Rackmount Servers: May 14th

[peterjhill]

Re: Re: G4 inside

Originally posted by Silver Dragon


The PPC processors, although they are getting warmer, run at a fraction of the power/heat that the other boys do. This means that Apple can stick the fastest G4 in a 1U box with very little concern for heat.

Not to mention that not many people care how many fans you put into a server to keep it cool. Hey maybe Apple could use that as a marketing point. Keep your server room quiet, oh even better, have a rack of servers in a library between some racks of books, with a sign on the side of the rack that says, quiet please. That would be an awesome ad 🙂

 

[mischief]

Here's a thought:

That mobo had it's airport cart way out in front, like it was made to be as close to it's faceplate-embedded antenna loop as possible.

If the case works as I think it will this puts all the antennas in an uninterupted line down one side of the rack.

I think we're gonna see OS X Server gain cluster-by-Airport capability.

I also think that the distance between the processor card slot and the pass-through cuts for standoff posts is large enough for a 4 CPU card.

 

[schalliol]

As for hardware announcements, didn't Steve pre-announce the original iMac at WWDC? Then there was a full release at MacWorld (that was awesome) with the actual product sale date one month later. So, I don't know that this is that out of character.

 

[Rower_CPU]

Re: Here's a thought:

Originally posted by mischief
That mobo had it's airport cart way out in front, like it was made to be as close to it's faceplate-embedded antenna loop as possible.

If the case works as I think it will this puts all the antennas in an uninterupted line down one side of the rack.

I think we're gonna see OS X Server gain cluster-by-Airport capability.

I also think that the distance between the processor card slot and the pass-through cuts for standoff posts is large enough for a 4 CPU card.

If that's the case, they better be moving to 802.11g for higher bandwidth...

 

[eirik]

wireless clusters?

I confess, this is a very sexy concept!!!

Now this competes with Fasternet and Gigabit Ethernet connectivity. For a simple cluster: a handful of nodes and daisy-chained (as opposed to fully-meshed), Ethernet, or Firewire for that matter, rule. The wireless concept sounds like it would kick-ass in a complex cluster of not only Mac nodes but with other devices too. That way, software would drive the complex connectivity. Software could dynamically adjust the configuration to accomodate changes in workload.

However, while this sounds great, this has two MAJOR problems. First, reliability! This is totally unproven and no one in their right mind would place mission critical resources on such a system.

Second, bandwidth, 802.11x is completely inadequate where success requires speed. Fortunately, some customers may not need speed. Even Apple's rumored wireless 100+ Mbps wireless offering sounds lacking in comparison to Gigabit Ethernet.

The good news is that higher frequencies could be used that allow for higher bandwidth. If this RF has an appreciably low amplitude, network administrators need not be concerned about BRAIN TUMORS.

I don't know, maybe money would better be spent on a super 3D switched fabric device that dynamically facilitates communication among the nodes, each plugged in via its Gigabit Ethernet.

All of this depends upon Apple's target market!!! Sadly, none of us really knows. If anyone thinks that Apple is launching an all out invasion of the enterprise market, think again. Apple doesn't have but a small fraction of the resources to do so. No, Apple has likely cherry-picked a few segments here and there for their racks. They'll establish a beach head and then grow from there.

So, which segments?

Eirik

 

[eric_n_dfw]

Wireless cluster - Kewl!

All they need now is that wireless power source and they could be completely cordless!

 

[3rdpath]

timing....

i think we'll see the ibook and the server show up together-and both using the G3 chips. that is, if what i've read is correct and most server apps don't need or use altivec.

and i don't think we'll see any new technology in the servers (DDR, Rapid I/O...).

i think its gonna be a long wait until july😱

 

[eirik]

No G3 in rack

Originally posted by 3rdpath
i think we'll see the ibook and the server show up together-and both using the G3 chips. that is, if what i've read is correct and most server apps don't need or use altivec.

and i don't think we'll see any new technology in the servers (DDR, Rapid I/O...).

i think its gonna be a long wait until july😱

While the G3 power draw is very attractive, the G3 is not fully MERSI compliant. That means it does not make a good CPU for multiprocessor configurations. I don't expect Apple to introduce a rack line without multiple processors per board. Only the low-end racks would offer single CPU's.

