Slashdot Mirror
Oracle Releases SPARC T5 Servers; Too Late?
First time accepted submitter bobthesungeek76036 writes "On March 26th, Larry Ellison and always with fashionable haircut John Fowler announced the new line of SPARC servers from Oracle. Touted as the fastest microprocessor in the world, they put up some impressive SPEC numbers against much more expensive (and older) IBM hardware. Is the industry still interested in SPARC or is it too late for Larry to regain the server market that Sun Microsystems had many moons ago?"
El Reg has a pretty good overview of the new hardware; the T5 certainly looks interesting for highly threaded work loads (there's some massive SMT going on with 16 threads per core), but with Intel dominating for single-threaded performance and ARM-based servers becoming available squeezing them for massive multi-threading, is there really any hope in Oracle's efforts to stay in the hardware game?
In b4 APK
In a few years, we'll be able to 3D print our own processors at home with all the power and capabilities we want! Engineering? Tools? Who needs 'em! Just download the file and 3D print anything you want! It's the future! Ten out of ten Star Trek viewers agree with me! Yay!
long live raspberry pi.
a beowulf cluster.
how many bitcoins per second?
did i win the buzzword bingo today, mommy?
While the T5 may be insignificant to a huge swath of the server market, there are many sectors (financial, energy, Federal, geo, etc.) that make significant use of SPARC platforms. The T5 is a huge advance to these markets. Oracle's not really struggling to stay in the hardware game is the Reg indicates. They produce much of their x86 gear because they use it in the Exa stuff. Their SPARCs are their bread and butter hardware in terms of raw server power. They will sell them as fast as they can produce them. Their recently announced move of manufacturing facilities from Mexico to Oregon is indicative of demand. They build their Exa's in Oregon. They worked a deal with the Oregon state Gov (tax incentives) to move their server manufacturing there in order to compress the logistics lag in getting the servers for the Exa to the kitting facility. Anyway, just my two cents.
Oracle is going to need to come up with a new game to make waves with the new processor. Simply improving a processor isn't going to change the fact that what people want are low cost processors without vendor lock in.
Sun made a name for itself with interesting hardware, but that was before processing power was a commodity. There are definitely organizations that still run SPARC, and some others who need top of the line performance that will at least give it a shot, but everyone who has a brain and a little industry experience knows that you can't just "try out" the new SPARC with Oracle in charge. If you walk any distance down that road, you start paying premium prices for every little feature you want going forward.
I used to work in exclusively Sun shops, and I've dealt with Oracle for years. There's little that the hardware and their database can do that can't be replicated by x64 and something like Postgres with some thought behind your architecture. For certain, the features they do have are not cost effective against the hundreds of thousands of dollars you pay for Oracle DB licensing, and the premium you pay for SPARC hardware and support.
The performance is just a factor, they can sell if Oracle prices it right, accounting for performance-per-watt of their stack vs. the competing ones.
Sparc being an exotic arch cuts both way, you sure have more trouble with ports, OTOH hackers have to adapt their tools to penetrate those servers, in many cases it's overall a plus.
The main obstacle IMHO is that those servers come from "we are indeed evil" Oracle ;)
---- MISSING MISCELLANEOUS DATA SEGMENT --- [sigdash] trolololol
I have not seen Sparcs in years. They are so 2003. Kind of sad as we move to generic x86 but Sun really screwed up marketing these and Oracle is not helping by requiring an Oracle RDBMS license whether they are an oracle shop or not does not help. Oracle has also been happy to tell people with perfectly good Sparc Ultra I's to go fuck themselves we wont patch your systems anymore unless you pay us $$$ for your 12 year old systems you already paid for!
You can tell I do not like Oracle so consider my opinion biased. True they can multithread really well but the performance is slow and the industry has moved to clustered low cost blades to spread things out instead in such programs. The issue with threading on a single big ass server is not as big as it once was but still used in limited circumstances.
I thought the UltraSparc was legacy at this point so I am surprised.
http://saveie6.com/
Here is an analysis of the SPEC and other benchmarks - http://benchmarkingblog.wordpress.com/2013/03/26/oracles-sparc-t5-and-m5-benchmarks-lather-rinse-repeat
Unfortunately, even in (sizable) niches like telecom, the days of exclusively SPARC shops are long over.
