If you can't lose performance to ECC when you have enough memory, how is it that you can lose performance to ECC when you are short on memory?
Either you lose memory performance, or you don't.
From what I can tell (not an expert, just read a few wikipedia articles), you'll see a 1-2% slowdown in memory performance with ECC enabled. Negligible and probably unnoticeable, regardless of whether you have enough system memory or not.
It can be non-trivial to make it work but it fits the rest of your requirements pretty well. It's gotten more user friendly in the last few years goo.
If your network supports multicast, AG will use it, which means you don't need a central server. This mostly means R&E networks, there is very little multicast availability on the commercial internet.
I have a OpenBSD box (486 old HP box) with 2 NICs in it. It runs PF and filters everything. That's all it does. It's far better, far more flexible and far more secure then any $40 Linksys "device" you can buy. Plunk a $2000 PIX down next to it and we'll talk.
Better in terms of functionality, almost certainly.
don't forget about the other implications of what you're doing, though. How much power does that 486 use compared to a linksys? How much does that cost you per year? How much carbon is emitted to keep it running?
Obviously there are always tradeoffs; just don't forget to include all the costs.
a reasonable alternative that combines both might be to use a linksys with the replacement firmware. That might take more work, and runs linux, not openbsd. (also let's not forget to include the environmental cost of producing & eventually disposing of the linksys;)
I'm shocked that nobody has posted Camino (mac web browser). (or else, i'm shocked that the search didn't find said post;). Uses the gecko rendering engine (so usually, if a page works in firefox, it works in camino), but throws away all the other crap from Firefox.
There is something similar called, i think, Galeon, for X11.
You're definitely thinking when you suggest this; cooling problems have been the bane of just about every data center i've ever been involved with (including my current project, sigh).
In a data center of this size (4000 sq. ft), with any significant density, a portable chiller and propping the doors really won't help.
I recently toured a Department of Energy research lab with a couple smallish but high-density data centers. (10 kw per rack; computing cluster with racks full of 1U servers).
They had designed the system with a big enough battery UPS to have 2 minutes of runtime after a power outage, so that they could gracefully shut everything down. They did *not* have enough battery to keep the A/C running.
They'd recently had a consultant tell them that if they had a power event and went to battery as planned, the temperature in the room would reach 140 degrees (F) in those two minutes.
this was *not* the power room with the batteries, although something tells me that room would have gotten warm too with all that discharge.
My experience with a horribly slow RT web UI was that it was the mysql database that was too slow, not RT itself (except inasmuch as RT was probably not being real smart about what SQL queries it was running).
I ended up throwing hardware at the problem; perhaps a mysql expert could do some good.
certainly if you're running the default mysqld config, you should replace my.cnf with my-medium.cnf or my-large.cnf from the mysql distro; that might help a lot, since it will tell mysql to use more memory than the default piddly amount.
The entire Internet2 network has very limited or no alternate routing capability.
Not really true - backbone nodes are connected to two other backbone nodes, and the network is deliberately engineered to provide enough capacity to route traffic the long way around.
It is more complicated than that with the new network, which operates at layers 1 through 3; the old network was layer 3 only. Layer 3 rerouting should happen automatically; layer 1/2 outages would require manual failover, I believe.
I'm not sure what the effects of the boston fiber cut were; they may have cut off access to the layer 3 backbone to new england until the noc could cut them over to an alternate fiber path to the NYC router; however, I believe said alternate fiber path is in place on the network already. It is definitely planned.
Here's a layer 1/2 Network Map of the new network.
disclaimer: i work for internet2, but am not a network engineer, so i may have something wrong.
"720p" TV sets are almost all 1366x768 or 1024x768 resolution, meaning they have to scale everything).
why the heck is this, anyway? I've always been surprised by this. In practice, they look pretty good, but i can't believe that a true 720-line display wouldn't look better.
for 1080-line content I could imagine that the extra 48 lines might help - but the scaler's job would be a lot harder, too - 720 is 3/2 * 480, and 2/3 * 1080, making the scaler's job a lot easier.
