Or they might run 20 or 30 or 50 of those Windows copies on that 4RU 4600. Or bunches of copies on their blades. Or fewer on !RU boxes. Depends on what you need. IMO a reasonable bunch of choices.
I'm glad we're on same page for definitions! I agree world is going parallel - since you can only push single CPU performance so far and we're already getting diminishing returns. I disagree that there will be fine grain parallelism at the operation level. That's been tried over and over since 360/91 and the pipelined, OOO machines since then, and that doesn't scale. Operand dependancies and branch prediction mean you really can't go that way as far as we need, and makes processors really complex. Look how IBM had to make POWER6 not have the OOO they highlighted in POWER5. No, parallelism at program unit level is the way to go for most workloads. But, it's worth you opening a blog on if you got argument to that says otherwise - go for it.
Above guy is confused partly because Sun reused the word "thread" to mean "hardware thread" when most people think of "software thread" like in Java or pthread_create().
Hardware threads are virtual CPUs sharing resources on a core so work can proceed when a thread stalls on cache miss - the hardware switches to a new thread in a SINGLE CLOCK. Not expensive. Time-slicing between many software threads on a few CPUs can be expensive but having many hardware threads to run those threads makes the problem tiny. In fact, this design is great way to make multithreaded applications run real fast.
You're right. This means that people will pick between one and the other based on whether app SMP scales or is single thread, and Sun will sell both alongside SPARC.
What a bargain, you dont pay extra to "upgrade" to the same amount of capacity you already have! (rolls eyes)...
So, for $95K you get a "Linux processor" that is crippled to not run z/OS, runs Linux slowly, and doesnt include DASD or software license costs, and is outperformed by $5K Xeon or AMD from Dell, HP, Sun, or IBM. Sign me up!
Like Dr Memory sez it's a 2002 box, not that old when you think how long MF boxes take to pay off. 2066-002 is ~350 according to Isham Research. Replacement box is still a few hundred thousand bucks. Hercules can emulate in software up to 80 MIPS on an AMD or Xeon laptop for maybe $1K. Think about it. Also think about why IBM has no specweb, speccpu, tpc or any other mainframe benchmark
"In all VM approaches the idea that one can freeze an entire system and look at it, or isolate it, or migrate it, is a very valuable one. It's done well for IBM on their mainframes."
Mostly right, but mainframe VM doesn't have the ability to migrate a virtual machine. Nor does it have a balloon technique to manage pressure on memory, as VMware does. Rumor is that is being worked on
A lot more heavy weight to move an entire virtual machine with full OS image from one box to another, then transfer IP address and SAN disk ownerships. VMware does it and so does Xen, but it's not a light thing to do. Moving a single app in its JVM is much smaller effort, much less disruptive
25 of those 30 years IBM tried to kill off their hypervisor, now they have 5 unrelated virtualization products they call under same name. They hope nobody notices that while they laid off their VM guys VMware and Xen came in and did things IBM never thought of like VMotion
Yes, will be faster Dell ('cos AMD faster than Intel) use less power + generate less heat (which can cost more than what you paid to buy it) plus use less space (1RU vs. 4RU). Have you paid real estate prices in NYC recently? Lotsa places have to run Intel racks half-populated because of the heat. Very expensive. Oh yeah, numbers I saw also said cheaper on purchase price
They should have said something about SPARC to prevent the guessing that SPARC is dead. Nope, it aint, but I expect they didnt want to say anything about it when doing this AMD launch.
Easy to explain - it's a hack job just as credible as the TCO reports MS pays for, except this time its IBM.
1. Report says hardware is more expensive on Solaris than Intel, even though Solaris runs on Intel and AMD too, so the price for HW should be == Linux. Runs on the same boxen. Even if we're talking SPARC I don't believe it, cos no numbers are provided!!
2. SW costs should be higher for supported hosts cos RH costs more than Solaris. Both are freee if you download them but companies dont run that way. RH and SuSE costs are way higher per box and per CPU for s/w and support.
3. Different people maintaining the OS and the JVM? At different costs? WTF is up with that...
4. Lets say that we're talking Solaris SPARC vs Linux Intel. Then the report is bogus again because it doesn't count the power and heat difference. SPARC hasnt kept up with Mhz like Pentium but it uses a lot less power and generates less heat. I see people partially populating racks on Intel so they have to pay for more real estate on Intel than SPARC too. Add 'em together: You gotta spend more bucks powering, cooling and racking 2-CPU Xeons than 2 or 4 CPU SPARC.(Lets see 4cpu DL580 ~1300W, 4cpu Sun 440 ~650W) Plus if I have a vert scaling app I can put in more compute power in less space and fewer network drops which also costs $.
What kinda baloney to have a "TCO study" that includes only purchase costs and no operational costs. Nah, it's just a hack job. Eye-Bee-Emm doesnt mind if it nicks Windows this is really just a dig at Sun. Linux has lots of good stuff but jobs like this are just using it as a club and could wind up reducing what people believe about it.