There will be some markets that continue to use Oracle hardware for business continuity sake (Sun/Oracle has ridiculously long hardware lifecycles by industry standards). But as a mass (server) market influence, I think Oracle is done.
Java is also from Oracle, and runs really well on SPARC, and Java is still hugely popular for writing enterprise applications, so there may be some potential there for them. Oracle Database obviously runs on SPARC as well, so it makes a pretty good platform for large enterprise applications.
The posts so far are missing the point. The point is that Oracle certifies their products to run on their hardware. They have a captive audience.
We're running away from SPARC as fast as we can.
Our unix shop used to be primarily SPARC-based, but with limited IT budgets, we're able to do far more with much less money using HP blades running CentOS.
For most purposes, SPARC hardware is far too expensive and Oracle seems to be doing all they can to kill Solaris.
We still run a handfull of SPARC systems that run specialized applications and a few Solaris zones, but nearly all other services have been pushed to natively hosted Linux systems, or virtual machines running Windows or Linux.
Don't anthropomorphize computers, they don't like it.
If you think SPARC doesnt have a place in the server world, you aren't thinking deep enough. At my work we use Sun/Oracle hardware for our most important feature, file storage as their equipment is built like a tank. I've had more hardware failures in x86 hardware then I ever had with SPARC.
Also much like mainframes, there are just some workloads that have not left the stability provided by such hardware. Also when it comes to security, I've learned from contractors that they are able to virtualize non-confidential and confidential servers on the same machine since they are able to at the hardware level keep them completely separate.
The new T5 servers are outrageously priced compared to x86 based servers. They only support Solaris OS and don't have support for other key operating systems ( MS, Red Hat) or virtualization technologies (i.e. VMware). The architectural design does not simplify operations and make IT more agile. Oracle is designing systems as if its 2005.
sure that b!y the
They are bulletproof! Right now I am working on ultra 20 NIS+ servers that have been forgotten for 5 years. No hardware issues; no reboots. However I have a large oracle RAC cluster on 5220s that is slow as a dog. I'd really like to see what the T5s would do there, but for the money I'm sure we could do better with x64 hardware.
I have. It's because the bad decisions of Scott McNealy doomed Sun Microsystems to failure (IMHO). Now Sun has failed--that's why Oracle could purchase Sun at such a cheap price. It's also because of the stability of Linux (Oracle has their own distribution) and commodity (dependable and easily replaceable hardware) servers can offer the same performance at cheaper prices for upfront and support costs over Sparc and Solaris.
I learned on Linux and Solaris (x86/SPARC) when I was 15, and I'm now 32 still using both (do the math).
A saying was told me to growing up, "Use the Proper Tool for the Job" which varies person to person, BUT for me SPARC and Solaris is the right tool. I see the OpenSource community as a great community. My WHOLE stack runs on OpenSource software. I beta-test/develop MUCH of my stuff on either Linux or OSX.
But when it comes to the production OS, I'm not some blanketed Linux bigot. I'm an *NIX Admin and an Architect at heart.
Professionally I'm a CTO (I do everything from programming php / data-center / network / DBA / UNIX / security / etc.) for an internet-based start-up that runs Solaris 10 and used SPARC CoolThread hardware in production. Baffled why? For a few reasons:
When I did a cost analysis of my time & the company's money vs Intel offerings and SPARC I eventually came away with these main points.
1.) SPARC hardware is still WAY superior with remote management than any x86 POS I've ever managed. The ALOM on a SPARC and a serial cable from my Mac works EVERY time. When I worked in past shops managing thousands of Linux Dells and HP's we had nothing but issues with ILOs from the hardware and OS side. Just pure donkey shit.
When you're a start-up buying used hardware it is a great way to cut cost where investors/owners LOVE. Frankly SPARC hardware in my experience can keep on chugging where those HPs and Dells are falling apart right and left. I don't have time to be fucking with hardware when I'm running the show of a million hats.
2.) LONG-term stability with Solaris 10 and maybe Solaris 11 (still evaluating) is a necessity to me. I work for a crazy ass mad-scientist type who does EVERYTHING custom. He's worse than the scientists that I worked with back at JPL-NASA. He has software that's been running for a decade, and the software/application I write with him now he wants to work years down the road as well. That means, I don't need to worry about a yum or apt get update that blows away some part that is critical to ONLY us and I gotta figure WTF happened. The OS is a critical back-bone element where I've seen "Linux dependency hell" fuck me so many times and cost me so many hours, that I PREFER building my own Solaris 10 packages and Solaris 11 (still in testing for me) packages (Yes, I'm a REAL UNIX admin no these lazy wanders) without worries that the OS will be compromised by something lame. In the long-run I have more freedom to enjoy time with my doggies.