Eirik

 

[edenwaith]

Personally I'm excited to see that Apple will actually be making rack servers. I don't think I'd get one myself, but it is good to see Apple really trying to get its foot into the door. Yes, they are selling their G4 servers, but I don't know of many people who are using them as a server. If I was going to set up my own server, I'd probably just use my own G4 Powermac, turn on the file sharing, and dig into my UNIX side to get things up and running.

One question is what type of servers will they be? Will they be built for the low end market of servers which also can be replaced by standard home computers, or is Apple shooting more for the high end, enterprise side of the market (which MS fortunately doesn't command, heh, heh). Perhaps they will have both, a low end (cheaper) and higher end. And if they can be all networked into a large computing cluster, that would be even cooler, which then could simulate a multi-processor machine.

 

[3rdpath]

Re: No G3 in rack

Originally posted by eirik


While the G3 power draw is very attractive, the G3 is not fully MERSI compliant. That means it does not make a good CPU for multiprocessor configurations. I don't expect Apple to introduce a rack line without multiple processors per board. Only the low-end racks would offer single CPU's.

Eirik

you know much more than i do about server needs-so i bet you're right.

but i'm still skeptical we'll see any of the new DDR etc on the server. as i posted earlier, i hope i'm wrong.

one week and we'll know for sure.

 

[blakespot]

Re: you're kidding right?

Originally posted by mozez
first off, that server board on ebay was a asus tech board, you can go to their site and see it, second, apple has never made anything first or cheaper than their competitiors, they rely on it being better, never cheaper.

No, it actually was almost certainly an Apple prototype board. Also, I know ASUS boards---it is definitely not one (nor a PC board at all). PC boardmakers don't usually put an Apple-specific power extension on their AGP slots...


blakespot

 

[blakespot]

Re: Re: i wonder

Originally posted by Rocketman


One theory is his hand was forced by the ebay issue.

The "eBay issue" had nothing to do with Apple's actions at WWDC. Leaks are unfortuante, but never shape Apple's PR action.


blakespot

 

[blakespot]

Re: Here's a thought:

Originally posted by mischief

I think we're gonna see OS X Server gain cluster-by-Airport capability.

What possible advantage would wireless networking give to clustered rack servers that are inches (or less) from one another?? Gigabit ethernet is built in.


blakespot

 

[peterjhill]

Re: timing....

Originally posted by 3rdpath
i think we'll see the ibook and the server show up together-and both using the G3 chips. that is, if what i've read is correct and most server apps don't need or use altivec.

and i don't think we'll see any new technology in the servers (DDR, Rapid I/O...).

i think its gonna be a long wait until july😱

No way. Not going to happen. Servers do not care about CPU power requirements. Servers will use the fastest processors available. Maybe they will have a range of processors available to choose from. There is no reason why a server app could not be written to take advantage of Altivec.

Think about what Server apps Apple will be targeting?

• Streaming Quicktime
• WebObjects
• Oracle
• Render farms
• A Mac version of "terminal server"
• E-mail server

Just a quick list off the top of my head.

 

[3rdpath]

Re: Re: Re: i wonder

Originally posted by blakespot

The "eBay issue" had nothing to do with Apple's actions at WWDC. Leaks are unfortuante, but never shape Apple's PR action.


blakespot

with all due respect to a macrumors God- i disagree. i'm not saying it was the ONE reason, but PR is all about information and perception. apple has a big interest in controlling the perception of info-most importantly info they didn't intend to release( which may or may not be accurate).

PR must be dynamic to be effective and taking the item off ebay was a PR action. so yes, leaks do shape PR action.

respectfully.

 

[eirik]

Re: Re: Here's a thought:

Originally posted by blakespot

What possible advantage would wireless networking give to clustered rack servers that are inches (or less) from one another?? Gigabit ethernet is built in.


blakespot

I'm not bullish on wireless clustered servers myself. But imagine a wireless solution that features dozens of channels, each with over 400 Mbps of bandwidth. Let's assume for a moment that fully-meshed internetworking of each of the nodes in a system of two dozen servers is too costly, too impractical.