Turning them off will save power but the fact that the adapter is rated at 12w doesn't mean they always draw that much; the simple way to be sure is with a power meter like the kill-a-watt.
That won't save you any power at all. A $20 AGP ATI Mobility card will use less power than an integrated chip, and still have better performance.
I looked around a while ago for an AGP card like that and couldn't find one - i didn't know the right search bait (I was trying things like "server AGP card"). After searching for one based with "agp ati mobility card" i've found a few items but can't tell if they are 4x/8x or 2x/4x (it matters, right? can't use one in the other?).
Ann Arbor is a great town, and has a reasonably good tech talent pool, and a major research university with a strong engineering school, CS department, and "school of information". We're essentially part of the metro detroit area, and as such, the entire economy here is going down the tubes courtesy of the auto industry (and unfortunately, the Pfizer research lab here is closing too). The place just isn't vibrant like i've experienced when I visit the coasts.
Even before the auto industry crash, we got no respect. Try finding venture capital for a tech company based in Michigan (one with a half-decent idea, even).
I love the place but I really doubt my wife and I will be here in 10 years.
I've had the same problem - duracell nimh's (just purchased a couple weeks ago) were too tight in my camera flash.
I had the brilliant idea to remove the plastic outer wrapper to make them a bit slimmer. Don't try this, it's a fine way to short out the batteries against one another. Luckily i didn't damage the flash.
Does anyone happen to have a set of recent energizer nimh's and a fairly precise measurement tool, to check the diameter for me? Sam's Club has them at an attractive price and the battery test web site mentioned above reviewed them strongly.
i'll try to check back here but i'm danpritts on yahoo if anyone happens to be able to do this. thanks!
I'm betting that power consumption on a 30kRPM would be "high". Make that "really high".
Even if the platters were essentially weightless, the motor components require power to
spin. Current 15k and 18k RPM drives gobble many watts and run hot.
I'm sure that a 30k RPM drive wouldn't be low-power. but it would be high performance, which is what he claimed this version of the device would be.
Notably, he did NOT claim that all versions of the drive would run at 30k.
Most Windows users (especially those doing anything serious) on their machines know the security landscape and have taken steps to mitigate the problem with a defense in depth strategy (including, importantly, not running as admin).
Can I have some of what you are smoking? "most windows users" wouldn't have a clue what this sentence means.
The name of the carrier will be announced once the agreement is finalized. There are various reasons for the carrier's name being kept private but the bottom line is that the carrier has asked that it not be named until the agreement is finalized.
Slashdot Mirror
If you can't lose performance to ECC when you have enough memory, how is it that you can lose performance to ECC when you are short on memory?
Either you lose memory performance, or you don't.
From what I can tell (not an expert, just read a few wikipedia articles), you'll see a 1-2% slowdown in memory performance with ECC enabled. Negligible and probably unnoticeable, regardless of whether you have enough system memory or not.
So you say "ZFS-Fuse is great".
You're the first person i've heard say that; everywhere else i see "horribly slow" and similar comments.
I take it you've actually used ZFS under FUSE on linux?
Better yet, if you don't know how to use "info", try reading the man page to figure it out. Just try.
flywheels to do what you're talking about would have to be incredibly massive.
For comparison, we have a data center with a 4 megawatt power input. The UPS system is flywheel-based.
It is designed to carry the full load for something like 15 seconds (might be 30; it's not more than a minute) while the generators get running.
There are 12 units, each of which has 2 wheels. I believe they are fully redundant, so 6 of them can provide the required 4 MW.
Each unit is about the size of two standard server racks.
They're probably a better solution than batteries for a UPS, but I don't think it's a practical energy storage mechanism for a long time.
Solaris has cachefs, which is a supported local NFS cache. Quick googling suggests AIX has something too
Linux has this: http://people.redhat.com/~dhowells/cachefs/
which appears to be actively developed although does not appear to be in the mainline kernel.
see http://www.accessgrid.org/software - this is "vic & rat" as mentioned above.
which is from the people who went on to do evo.
