back to article IBM lets fly single-socket Power7 Penguin server

Big Blue has launched a third Power-powered Penguin box and pitched it against x86 server rivals. The machine, called the PowerLinux 7R1, is a single-socket machine that complements the two-socket PowerLinux 7R2 rack and tower server, and the Flex System p24L half-width node for PureSystem modular servers. Back in April IBM …

COMMENTS

This topic is closed for new posts.

Not the same as Linux x86

Linux on POWER / SPARC / Itanium / IBM's Mainframe is similar to Windows 8 on ARM... not the same than x86 versions... After all, on most machines you have to install 3rd party Software and not all of them are from IBM. These 3rd party software companies won't invest on different Linux platforms other than x86 unless:

- The market share becomes relevant for their business, thus justifying the port to other processor architecture

- IBM give them some financial incentives... I remember Intel doing that to leverage Itanium, but after billions spent HP and Oracle are still fighting about its death.

0
3
Anonymous Coward

Re: Not the same as Linux x86

You are right about the ISVs. There is a different set of support at companies like Oracle, for instance, for RHEL on x86 vs. Power vs. z.... It is a shame because this Power server will likely blow the doors off of a single socket x86 server with better RAS for about the same price. I imagine this is at least predominantly targeted at custom developments on LAMP and the like.

0
0
Silver badge

Re: Not the same as Linux x86

> about the same price.

Everything else may be true but that won't be. Unless you consider 2x about the same price.

0
1
Flame

Re: Not the same as Linux x86

No, it's entirely different to Windows RT on ARM.

With the huge open source software ecosystem already available and already built across multiple architectures, you already have a hell lot available for linux on non-x86 architectures that you don't on windows.

As for closed source, well what do you need on a server that IBM don't provide? Why would you want non-IBM software?

;)

5
2
Anonymous Coward

Re: Not the same as Linux x86

"As for closed source, well what do you need on a server that IBM don't provide? Why would you want non-IBM software?"

True, IBM does have the complete stack and they are tight with the Linux community (as they are a large part of the "Linux community"). Provided you don't want to run Oracle or MS, you can run everything else. IBM's Power systems are really solid. If the price is about the same, why not.

2
0

Re: Not the same as Linux x86

Java works just fine on PowerPC (with the exception of data-racy application may blow up due to the weaker memory model compared to x86).

Plus, I bet you can shop any of the non-open-source software from IBM

0
0
Linux

Re: Not the same as Linux x86

Won't make much difference if you are running open source apps. Most FOSS apps are available across architectures.

But if you insist on running Proprietary x86 apps, you shouldn't be reading this article anyway ... go back to your ball and chain.

0
0

But an AIX port would be easily adapted to Linux...

The benefit for the user is that IF the vendor (or IBM) does an AIX port, then that port can be easily adapted to Linux-on-Power. Think in terms of Oracle-Database, WebSphere and NetBackup. There are MANY more products in this domain.

0
0

Re: But an AIX port would be easily adapted to Linux...

Oracle just canned Power on Linux support. They still support Linux on System z, it is really popular and cost effective for Oracle DB, but not Power - Linux. You are right about the rest though. WebSphere would work well. DB2 would work well. All of the open source app stuff is generally supported for Power - Linux.... Oracle is the only major hold out. I think this is going to be targeted at the big web shops.

0
1
Silver badge

pricing

Why pay 2.5 times what the same spec x86 box will cost from Dell?

"similar pricing" applies if you get a x86 Linux quote from IBM - which I did a couple of days ago. Rape and pillage on memory applies across all their platforms.

0
1
Anonymous Coward

Re: pricing

Because IBM makes way better servers in every respect. Look at the RAS features on a Power server and look at the non-existent RAS features on a Dellapart box. Not even in the same league. IBM Power has a completely different IO architecture and hypervisor which allows them to run at much higher utilization rates than Dell's commodity architecture. If you can squeeze twice as many socket licenses on a Power server, the cost of the one socket server really doesn't matter.... IBM's build quality, management software (especially on Power), support is way better.

1
1
Silver badge

Re: pricing

I don't doubt that IBM make better hardware.