The servers are running a tightly integrated WebObjects solution that fully levers load balancing. In this scenario, the loadbalancing manager detects unacceptable latency due to one of the databases incurring too much workload. So, it splits this database in half and assigns another server to this second half for processing (this might require the system to move an application off of an existing server to make room for the database). Now, overall system performance would suffer if this second database server were internetworked such that all of its inbound and outbound traffic had to go through yet another server due to the connectivity constraint that I mentioned (fully-meshed connectivity impractical). So, the load-balancer reassigns the wireless channels so that the second database server it optimally placed in the topology.

What is really cool about this is that a network administrator wouldn't have to disconnect and reconnect Gigabit Ethernet cables to load-balance and optimize the topology of all of the nodes. It would be totally dynamic. Now maybe a simple Ethernet swtich board (if one exists) would be cheaper and nearly as effective (this board would be similar to boards that people employ to use a single keyboard/monitor to support multiple desktop computers). So, at the least one wouldn't have to deal with flipping switches on such a board. At the most, one wouldn't have to search, connect, and verify Ethernet cable connections.

So, if a wireless solution were this robust, then it would be attractive. The entire topology of the cluster would be completely soft-switched (software based) without any mechanical actions.

That said, I doubt we'll see a wireless solution that robust anytime soon.

Eirik

 

[eirik]

Re: Re: timing....

Originally posted by peterjhill


No way. Not going to happen. Servers do not care about CPU power requirements. Servers will use the fastest processors available. Maybe they will have a range of processors available to choose from. There is no reason why a server app could not be written to take advantage of Altivec.

Think about what Server apps Apple will be targeting?

• Streaming Quicktime
• WebObjects
• Oracle
• Render farms
• A Mac version of "terminal server"
• E-mail server

Just a quick list off the top of my head.

Peterjhill's statement, quoted above, is often true. However, for high density cluster solutions or just high density racked server solutions, power requirements become a huge factor.

This is the driver behind the latest craze in server markets whereby blade configurations are being offered.

IBM, with its relatively low-power Power4 CPU's use a fraction of the power of an x86 CPU, recently started offering a blade configuration that is turning a lot of heads in high-end enterprise applications.

When the density of CPU's gets to some thresholds, enterprises find that they have to upgrade their power systems (power grid from utility, UPS's, internal wiring, and other power infrastructure) as well as upgrade their climate control. This costs tens of thousands of dollars or even hundreds of thousands, depending upon the density. x86 CPU's run too hot for blade configurations or at least they cannot be as densely packed. This means higher volume of rackspace. That too is limited and expensive, real estate costs for example.

So, power is a major driver in server sales when the density is high. But as P... says, most server needs are not this high, so power is not a critical factor.

Eirik

 

[mischief]

Bingo!

I'm aware that 802.11 isn't wide enough for main server load. I was suggesting using it for managing the cluster not the main I/O. It would be coolest of course to use as many interconnects as the OS can handle. Say the main load goes out by Gigabit, most intra-cluster traffic goes by FW or FW2, and Airport does the admin. How much total bandwidth is that?😀 😱

 

[schalliol]

Also, keep in mind that even the G4 runs with less power and heat than x86 systems.

I would be suprised if Apple uses something other than a G4 (or perhaps G5). Even the ANS/700 units used the standard processors.

 

[peterjhill]

Re: Re: Re: Here's a thought:

Originally posted by eirik

<snip>
But imagine a wireless solution that features dozens of channels, each with over 400 Mbps of bandwidth.
<snip>

Umm...
1. All wireless implementations that I know about are the equivalent to hubs
2. Current 802.11 that Airport uses has 3 effective channels, sure you can set the channel to 11 different values, but the channels overlap frequencies, thus "dozens of channels" is moot. The bandwidth that the FCC is setting aside for this unrestricted use is too small for 400 mbits/sec. Your scenario would require something like a gigahertz of EMF bandwidth

If 100 mbits/sec switched is not fast enough, you could always get gigabit ethernet switches to connect all the servers together.

In a fantasy world would wireless clusters make sense, but in a real world, it would be silly and ineffective.

OTOH someone told me that they heard about servers that had a wired and wireless NIC, and if the wired NIC went down, you could still access the device with the wireless NIC. That is kind of cool, as long as your using SSH or SSH tunnelling to secure your connection.

