Oracle's bright new Sonoma SPARCs hint at own-tech cloud

Ever since Oracle revealed its new hyperscale Sonoma SPARC chippery, I've been trying to figure out who would buy it and why. And I can think of only one customer: Oracle itself. There's very little enthusiasm around the world for clouds built on anything other than x86. Sure, the occasional ARM server has made inroads here …

[iOS6 user]

M7 on doing SQL memory scan can do 120 GB/sec, whereas an x86 runs queries at ~5Gb/sec (in both cases it is max theoretical speed) .. and this is why in M7/Sonoma CPUs is build IB controller.

Which one ARM CPU can do what x86 does with own memory subsystem?

Please try compare how many x86 machines running parallel will be necessary to have to have the same avg speed. After this try to compare price of the x86 and Sonoma HW. On top of this you must add power consumption of both solutions. At the end multiple powering costs by two to have real powering and cooling costs. Try to compare DC footprint costs as well .. maintenance costs ..

Please remember as well that we are talking about raw memory scan. M7 DAX (in CPU database accelerator subsystem which is present in Sonoma as well) can speedup those 120 GB/sec by almost factor of compression ration with which database database data can be compressed. With columnar compression Oracle DB data can be compress with even 5 to 10 or more compression ratio.

Trust me only in London area you can find more than few dozens of Oracle customers with so high needs to have so high SQL scan speed.

Sonoma probably would be not able to gain speed of full M7 but even this will attract enough number of customers to have income higher than costs.

PS. I'm really surprised how shallow knowledge has author about high end systems.

[Nate Amsden]

I think the author is arguing that that high end market(that continues to shrink as x86 power grows) you talk about isn't going to be enough to make investment in SPARC worth while for Oracle, that they want to reach into the more general purpose server market, like Sun tried to do I remember with at least the T series? On paper at the time they seemed to me anyway(from what I recall) to be decent chips but they never went anywhere(relatively speaking market share wise).

Not even Intel has been able to convince customers to get off of x86. Myself I think ARM is a dead end for servers Intel will stomp on them. I do miss AMD's high(er) end CPUs (I have a bunch of HP DL385 G7s with Opteron 61/6200s) but AMD abandoned them years ago now, and bet the farm on ARM as well (too bad, and they burned their engineering talent and product pipelines so getting back into high end X86 at this point is probably impossible for them).

There will be a market for high end CPUs, though I suspect SPARC will go away, Oracle is pretty ruthless on profits and stuff. IBM will probably keep Power running for the foreseeable future, I don't see them changing strategy especially since they sold off their X86 server line.

[iOS6 user]

> I think the author is arguing that that high end market(that continues to shrink

https://finance.yahoo.com/q/is?s=ORCL+Income+Statement&annual

Oracle Income before tax in 2013 was 12,834,000. In 2014 it was 13,704,000. I Don't see where author see this shrinking?!?

In previous years was the same.

BTW existence on market bigger and bigger 3rd part ?asS providers to be hones even opens this market for Oracle because if those companies offers DBs as services high end database installations may provide lowering the costs of running huge number of customers databases on such high end HW platform like Sonoma instead running them on x86.

Try to think only about one aspect of running one consolidated DB engine on such HW like making backups. In case of Linux there is no so effective technique like like that one which is available on Solaris like flushing all not written data from memory to storage -> block access to storage for fraction of the second to make snapshot and -> unlock storage -> create off site backup using zfs send.

With massive scale databases binary backup is only solution.

ZFS is on the market more than 10 year (few months ago was 10 years anniversary). After 10 years Linus with btrfs is faaar behind ZFS. OpenZFS in longer term is less and less alternative (distance between ZFS and OpenZFS is only growing).

[Steven Jones]

I don't think evidence about Oracles revenue growth necessarily contradicts the notion that the high end market (for specialist SPARC servers) is declining as much cheaper x86 hardware continues to make inroads in that area. The market for servers using other that x86 architecture servers is increasingly a niche one.

However, I'm not sure that the author's hypothesis that Oracle's internal use of SPARC for cloud-based services is enough to continue the level of investment.