The question is "Are they so much better that I can't just buy 2 Dells?" - and this is referring to near-identically specced x86 boxes, not Power ones.

Socket licenses aren't generally an issue on Linux or associated software and they're certainly not an issue with the software we're running here (scientific analysis stuff which would run well on Power)

1
0
Anonymous Coward

Re: pricing

On identically speced x86 boxes, probably not. On single socket and low end double socket, there are no major technology differentiations. They are commodity hardware, one step above PCs. It is more about IBM just being better at building things than Dell at the low end. Twice as much seems high though. IBM is usually slightly higher than Dell, but like 10-15% not twice the price. The VAR is probably trying to make some margin on you.

IBM's major differentiations on x86 come into play with their eX5 chip set servers, like x3690 (two socket rack), HX5 (two socket blade), x38/950 (four and eight socket). That is where IBM uses a completely different chip set and I/O architecture than commodity x86. They scale memory, which is usually the workload bottleneck, to 96 memory slots and 3.5 TB of memory. They have a sweet memory buffering technology that ensure there is no latency between those DIMMs that are loops away the procs. As they eliminate the bottlenecks which prevent x86 servers from going over about 20% CPU utilization, you can now scale these things to the moon. The VMware VMark record for four sockets is the x3850 with something like 365 VMs on a four socket. Like I mention, if you can double or triple the amount of workload on a socket, the servers are essentially free. The RAS features are not Unix level, but they are a helluva lot better than Dell.

0
0
Anonymous Coward

It's the software (mostly) that matters these days (citation needed for Dell/Intel comparisons)

Who remembers when DEC tried this with Alpha boxes (servers and workstations) that cost less if you only wanted the NT/Linux version and a bit more if you wanted the VMS or UNIX(OSF/1+) version? The hardware was more or less identical (though at one stage the NT/Linux ones were a different colour from the expensive ones).

Who thinks it will work better for IBM and Power than it did for DEC and Alpha? Two factors make it more promising now than it was back then: IBM have rather more clue about selling not just engineering (DEC didn't, mostly), and there's rather more Linux around now than there was back then.

I'd be interested to see pointers to sensible evidence of why IBM's IO is significantly better than Intel's. HP have been trying that game with IA64 vs x86 ever since it came in. It wasn't very credible initially and it's got less credible year by year. I haven't followed any similar Power vs x86 discussions.

0
0

IBM's Power IO vs Dell's IO...

You can certainly make Dell's IO perform similarly to IBM's Power IO. It is just that the cost-advantage to the Dell machine disappear. UNISYS had sold million-dollar Intel machines for years.

Natively, IBM's CPUs have faster busses and more of them (especially as Intel goes to high-core sockets). These busses facilitate less-expensive adaptation for high IO throughput.

1
0
Anonymous Coward

Re: IBM's Power IO vs Dell's IO...

"Natively, IBM's CPUs have faster busses and more of them (especially as Intel goes to high-core sockets). These busses facilitate less-expensive adaptation for high IO throughput."

Really? When compared with HyperTransport and its Intel knockoff, Quickpath (formerly Common System Interconnect)? References welcome.

"UNISYS had sold million-dollar Intel machines for years."

Really? Presumably ES7000/x86, and relatives, right? Yes, they're in the slideware etc, but I remember when Compaq tried to sell them, rebadged as Proliant 9000 (?). Resulting sales: negligible; product withdrawn after a very short lifetime. I also have a contact in Unisys sales, who confirms that Unisys's own sales of ES7000/x86 are limited, in part due to them not being "qualified" in a timely manner for essential x86 software (VMware etc).

"You can certainly make Dell's IO perform similarly to IBM's Power IO. It is just that the cost-advantage to the Dell machine disappear."

That's entirely plausible, but again could use some detail.

Cheers!

0
0

This post has been deleted by its author

Re: IBM's Power IO vs Dell's IO...

Soooo, let us get this right...

You are an anonymous coward who counters fact with thing you thought you hear from a friend...

Am I getting this right.

