systems.


+ Reply to Thread
Page 1 of 2 1 2 LastLast