 

[mcrain]

Re: Re: Re: i wonder

Originally posted by blakespot

The "eBay issue" had nothing to do with Apple's actions at WWDC. Leaks are unfortuante, but never shape Apple's PR action.

blakespot

Sorry Blakespot, but leaks of confidential information in a publicly held company DO shape PR action. If a PR department says anything that can be construed as denying the content of the leak, the company will be sued for securities fraud in the event the investor relies upon the PR department's statement, buys or sells stock, and then suffers financial loss. The 34 Act (I think) of the SEC is pretty clear on that point.

 

[peterjhill]

Re: Re: Re: timing....

Originally posted by eirik

This is the driver behind the latest craze in server markets whereby blade configurations are being offered.
Eirik

My impression of the "blade servers" is that it allowed you to put more servers into less space. For example, instead of just one server per RU, you can put 6 dell servers into 3 RUs . This would be handy for web servers of dns servers where you want to use round-robin dns to make multiple servers appear as one.

When it comes to reducing the power requirements of a server, the CPU is only a small part of the equation.
The top of the line G4 uses only 30 watts, the one they put in the new Powerbook uses something like 19 watts.
a nice scsi harddrive uses 10 watts just idling (or 13 watts here )
Then you add the memory requirements for the ram, the fans...

A 2 Gig P4 uses 43.7 watts
A 1 gig P3 runs at 29 watts

So if Dell can stick 6 servers into 3 RU's using P3s, Apple can put the same number using the most powerful G4 that Motorola has listed.

 

[percolate]

Re: Re: timing....

Originally posted by peterjhill


No way. Not going to happen. Servers do not care about CPU power requirements. Servers will use the fastest processors available. Maybe they will have a range of processors available to choose from. There is no reason why a server app could not be written to take advantage of Altivec.

Think about what Server apps Apple will be targeting?

• Streaming Quicktime
• WebObjects
• Oracle
• Render farms
• A Mac version of "terminal server"
• E-mail server

Just a quick list off the top of my head.

If you'll notice, the only comments Jobs' made about added features and such all were directed more towards the traditional file and application serving markets - improved "server" JVM, Python, Fast CGI, TCL, Ruby, disk, print and mail quotas, et al. IANAE on this but I don't think these really serve much purpose to render farm nodes, although the JVM may help controllers. Normal app and file servers will see little to no gain using G4s instead of G3s currently and most likely won't ever because vectorization doesn't fit most of the algorithms needed in most server apps. You're typically not just adding or multiplying arrays of ints or floats, you're looping through threads, connections, doing queries, serving files or trying to load balance operations.

Also, many people using servers do care about power requirements and heat just as they care about space. The less of any of them, the less problems and costs they'll have. It becomes more pronounced the larger the number of servers needed. I think they'll be releasing 1 GHz Sahara G3s, and possibly G4s at the same time or a little later on. Hopefully if the server OS costs significant money, they'll offer a G4 version with the standard OS X then, too. If you just need computation nodes paying a couple hundred extra for server software would really stick in your craw.

 

[eirik]

Re: Re: Re: timing....

Originally posted by percolate


....
I think they'll be releasing 1 GHz Sahara G3s, and possibly G4s at the same time or a little later on.
...

Is the Sarhara MERSI compliant? If not, I don't see it going into the rack. If yes, it would be very much worthwhile for highly dense servers.

Eirik

 

[gjohns01]

Re: Re: timing....

Originally posted by peterjhill


No way. Not going to happen. Servers do not care about CPU power requirements. Servers will use the fastest processors available. Maybe they will have a range of processors available to choose from. There is no reason why a server app could not be written to take advantage of Altivec.

Think about what Server apps Apple will be targeting?

• Streaming Quicktime
• WebObjects
• Oracle
• Render farms
• A Mac version of "terminal server"
• E-mail server

Just a quick list off the top of my head.

I agree with everything except the power and Oracle part. That arse Ellison hasn't ported it to OS X. Don't think he plans to. Sybase may be coming. Maybe IBM will be kind enough to port DB2. I don't think Apple will hit the high-end. Look for more of a mid-range "workgroup" server. Going high-end will put them into competition with the big boys and I don't think they are prepared to spend the kind of money those guys do on support etc.