Baloney. What part of their software have they provided to the community? Sure, they'll fund open source projects (so does Sun and lots of others) but what of their own stuff do they give away. JFS? Big whoop. What about the stuff they care about: AIX, zOS, DB2, WebSphere, CICS, blah blah blah.
Even their paying CUSTOMERS cant get source to those products!
Yup that's true. Naughty Linux for stealing the id in the first place! Speaking of partitions it is confusing that Solaris has its "slices" within a single partition that fdisk sees. Not bad or dumb, just different.
I put it on 2 laptops. No problem with ethernet. X is fine now that its on Xorg. what release are you trying?
IIRC, Solaris had partition id 82 before Linux grabbed it without asking "may I". Solaris 10 now uses partition id bf because of that.
IBM hasn't dropped DB2 for Solaris - they just haven't done it (?) for Solaris on peecee. They still do lots of it on SPARC.
Oracle already screwing Sun and their cost/core really hurts total purchase price. This way Sun can say 'you want a DB app for free on our OS - here ya go'
Nope not in the initial release. Should come out in an update. Spring? Summer? Veritas is pricey alright - but some of its features already are in Solaris with UFS even before ZFS. Take a look and see if you really need it anymore
Re:Solaris 10 on Sun Ultra 5/Ultra 10 questions
on
Solaris 10 Released
·
· Score: 1
I got it on 300Mhz Ulta 2 with almost a gig of RAM, works great. Ditto on Ultra 5 with 384MB RAM. No problem. Go buy cheap RAM on eBay - problem solved
Or, put Solaris on a spare peecee Pentium III or later. Same rules: 256MB is okay, more is better.
Why would you pay a large sum of money for Solaris when it's a free download? Just go get the bits.
One difference between Solaris and AIX+HPUX: Solaris will run on your PC (most likely - really! I've seen it on Dells, Gateways, Toshibas, Fujitsu, Compaqs, Sony), while AIX or HPUX you gotta go buy an IBM or HP workstation for lotsa bucks.
Maintenance and support are the big wins-no fixing up machines that a user screwed up. It is a LOT less admin hassle to support a passle of users.
Also: next version of Sun Ray server runs on Linux, so you can run it off a standard peecee server and drive the Sun Rays from that. There are pluses and minues of this compared to doing on Solaris on SPARC, but without getting into the whole thing: it opens up access to Wine for those who haven't kicked the Office habit
Well I sure as heck have, and even breaks going from say RH 7.0 to 7.1 to 7.3 to 8 to 9! Let alone cross distro. Yes, recompiling from scratch usually gets you there, or changing environment variables to use old threading model, or installing 800 prereq RPMs that you don't already have, or removing some RPMs you have that break the code you want or.... Sure, no problem! Seriously - it IS a problem!
Slashdot Mirror
no. gotta have a z9 or z10. it wont work on a z800 or z900 either.
this tells you all you need to know about the whole project
Or they might run 20 or 30 or 50 of those Windows copies on that 4RU 4600. Or bunches of copies on their blades. Or fewer on !RU boxes. Depends on what you need. IMO a reasonable bunch of choices.
I'm glad we're on same page for definitions! I agree world is going parallel - since you can only push single CPU performance so far and we're already getting diminishing returns. I disagree that there will be fine grain parallelism at the operation level. That's been tried over and over since 360/91 and the pipelined, OOO machines since then, and that doesn't scale. Operand dependancies and branch prediction mean you really can't go that way as far as we need, and makes processors really complex. Look how IBM had to make POWER6 not have the OOO they highlighted in POWER5. No, parallelism at program unit level is the way to go for most workloads. But, it's worth you opening a blog on if you got argument to that says otherwise - go for it.
Hardware threads are virtual CPUs sharing resources on a core so work can proceed when a thread stalls on cache miss - the hardware switches to a new thread in a SINGLE CLOCK. Not expensive. Time-slicing between many software threads on a few CPUs can be expensive but having many hardware threads to run those threads makes the problem tiny. In fact, this design is great way to make multithreaded applications run real fast.
You're right. This means that people will pick between one and the other based on whether app SMP scales or is single thread, and Sun will sell both alongside SPARC.
Sure there is, called Chime, and on opensolaris.org. harder? compare setting up logical volumes on Linux compared to much easier ZFS on Solaris
What a bargain, you dont pay extra to "upgrade" to the same amount of capacity you already have! (rolls eyes)...
So, for $95K you get a "Linux processor" that is crippled to not run z/OS, runs Linux slowly, and doesnt include DASD or software license costs, and is outperformed by $5K Xeon or AMD from Dell, HP, Sun, or IBM. Sign me up!