0
0
Anonymous Coward

Re: IBM's Power IO vs Dell's IO...

If you have evidence that there have been significant sales of Unisys's cellular multiprocessing stuff, with or without Compaq badge-engineering, feel free to quote it.

Meanwhile, it is demonstrably true that Compaq announced, and then relatively quickly withdrew, the Compaq-badged version of this Unisys box. It is also demonstrably true that although Unisys initially announced an IA64 flavour of this box, there have been no significant sales. It is also demonstrably true that there are significant pieces of modern x86 enterprise software stack that were (or are) not qualified on this box.

Or can you show otherwise?

Address the message, not the messenger, please. That's (imo) a major reason FOR being anonymous - focus on the content, not who's delivering it.

0
0
Bronze badge

IBM preparing for AIX retirement ?

They surely must see internal competition between same-specced machines with only different OS. One of them is rapidly becoming less competitive, which will obviously lead to drop in sales. The only component unique to these less competitive machines is AIX. Any other conclusion I might have missed?

1
1

Re: IBM preparing for AIX retirement ?

I don't know if AIX is going to be retired or not. 'Boffins' here have predicted its demise for over 10 years, though.

IBM has announced a commitment to Linux on all of its platforms. They did not accompany that with an announcement concerning retirement of zOS, zVM, zVSE, AIX or i5/OS. They appear to be sanguine about keeping all of them alongside their Linux strategy, for now...

1
0
Anonymous Coward

Re: IBM preparing for AIX retirement ?

No, AIX has never been doing so well in its history. Highest market share in Unix, highest revenues ever, etc. Now, it is true that the wider Unix market has been declining for a number of years, but AIX has gone from third place (of three) in the early 2000s to dominating the market. Most AIX's gains have been as a result of HP-UX - Itanium and Solaris - Sparc becoming uncompetitive and nearing end of life. I think that IBM realizes that, over time, AIX cannot keep growing by taking away Unix market share because eventually they are going to run out of Unix market share to take. They want to add more users to the Power chip platform which would otherwise be running x86 - Linux or Microsoft as a new gain, not as a cannibalization of AIX.

1
0
Anonymous Coward

Re: IBM preparing for AIX retirement ?

Right, it is IBM trying to capitalize on the fastest growing part of the market, Linux, with their Power and z platforms. IBM has had Linux on Power for over a decade. This isn't new, it is just a lower cost version. As their two primary Unix competitors have fallen apart, IBM's AIX sales have been booming (towards the end of a chip cycle). Unix is not going to be a growth platform in the future as Linux is becoming pretty comparable, but there will still be a market for AIX... or, rather, there will still be a market for a vendor who takes ownership of the platform from silicon through OS with pre-integrated firmware, hypervisor, utilities, patching/updates, etc. The primary problem with x86 is not that x86 servers are not enterprise ready or that Linux is not enterprise ready or that VMware is not enterprise ready, but that all three put together are not enterprise ready because they are not integrated. All of the x86 platform vendors are releasing their own patches and doing their own thing. The most common reason for an outage on x86, in my experience, is because of some small compatibility issue between some driver and VMware or some firmware update being incompatible with with some NIC. All of the independent components still work well, they just don't work well together because there is little pre-integration... understandably so, Microsoft, VMware, Oracle, Red Hat, etc all hate each other. There will always be a high end market for those that want to own the whole deal and provide a superior experience.

3
0
h3
Bronze badge

Does IBM has the transitive tech as part of powervm ?

When I tried that prior to IBM acquiring it then it was pretty good for running any old junk on anything else.

0
0

IO on Power vs Quickpath..

We are just talking about IO here, right?

Quickpath IO is a LOT like Power 6 inter-node-links, albeit without the firmware structure that facilitates VIO & other firmware that facilitates very high IO rates. Power 7 doubled the GHz of all internal busses over Power 6.

IBM publishes 40 Gbit QDR connectivity to its IO drawers, which dumb-it-down for PCI, etc.

(http://public.dhe.ibm.com/common/ssi/ecm/en/pod03032usen/POD03032USEN.PDF)

0
0
This topic is closed for new posts.

Forums