Of course there is still a market for what might be called "legacy" customers of SPARC servers, but in my experience, most would wish to move to a more cost effective platform (read x86 architecture), but are deterred by the cost and complexity of migration. If you have an existing application running of an Oracle DB on SPARC servers and need to keep a service going with minimal downtime during a migration to x86, then you are faced with a major technical and logistical challenge. Yes, there's Goldengate, but that's neither cheap to licence or simple compared to uplfiting a SPARC server.

SPARC will limp on (like many legacy platforms), but I can't see it's fortunes reviving. Not many new projects are pointed towards SPARC.

[iOS6 user]

> I don't think evidence about Oracles revenue growth necessarily contradicts the notion that the high end market

Size of this market is determined only by number of companies with enough big needs.

I don't see any evidences that number of such companies is declining.

Growing revenue such companies like Oracle definitely is not evidence here.

Do you see any other data suggesting decline trend?

[Spaller]

Sparc is dead in EDA

EDA, or Electronic Design Automation, was dominated by Sparc machines for a *very* long time. Now no one in EDA supports Sparc machines. We've even been busy at removing the cursed #ifdef's that supported it since we never expect it to come back (same for HPUX and AIX, but they were never dominant like Sparc/Solaris was totally dominant). And I won't be missing register window traps. In other words, even if this chip is a great success, it is extraordinarily unlikely that this engineering community will move back to Sparc. Oracle would have to pay to get a company like ours to provide a port of our software to Sparc now. Others still try, like IBM or HP, to pump up their application base on their architecture. But after decades of seeing these deals done, they matter not. Once you have to pay for the port, you're forever a niche player in that area.

Fabs are designed are around the hardware, and there is no fab that uses Sparc machines to run it for things like LVS, DRC, OPC, fracture, ... This was not the case before. Thousands upon thousands or cores are necessary to validate a chip for manufacture. Oracle is completely out of this manufacturing space except in the case of using Oracle DB for tracking chip defects.

So in essence, in the past ten years, Sparc can no longer be found in these engineering communities. It seems to have holed up in finance communities and some webbies.

Full disclosure: I work in this space.

[Steven Jones]

Oracle's revenues are dominated by software licencing , support and services. For years they have increased turnover by buying up companies selling enterprise applications. So yes, they have plenty of very large corporate customers, but the considerable majority of that is not in hardware sales. (Also, cloud services aren't, as yet, very high either, which rather questions the number of SPARC servers that Oracle might be able to absorb internally unless there was a vast change).

In US $ the hardware turnover is on a gentle decline. There might still be decent returns, but there's no sign of some recovery in SPARC platforms. It's just the continuation of a story from the time when SUN was the dominant player in the UNIX market. They are just losing out to x86 servers and will, no doubt, continue to do so.

http://investor.oracle.com/financial-news/financial-news-details/2015/Total-Revenue-Unchanged-at-93-Billion-but-Up-6-in-Constant-Currency/default.aspx

[iOS6 user]

> In US $ the hardware turnover is on a gentle decline. There might still be decent returns, but there's no sign of some recovery in SPARC platforms

You don't see one very important fact that in last years progress on hardware overcome the same on software area. Only by this in many cases many workloads can be handled by investing in hardware. Oracle IMO has descent or enough balance between investing in hardware and software.

I'll mention only again ZFS as very basic technology. Providing ZFS compression on enough big scale can provide huge savings making king of contradicting result that Linux which is theoretically for free can be on some exact conditions more expensive than Solaris only because to use it it will be necessary to buy much more powerful end expensive hardware.

Behind this boundary begins area where Oracle has something to offer than other companies for his customers.

People sees decline of high end market where they see that some mid range or low end solutions are entering on area handled by companies offering enterprise or high end solutions.

This what happen probably happen with EDA. At some time this was something new build by highly specialized engineers as high end solution and by this they some exact solutions been build on top high end technologies. After this EDA solutions matured and competition causes lowering cost of building those automation. What makes such solutions high end? Not to much anymore ..

Moving up lower boundary of enterprise marker is constant process .. "nihil novi sub Sole". What is less obvious is that upper boundary of this area constantly is moving up as well. Why it is less obvious? One very simple fact: number of people working for this marker is it was and still is fraction of whole IT employs.

[tom 99]

IB controller for memory scans..?