When you work for a company that builds custom crap that. Everything it talks to regarding the OS needs to work without question. I have always have had that with Solaris SPARC and with Support till 2018 or extended 2021 by then I should be retired from the gig! But I KNOW nothing funky will happen with the OS while I'm working here. For each new x86 hardware update for Linux, it's a whole new 'testing' to make sure it doesn't blow up the OS on the next reboot. Never had that with SPARC of maintained properly.
With that long-term support and marriage to the hardware I know the relationship is TIGHT, that can be VERY useful when you're concerned with down the road support or integration. Dell or HP does a hardware update and the RedHat or Debian kernel or images haven't been added, then you gotta do a post image. FUCK THAT NIGHTMARE! SPARC WORKS end of story.
3.) Threads! NOTHING compares to SPARC when it comes to multiple threads and what not. My T2000's running 32 cores make damn good web-servers. They also save space in the rack as well!
4.) Virtualization is WAY superior than KVM or VMWare. I've used many of the OpenSource VM solutions and frankly non compare to the control that I can do with either LDOMs or Solaris Containers/zones.
5.) ZFS yeah, Linux we hear your promises of a bad-ass filesystem, I'm still waiting.
So, is Oracle and SPARC dead? Popularity may go down, that's normal, but it's not "dead" to anyone who has a reason/purpose to use the OS/hardware offered.
The world isn't one big LAMP stack.
Again, I'm not *against* Linux, I use it for development and personal shit all day. However, I'm not a blind follower either.
Frosty piss or not, put them tiggaflops to bouncin' on some PDRK headshots and nano-powered super leaps.
How does this even make sense? Are we supposed to believe every core replicates every functional block 16 times? Or are we supposed to believe each of the 16 threads just so happens to only need a different 1/16th of the usual functional blocks found within a core?
Of course not. Of course not. We are talking server software functions re-purposed as hardware hacks in the core for speed boosts in incredibly limited circumstances. The PC or Android tablet you are reading this on also has 16+ threads per core, if by that you mean how many currently active threads will eventually run on the same core, as the scheduler gives each in turn some 'run-time'.
The problem with Oracle's horrible hard hacks in SPARC's CPU is that such hacks only offer a possible advantage while the apps are coded in highly specific ways. SPARC defeats the whole purpose of a general programmable CPU. If someone comes up with new data structures and algorithms for server side code bases, SPARC becomes a most unsuitable platform.
The solution to the 'inefficiency' of thread switching is NOT more hardware hacks, but LESS thread switching. For a single app to create masses of instances via masses of threads breaks most general CPU architectures, because of how the memory sub-systems work. The answer is to 'abstract' the threads in the code itself, rather than issue real threads to the OS. So that "16 threads per core" should be recoded as "1 thread per core" with "16 connections per thread".
To me database is just a resource, so I may be looking at this wrong, but it seems that the Oracle server is just an appliance, and it doesn't matter to me what architecture it runs on as long as it performs well. So buying an Oracle server on Sparc isn't really a matter of "it's not x86" or "it's not Power". Since I'm not going to use the box for anything except Oracle, the cpu architecture is immaterial. Not even the OS is important, as long as it works well.
Oliver's law of assumed responsibility: If you're seen fixing it, you will be blamed for breaking it.
Having PO'ed every technophile in the world, Oracle actually expects anyone with a brain to recommend them as a path to go down? I recommend all of my clients move away from them at warp speed, and embrace other - faster, better - technologies. Oracle has managed to destroy everything it touches - I see no reason to expect they will do anything difference on this.
To recap, the T5 chip has 16KB of L1 data cache, 16KB of L1 instruction cache, and 128KB of L2 cache for each of its sixteen cores, plus an 8MB L3 cache that all of the cores share.
8MB L3 cache for 16 cores?!
Are they kidding??
IMO: DOA.
All hope abandon ye who enter here.
Not many people noticed this:
Oracle charges a big fat $0 dollars for their Linux port, including free updates.
