Re: Even x86 is an option for legacy SPARC, these days
Why on earth would you emulate SPARC, when all Oracle software (including Solaris) runs just fine on x64? If you need SPARC, use SPARC. If you don't, then use x64 native.
When reports emerged early in September that Oracle was shedding hundreds of workers, it might have been dismissed as just the latest in a series of such moves. What caught the attention of industry watchers was where the axe was falling. Sources indicated those being laid off included many of the engineers responsible for its …
This post has been deleted by its author
This post has been deleted by its author
What I would like to know is what I don't know. OK, seems I have had a long day. So I understand that x86 is the most popular architecture and has the most applications available to run on it. That makes it a no brainer for most. I've never worked with SPARC, though there were plenty of Sun boxes sitting there doing their thing with no hassles back in my Telewest days.
What I don't understand is (sorry for bullet points its how my mind works):
- What does SPARC have to offer that x86 can't do as well?
- What would a SPARC development pipeline look like? (I.e. is it fine already?)
- If Oracle wanted to push it hard as an alternative to another architecture then how would they get value from that investment? (In the face of x86, failed Power, and upcoming ARM)
SPARC is prized among developers because it is NOT x86. When your code compiles cleanly on Linux and SPARC/Solaris, then you are reasonably sure that its portable. There is lots of compiler support for SPARC.
Microsoft once did exactly this thing. The original NT kernel was first prepared on MIPS. I think that x86 was the 3rd target platform.
The SPARC M2 is the most powerful 64-bit CPU design that is completely open and free for anyone to adapt. Alas, that has never excited great interest, even though it should.
One of the reasons I use SPARC is because it can't run code on its stack. That makes it immune to a class of attacks that can bypass firewalls. It is much harder to write Return Oriented Programming exploits for SPARC (and MIPS and ARM and POWER) than it is for X86.
It is also a much smaller target than x86 for hackers so they often just skip the systems.
Code development is just like Linux. Linux has always copied many things from Solaris and while Linux supports far more hardware, is it still catching up with 10 year old features from Solaris such as ZFS and Dtrace and Basic Security Modules. I can get an Linux system to audit every system call but the machine will be so busy doing audit work, it won't be able to do any real work. Any program that will compile on FreeBSD and Linux will most likely compile on SPARC without issue except for the coders that don't understand the "all-the-world's-a-VAX syndrome".
The disadvantage of SPARC today is that I simply don't need the new hardware. I figured their current 1RU offering is something on the order of 4 million percent faster than the $70,000 Suns we had in the mid 90s. Their newest 1RU box would replace a rack and a half of the last new 1RU SPARCS we bought new for the price of 4 of the those machines. I need more than one box for hardware failover, and my load isn't close to stressing the thing out.
I don't see SPARC increasing share unless they come out with a low cost appliance server but I can't see Oracle doing that.
I am with you Tim, except that I run OpenBSD and not Solaris. I too don't need more than one Sun box, except to support upgrades and failover, and multiple locations.
I would like to see Sun doing boxes with about the size, power and build quality of a T2000, but with the more recent architectures, and even lower power consumption - fanless operation?
Since colos charge by the Amp these days, the threat to Sparc is Arm, not x86.
The Cloud is not a better solution than when it was called Multics. its just that some people are too young to remember Multics.
When Sun4 workstations with SPARC chips came out in the mid 80's they were streets ahead of anything from the PC or DEC/VAX stables at that time, so its sad to see them fading away. But a 30 year innings is not bad for a wholly innovative development from a startup company that went against all the industry conventions at the time. Well done Sun!
Having been in IT for over 40 odd years, I've seen lots of great technologies come and, undeservedly go. We are pretty much faced with an Intel only future at one level, and in the cloud a decreasing number of players. The ultimate loss of SPARC, will leave IBM and ARM as competitors to Intel, AMD et al don't really count, and it could be said, neither does IBM.
IT is a bit too much of a "Me Too" (Martin Butler, of Butler Group) industry, unwilling to be different, but then complaining when there's no choice. It comes down, as always, use it or loose it, very few, comparatively, people use non-Intel based servers, certainly not enough to attract the required investment to develop, in any meaningful way. And fewer and fewer willing to accept the risk of doing different. Oh, and your Open Source new best friends, they are beginning to behave more like Oracle, I wonder how long it will be before they all do, those that are left that is.
The bright spot is that monopolistic behaviour allows disruptive technology to develop and be bought by the monopolies, or become new ones.
Opening up the Solaris source (again) would make life very ... interesting ... for illumos.
We've diverged quite a bit, and have enough changes of our own, to the point that pulling in changes would actually be quite difficult. Hardware support for current SPARC models would be good, but I'm not sure there's all that much else.
x86 cannot compete with SPARC as SPARC M7 is typically 2-3x faster than x86, up to 11x faster on some Enterprise workloads such as databases. Now the new SPARC M8 is twice as fast as M7, which means x86 is really slow in comparison. For instance, the SAP benchmark top spots are all dominated by SPARC. x86 is far below.