> M7 on doing SQL memory scan can do 120 GB/sec, whereas an x86 runs queries at ~5Gb/sec (in both cases it is max theoretical speed) .. and this is why in M7/Sonoma CPUs is build IB controller.

I'm sorry, but you probably miss some basic CPU architecture concepts. You do not need IB to do memory scans. In fact you don't do any I/O to access RAM.

You should compare systems carefully instead of believing what Oracle marketing says. Compare socket to socket performance or $ to $ performance. 5Gb -- did you mean 5 gigabits or gigabytes? You can scan more than 5 gigabytes /s on modern notebook.

[iOS6 user]

Re: IB controller for memory scans..?

> I'm sorry, but you probably miss some basic CPU architecture concepts.

This concept is not fixes and from first 4004 Intel CPU still is evolving.

>You do not need IB to do memory scans. In fact you don't do any I/O to access RAM.

OK. Lets say that you just rebooted the system which is running in-memory-database and this database uses 2TB or RAM (max amount of RAM per CPU socket in case Sonoma/M7).

How long will take warming up this DB over FC/SAS/SAT/10Bb NIC(s) links?

And another thing. Remember that using IB is not only about bandwidth but about lower latency of IOs.

> You should compare systems carefully instead of believing what Oracle marketing says. Compare socket to socket performance or $ to $ performance. 5Gb -- did you mean 5 gigabits or gigabytes? You can scan more than 5 gigabytes /s on modern notebook.

Really don't understand why you are assuming your laptop CPu does what are doing high end CPUs like Sonoma/M7.

[Anonymous Coward]

High end

If you want to talk about high end why not talk about actual high end: what proportion of the top500 supers use SPARC? Do you really think the people who build these things are not measuring performance (and in particular I/O performance: anyone can fill a room with cores, but that's not a supercomputer)?

[iOS6 user]

Re: High end

> what proportion of the top500 supers use SPARC?

None of the HPC installations is about working with highly demanding databases.

Top500 and HPC is about processing data quire often using OpemMPI API.

[PlinkerTind]

SPARC market is not shrinking because...

If we talk about the "actual high end, the top500 supercomputers", so no, top500 is not high end. Supercomputers are just clusters, and clusters can not run business enterprise software (as explained by SGI). The high margin lucrative market is business servers, with as a many as 16/32 sockets. For instance, one single 32-socket IBM POWER P595 server used for the old TPC-C record, costed $35 million. No typo. These large scale-up servers costs very much money, millions. Whereas a cluster is basically the cost of a bunch of nodes and a fast switch - very cheap. And a large cluster such as SGI UV2000 with 10.000 of cores and 64TB RAM - can never run business enterprise software. You need a large scale up Unix server with 16/32 sockets such as SPARC or POWER. Until a couple of months back, there did not exist larger x86 servers than 8-sockets. It is very difficult to build large 16 socket servers, x86 has tried for decades and failed. Now recently SGI released their UV300H which has 16-sockets, but I suspect performance is awful as it is the first generation 16-socket server, whereas SPARC goes up to 64-sockets today. And 64-sockets beat 16-sockets.

Regarding the shrinking Unix market. It is true that Unix market shrinks, but the Oracle engineered tailor made black boxes designed to run business software such as Oracle databases, is increasing very fast. That market is increasing whereas Unix market shrinks. The only time you need large Unix servers today is if you are going to run very very large workloads on business enterprise software, such as SAP. Check the SAP benchmarks, it is RISC all the way at the top. SPARC has top spot with 840.000 saps, whereas the best x86 server has 320.000 saps. x86 does hardly scale to larger than 8-sockets, which is nothing compared to 64-socket SPARC servers.

So if you need extreme enterprise business performance, or extreme RAS reliability, you must choose SPARC/POWER. Otherwise, x86 is fine for the low end.

BTW, the largest POWER8 server is E880 which scales up to 16-sockets and 16TB RAM. The largest SPARC is Fujitsu M10-4S with 64-sockets and 32TB RAM (soon 64TB). This year the Oracle SPARC M7 will be released with 32-sockets, 1.024 cores, 8.192 threads and 64 TB RAM. It can tackle the largest business workloads. Nobody else can, POWER8 can not, x86 can not.