Oracle was kind enough to drop free updates from Solaris, drop Ultrasparc support, drop OpenSolaris, and by extension drop the userbase. Why pay when a sibling product is free?
Oracle intended to drive customers out the door. What other conclusion can there be?
I've worked with T1 and T2's they are impressive hardware. But that's 4 threads and 8 threads on 8 cores, respectively, so a T5 that supports 16 threads was a foreseeable development. But AMD sells 16 core cpu's for about $500/cpu, and you can get them in 4 way configurations that support up to 512GB of RAM and they are impressively energy efficient. Physical cores should beat threads, especially when we are comparing RISC to CISC, and the AMD's are running at 2-3 Ghz.
Why the Opteron is not eating everyone's lunch is a mystery to me. I guess you don't get fired for buying Xeon.
Apparently Apple seems to be the exception to your rule, being all about sexy, and not of actual technological prowess.
This is the fastest CPU? I thought that the POWER7 was the fastest, by a mile.
I hope so.. T4 systems suck. We have problems all the time with the LDOMs spinning up and taking out the entire frame. Also, isn't it about time Solaris died?
i haven't seen such a well constructed troll post in years. Seriously, this is good shit.
--The world isn't one big LAMP stack. HAHAHAHAHAHAHA
Every 15 year old thinks he knows everything. At 32 you should at least realize you don't know 1/2 the shit you SHOULD know.
but you gave yourself away with -and a serial cable from my Mac works EVERY time.
Still, another 10 years or so you are probably gonna be pretty good.
Fujitsu owns the SPARC64 architecture, so in the end, if Oracle drops it or diminishes, Fujitsu would be left as the default owner of the SPARC architecture. They have too much invested in SPARC to abandon it. Although they could go for Itanium, POWER or MIPS if they are for any reason forced to abandon SPARC.
But I don't see why that will be necessary. With current process nodes and technology, Fujitsu can simply take the latest and greatest CPU they have, make a multi core design out of it, and use it in whatever computers they want. That way, they preserve compatibility as well. Multiprocessing used to be a bad idea when applications weren't well threaded or didn't run in separate processes, but that's been a thing of the past for a while now. So if Fujitsu just takes the latest T5 or its own SPARC64, shrinks and replicates the core and runs w/ it, it would be a simple logical step forward.
One question - does Fujitsu use Solaris, Linux or some homegrown Unix on their SPARC64 computers?
Funny thing is, just a couple days ago I was reading a presentation about the obscene amounts of money MorganStanley was saving by switching their OpenAFS cluster from Linux over to Solaris on SPARC.
The advantages are mostly around ZFS, plus dedupe and compression, but they throw in DTrace as well, as having helped track down performance issues with OpenAFS.
http://www.ukoug.org/what-we-offer/library/openafs-on-solaris-11-x86-robert-milkowski/
This is decidedly one place where Linux is lagging massively behind. Linux could eventually catch up, but only if Oracle sits on their ass and doesn't keep improving things (which I expect they wont). FreeBSD seems like the only alternative, but ZFS on FreeBSD lags behind in features, performance, and stability, so Solaris could stay king if Oracle kept throwing money into it (which they won't). FreeBSD also doesn't have the benefit of being built for only one vendor's hardware, so it can have just one set of drivers built-in and heavily tested and documented.
The Oracle/HP lawsuit left both sides hurting. HP may have a âoeMilkâ the Customer Business Strategy but I'd argue ALL proprietary Unix system vendors, including Oracle, have been doing EXACTLY the same thing ever since AMD's Opteron came out, and brought the end in sight. HP, Oracle, and IBM will continue squeezing the last drops of life out of whatever hardware improvements were already in the pipeline, then just act as rent-seekers, offering maintenance on all those legacy boxes, until the economics makes migrating all those legacy applications to Linux, unavoidable.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
I work in what used to be a heavy Sun/Solaris enterprise. We still use a lot of legacy hardware (up through the T2 processors) but pretty much stopped buying when Oracle took Sun over. We also use a lot of Oracle database software. Oracle's has given up on the lower end and are exclusively pitched at the top end of the market. Sun had a lot of problems with their identity over time but they always understood that they needed to create an on-ramp for their brand by supporting the low end. Oracle's systematically destroyed Sun's market ecology, in my opinion, quite deliberately. As usual Sun's hardware stands head and shoulders above anything else in raw performance and elegance. (As do Oracle database products.) But it doesn't matter unless you need the absolutely highest level of performance. And that's a vanishingly small percentage of the market. My management is interested in elasticity and low capital costs now. The cloud and Amazon Web Services seem to be where Sun used to be in mindshare. Oracle might pull it off, but I've seen absolutely no inkling that Larry Ellison cares about anything other than extracting as much margin as he can from the absolute highest paying customers who share his margins. So, probably the T5 and M5 are too late to matter anymore except for a tiny market that needs 8 cores and 32 threads per CPU and has money to burn on very large systems and support contracts. They'd be pretty much the last vendor I'd consider these days for general computing.