The last benchmark Oracle released for SAP was 2016, An 8 Socket Machine with 256 Cores.
Giving a SAPs per core performance of 2787. A total SAPs score of 713480.
In the same year a Dell Poweredge R730 had a SAPs per core of 2677. Yes this was only a 2 Socket box with a SAPs rating of 117780.
The same server with Skylake has a per core SAPs rating of 3129.
So yes SPARC won on overall numbers. However I am yet to meet the customer who has a DB server requirement of 713480 SAPS.
Even a requirement of 117780 SAPs for a DB server is very rare. Yes the whole landscape might require lots of SAPs but a bunch of 2 Socket x86 servers will a) be cheaper, b) won't have all your eggs in one basket.
Yes there are workloads that have been tuned for SPARC, likewise for intel. At the end of the day it comes down to cost. x86 is a good enough workhorse at a far cheaper price.
> SPARC M7 is typically 2-3x faster than x86, up to 11x faster
SPECint_rate numbers for SPARC M7 are here.
SPECint_rate numbers for Intel Xeon are here.
Intel Xeon 8180 outperforms SPARC M7. Comfortably so.
There are no recent numbers for SPECint SPARC. The most recent numbers for SPECint SPARC are from 2011. When compared to an Intel Xeon from 2011, the SPARC SPECint numbers look dismal.
Yes, SPARC M7 was typically 2-3x faster. Look back here:
I did not know about the new Intel 8180 cpu, thanks. As the SPARC M8 is faster than the M7, we have to wait and see the results of M8. I am confident M8 will be faster.
This article is complete hogwash. Oracle isn't cancelling SPARC nor Solaris and neither is Fujitsu cancelling SPARC either. This is all fact less rumors from anonymous cowards who clearly may be working for the competition. Unfortunately, even the analysts are being biased by these fake news sites. There are still 1000's of engineers dedicated to SPARC & Solaris and while its true that Oracle has done some layoffs in certain areas, it has significantly ramped up its hiring in other areas like IaaS which significantly includes SPARC & Solaris and will be releasing SPARC M8 as a service shortly-at the SAME EXACT PRICE AS INTEL X86 Bare Metal. SPARC M8 is by far the most significant processor developed in the world, showing a very significant 2x to 7x PER core performance advantage over the latest Intel Skylake processors (even faster for certain JAVA/JVM workloads that leverage the onboard DAX) and with 32 x cores per SPARC M8, is easily 2-4 times more powerful on a per chip basis compared to Intel and therefore lower system costs on a comparable performance scale. From a security perspective, SPARC M8 is the only CPU in the world to support/integrate the latest SHA-3 security standard being implemented by the worlds most securest banks and governments. SPARC M8's on-chip security encryption algorithms are easily 3x to 9x faster (using the SHA512 digest benchmark) versus Intel Skylake. If anyone is serious about security and worried about all the Lintel hacks/ransomware going on, wake up and evaluate SPARC! SPARC not only powers Oracle's SPARC servers, but also SuperCluster, MiniCluster as well as all ZFS storage products and many of the US government agencies and especially top security agencies run Solaris/SPARC for a reason. Even Oracles global single-instance ERP system is running on SuperCluster. Its proven, its bullet proof and its secure. And finally, SPARC is an IEEE standard and is one of the few non-proprietary Open-standards based processors in the world. Theres even open-source fully free versions of SPARC available. God forbid we depend on Intel as they have not delivered any major performance/core advantages in over 5 generations! Remember folks, if you move to the cloud, you're subscribing to cores!
M8 is fine, sure, but that was just about done. What about the roadmap after M8? My contacts inside of hardware tell me that the entire CAD group is gone, pretty much, as is much of design. Sustaining is in relatively good shape, but there were layoffs even there.
If there is a future for Sparc all Oracle has to do is to be open about it and the layoffs. But, it being Oracle, not likely.
Oracle doesn't tell its employees any more than it tells the outside world. Someone I know who is in the former PeopleSoft group had no clue about this until I told her.
Back when I was a customer I bought some of the first Sparc workstations. Too bad they couldn't sell it any more.
I say look up the Oracle SPARC roadmap from 2011 when Oracle first acquired Sun and the SPARC/Solaris team and compare to what has actually been delivered and you will see that Oracle has actually delivered new SPARC CPUs quicker to market and at much higher performance than the roadmap estimates portray.
Since the Sun acquisition in 2010, Oracle has released a total of 8 different generations of SPARC processors (Intel is on its 5th generation since Nehalem in 2010) having increased CHIP performance by atleast 10x on SPECint2006rate and 12x faster/core on OLTP showing a full 2x faster/core than the latest Skylake, how long will it take Intel to catch up with its ~15% increase every generation? Also, look at the Solaris releases. Oracle released 4 major updates to Solaris in the last 7 years having delivered leading edge technologies like:
Fully Integrated OpenStack distribution
Full Automation of software patches and updates, and other packaging improvements
Live migration of Solaris Kernel Zones
InfiniBand support for Kernel Zones
Virtual Clocks for Solaris Zones
VNICs on IPoIB
Periodic and Scheduled Services
Tailored Compliance Reporting
OpenBSD 5.5 Packet Filter
Integration with OpenStack Juno
And wait to see whats in Solaris 11.4. Its 5+ years ahead of Linux.