Like Dr Memory sez it's a 2002 box, not that old when you think how long MF boxes take to pay off. 2066-002 is ~350 according to Isham Research. Replacement box is still a few hundred thousand bucks. Hercules can emulate in software up to 80 MIPS on an AMD or Xeon laptop for maybe $1K. Think about it. Also think about why IBM has no specweb, speccpu, tpc or any other mainframe benchmark
Mostly right, but mainframe VM doesn't have the ability to migrate a virtual machine. Nor does it have a balloon technique to manage pressure on memory, as VMware does. Rumor is that is being worked on
A lot more heavy weight to move an entire virtual machine with full OS image from one box to another, then transfer IP address and SAN disk ownerships. VMware does it and so does Xen, but it's not a light thing to do. Moving a single app in its JVM is much smaller effort, much less disruptive
25 of those 30 years IBM tried to kill off their hypervisor, now they have 5 unrelated virtualization products they call under same name. They hope nobody notices that while they laid off their VM guys VMware and Xen came in and did things IBM never thought of like VMotion
Uh, Chaffar - the parent post you object was a "joke". Geddit now?
They should have said something about SPARC to prevent the guessing that SPARC is dead. Nope, it aint, but I expect they didnt want to say anything about it when doing this AMD launch.
1. Report says hardware is more expensive on Solaris than Intel, even though Solaris runs on Intel and AMD too, so the price for HW should be == Linux. Runs on the same boxen. Even if we're talking SPARC I don't believe it, cos no numbers are provided!!
2. SW costs should be higher for supported hosts cos RH costs more than Solaris. Both are freee if you download them but companies dont run that way. RH and SuSE costs are way higher per box and per CPU for s/w and support.
3. Different people maintaining the OS and the JVM? At different costs? WTF is up with that...
4. Lets say that we're talking Solaris SPARC vs Linux Intel. Then the report is bogus again because it doesn't count the power and heat difference. SPARC hasnt kept up with Mhz like Pentium but it uses a lot less power and generates less heat. I see people partially populating racks on Intel so they have to pay for more real estate on Intel than SPARC too. Add 'em together: You gotta spend more bucks powering, cooling and racking 2-CPU Xeons than 2 or 4 CPU SPARC.(Lets see 4cpu DL580 ~1300W, 4cpu Sun 440 ~650W) Plus if I have a vert scaling app I can put in more compute power in less space and fewer network drops which also costs $.
What kinda baloney to have a "TCO study" that includes only purchase costs and no operational costs. Nah, it's just a hack job. Eye-Bee-Emm doesnt mind if it nicks Windows this is really just a dig at Sun. Linux has lots of good stuff but jobs like this are just using it as a club and could wind up reducing what people believe about it.
Even their paying CUSTOMERS cant get source to those products!
Yup that's true. Naughty Linux for stealing the id in the first place! Speaking of partitions it is confusing that Solaris has its "slices" within a single partition that fdisk sees. Not bad or dumb, just different.
I put it on 2 laptops. No problem with ethernet. X is fine now that its on Xorg. what release are you trying?
IIRC, Solaris had partition id 82 before Linux grabbed it without asking "may I". Solaris 10 now uses partition id bf because of that.
Sorry you're going. V40z is fast like a scalded lizard. Only prob is that the thing is LOUD. Not good for desktop box...
IBM hasn't dropped DB2 for Solaris - they just haven't done it (?) for Solaris on peecee. They still do lots of it on SPARC. Oracle already screwing Sun and their cost/core really hurts total purchase price. This way Sun can say 'you want a DB app for free on our OS - here ya go'
Nope not in the initial release. Should come out in an update. Spring? Summer? Veritas is pricey alright - but some of its features already are in Solaris with UFS even before ZFS. Take a look and see if you really need it anymore
I got it on 300Mhz Ulta 2 with almost a gig of RAM, works great. Ditto on Ultra 5 with 384MB RAM. No problem. Go buy cheap RAM on eBay - problem solved Or, put Solaris on a spare peecee Pentium III or later. Same rules: 256MB is okay, more is better.
Why would you pay a large sum of money for Solaris when it's a free download? Just go get the bits. One difference between Solaris and AIX+HPUX: Solaris will run on your PC (most likely - really! I've seen it on Dells, Gateways, Toshibas, Fujitsu, Compaqs, Sony), while AIX or HPUX you gotta go buy an IBM or HP workstation for lotsa bucks.
Maintenance and support are the big wins-no fixing up machines that a user screwed up. It is a LOT less admin hassle to support a passle of users. Also: next version of Sun Ray server runs on Linux, so you can run it off a standard peecee server and drive the Sun Rays from that. There are pluses and minues of this compared to doing on Solaris on SPARC, but without getting into the whole thing: it opens up access to Wine for those who haven't kicked the Office habit
Well I sure as heck have, and even breaks going from say RH 7.0 to 7.1 to 7.3 to 8 to 9! Let alone cross distro. Yes, recompiling from scratch usually gets you there, or changing environment variables to use old threading model, or installing 800 prereq RPMs that you don't already have, or removing some RPMs you have that break the code you want or.... Sure, no problem! Seriously - it IS a problem!