The SPARC consortium has its members, but other than Fujitsu, who makes SPARC computers any more? I do think that any company that does ought to go w/ FBSD, OBSD or Linux, so that their fortunes are not tied to anything Oracle does.
Open Solaris, in terms of this discussion about the SPARC, is indeed dead. OpenIndiana does not have any SPARC support, while a new fork from Open Solaris - Schillix - which does plan to support SPARCs, has just gotten started. So better not go the OpenSolaris/OpenIndiana/Schillix route at all if you have a SPARC server. Instead, load it up w/ FBSD if ZFS is important to you or OBSD or, if one prefers Linux, Debian, and take things from there.
1.) SPARC hardware is still WAY superior with remote management than any x86 POS I've ever managed. The ALOM on a SPARC
You're using hardware that old? My understanding (which could be flawed) is that the SPARC systems have used ILOM for a number of generations now.
and a serial cable from my Mac works EVERY time. When I worked in past shops managing thousands of Linux Dells and HP's we had nothing but issues with ILOs from the hardware and OS side. Just pure donkey shit.
I here you there. Modulo the stupid pause-on-Break behavior that went away quite a while ago, Sun has mostly understood remote management via serial console for years. This is something that Dell, IBM, and even Cisco people that I've talked to just don't understand. HP recently has shown some understanding and improvement: iLO 2 sucked hard, iLO 3 was a little better, but now iLO 4 version 1.10+ works reasonably well for initial setup via serial console and subsequently the HTTP interface once one figures out how to configure IP, but before that I saw all sorts of crap, and entering the curses-style ORCA even today from the console is nearly impossible. I still haven't figured out a way to get a real host console when SSHing to iLO, but the serial connection mostly works now, once we found an adapter that works on their anachronistic DB9 connector. I demo'd an IBM system last year, and even IBM's techs couldn't give me a way to make the serial console work.
When you're a start-up buying used hardware it is a great way to cut cost where investors/owners LOVE. Frankly SPARC hardware in my experience can keep on chugging where those HPs and Dells are falling apart right and left. I don't have time to be fucking with hardware when I'm running the show of a million hats.
FWIW HP hardware has been reliable for us so far. We were stuck with a number of G6 systems that are mostly gone, a few G7's and quite a few G8's now. My experience has been that the G8 systems are massively better than what preceded them. The DL580 G7 on the other hand, I have little confidence in.
For each new x86 hardware update for Linux, it's a whole new 'testing' to make sure it doesn't blow up the OS on the next reboot. Never had that with SPARC of maintained properly.
I've seen a couple of Solaris patch bundles over the years that could render a system unbootable (modulo netbooting or an alternate LU environment), and in the last year a couple of stupid RHEL/Linux kernel bugs that also could render a system unbootable (no LU or way to revert changes, only a netboot rescue environment)
4.) Virtualization is WAY superior than KVM or VMWare. I've used many of the OpenSource VM solutions and frankly non compare to the control that I can do with either LDOMs or Solaris Containers/zones.
I haven't personally played with Linux virtualization, but my experience with zones was disappointing.
5.) ZFS yeah, Linux we hear your promises of a bad-ass filesystem, I'm still waiting.
Damn straight. Btrfs sure isn't it. I set one up recently to test it, and found that for the thing to mount at boot-time I had to enumerate all of the component disks in the friggin' fstab, and then enabling compression resulted in space utilization *inflating* by 2x, compared to 20x compression that ZFS gets on the same data.
The world isn't one big LAMP stack.
Agreed.
Calling the T5 "the world's fastest microprocessor" is akin to saying 1000 lawnmowers is "the world's largest engine." Oracle has optimized SPARC for a 20th Century workload. Time to move on Larry, the rest of the world has.
Organization? You must be joking..