Please help me understand which other OS out there has such developments and advancements and is as integrated and complete as Solaris?
@AC, are you an Oracle employee? If so, you should say so. If not, you should say so. Posting such uplifting content as an AC makes no sense.
Yes, Solaris is great. No, it is not setting the world on fire. No, you can't blame a marketing budget. Oracle Marketing have enough cash to buy a small country (and sail to it, on an Americas Cup yacht, no less). Do you think Oracle's reputation for bait and switch and their track record with previous acquisitions might have something to do with it? We buyers simply can't rely on what they tell us no matter how grand it sounds.
As far as performance goes, SPARC is great, x86 is great, benchmarks are meaningless. We El Reg readers know that even if we sometimes like to compare chip weiners.
As far as technology innovation, I can't help but notice how much of it is improvements to a technology made elsewhere. OpenStack. BSD pf. CIFS/SMB. Not quite like the glory days of NIS, NFS, Java and so on. But I do agree that Solaris virtualization support is years beyond Linux.
> benchmarks are meaningless
SPEC benchmarks aren't based on some artificial and unrealistic programs that have no correlation to real world usage. The benchmarks are all based on real world open source programs. I fail to see how they can be labeled meaningless.
Are you suggesting that the speed at which a file can be compressed or decompressed with bzip2 - one of the benchmark tests in SPECint and SPECint_rate - is irrelevant or meaningless for a reald world use case?
Whether or not you agree with the selection of programs included in the benchmark is a moot point. Everyone runs the same benchmarks.
I'm sad to see Solaris fading out, but most of us who develop not-Oracle S/W products have already seen it fading for years. Before the Oracle purchase, Solaris (both SPARC and x86) was easily the biggest market share for the product I currently work on. After the purchase, new Solaris sales nosedived and never recovered - replaced by Linux, even though I think our salespeople are pretty HW/OS agnostic (Windows has stayed a a steady #2). Even migrations from Solaris to Linux have been heavy, much more common than HP/UX or AIX migrations to Linux.
A couple minor complaints with the article:
"[Sun] was largely responsible for the first commercial Unix-based workstation"
I am pretty sure Apollo gets that credit, though it was eclipsed in market share by Sun.
"Die-hard customers can be very stubborn"
In this case, I think "stubborn" isn't a fair description for many/most enterprise customers. It is about money. If a vendor is willing (and can be trusted) to continue decent, affordable support for years, then staying on a fading platform is not necessarily a bad choice. I remember a DEC VP back in the early 90s giving a presentation on why PDP-11 was (at that time) still a ~US$250M per year business for DEC. PC vendors would come into factories and show how a PC-based factory control system could save tens of thousands per year on hardware and support costs vs. their existing PDP. But the customer would find that the migration price tag was often US$ hundreds of thousands (much of it rewriting and testing custom device drivers). Suddenly those PDPs didn't look quite so expensive. So, in summary, many just take a "if it isn't broke, don't fix it" approach.
(Anon because of the first para)
Oracle *loves* SPARC so much, they want *all* of your money.
How to make sure a technology goes unused...
Live migration of guest domains with whole cores and a CPU cap does not conform to the terms of the hard partition licensing. When live migration is used in an Oracle VM server pool, hard partition licensing is not applicable. You must determine the number of guest domains running the Oracle Software and then license the same number of physical servers (starting with the largest servers ased on the CPU core count) up to the total number of the physical servers in the pool. For example, if a customer has a server pool with 32 servers and 20 guest domains running Oracle software within the server pool, the customer must license the 20 largest physical servers in the ool. If the customer is running 50 guest domains with Oracle Software in a pool of 32 physical servers, they need only to license the 32 physical servers in the pool.
I think this uncertainty over SPARC plays into a general dissatisfaction rising in the Oracle user community over the quality of support they get.
My own upline has gotten VERY militant and hostile to Oracle staff over demonstrated failures to deliver product features and resolve issues in a timely manner, and we've been a loyal customer of both Sun and Oracle for decades.
We have a couple of IBM power series "mainframe" stacks only because of the lack of directional strategy coming out of Oracle regarding their plans for Sun-legacy software and hardware in the six months after the buy-out, and that move was so popular we had an SA who outright refused to work on them. (Fortunately for him and his family he had his work cut out on our remaining Sun kit.)
It is obvious that big computer companies want to go the appliance or cloud routes - that way they get to sell you the kit AND dun you for the SA work needed to get it all humming.
It's funny that after all these years I see the industry returning to the "agency" model of computing that was a very popular option when mainframes rules but only the very rich could afford them.
Biting the hand that feeds IT © 1998–2020