It can be non-trivial to make it work but it fits the rest of your requirements pretty well. It's gotten more user friendly in the last few years goo.
If your network supports multicast, AG will use it, which means you don't need a central server. This mostly means R&E networks, there is very little multicast availability on the commercial internet.
Better in terms of functionality, almost certainly.
don't forget about the other implications of what you're doing, though. How much power does that 486 use compared to a linksys? How much does that cost you per year? How much carbon is emitted to keep it running?
Obviously there are always tradeoffs; just don't forget to include all the costs.
a reasonable alternative that combines both might be to use a linksys with the replacement firmware. That might take more work, and runs linux, not openbsd. (also let's not forget to include the environmental cost of producing & eventually disposing of the linksys
I'm shocked that nobody has posted Camino (mac web browser). (or else, i'm shocked that the search didn't find said post ;). Uses the gecko rendering engine (so usually, if a page works in firefox, it works in camino), but throws away all the other crap from Firefox.
There is something similar called, i think, Galeon, for X11.
You're definitely thinking when you suggest this; cooling problems have been the bane of just about every data center i've ever been involved with (including my current project, sigh).
In a data center of this size (4000 sq. ft), with any significant density, a portable chiller and propping the doors really won't help.
I recently toured a Department of Energy research lab with a couple smallish but high-density data centers. (10 kw per rack; computing cluster with racks full of 1U servers).
They had designed the system with a big enough battery UPS to have 2 minutes of runtime after a power outage, so that they could gracefully shut everything down. They did *not* have enough battery to keep the A/C running.
They'd recently had a consultant tell them that if they had a power event and went to battery as planned, the temperature in the room would reach 140 degrees (F) in those two minutes.
this was *not* the power room with the batteries, although something tells me that room would have gotten warm too with all that discharge.
My experience with a horribly slow RT web UI was that it was the mysql database that was too slow, not RT itself (except inasmuch as RT was probably not being real smart about what SQL queries it was running).
I ended up throwing hardware at the problem; perhaps a mysql expert could do some good.
certainly if you're running the default mysqld config, you should replace my.cnf with my-medium.cnf or my-large.cnf from the mysql distro; that might help a lot, since it will tell mysql to use more memory than the default piddly amount.
As it turns out - not necessarily.
http://en.wikipedia.org/wiki/Lactation_consultant
Not really true - backbone nodes are connected to two other backbone nodes, and the network is deliberately engineered to provide enough capacity to route traffic the long way around.
It is more complicated than that with the new network, which operates at layers 1 through 3; the old network was layer 3 only. Layer 3 rerouting should happen automatically; layer 1/2 outages would require manual failover, I believe.
I'm not sure what the effects of the boston fiber cut were; they may have cut off access to the layer 3 backbone to new england until the noc could cut them over to an alternate fiber path to the NYC router; however, I believe said alternate fiber path is in place on the network already. It is definitely planned.
Here's a layer 1/2 Network Map of the new network.
disclaimer: i work for internet2, but am not a network engineer, so i may have something wrong.
why the heck is this, anyway? I've always been surprised by this. In practice, they look pretty good, but i can't believe that a true 720-line display wouldn't look better.
for 1080-line content I could imagine that the extra 48 lines might help - but the scaler's job would be a lot harder, too - 720 is 3/2 * 480, and 2/3 * 1080, making the scaler's job a lot easier.
Turning them off will save power but the fact that the adapter is rated at 12w doesn't mean they always draw that much; the simple way to be sure is with a power meter like the kill-a-watt.
I looked around a while ago for an AGP card like that and couldn't find one - i didn't know the right search bait (I was trying things like "server AGP card"). After searching for one based with "agp ati mobility card" i've found a few items but can't tell if they are 4x/8x or 2x/4x (it matters, right? can't use one in the other?).
I just moved from a setup like this with an 800Mhz C3 to an athlon. I kept finding myself wishing the C3 were faster and finally just gave up.
The thing that broke the camel's back was running slimserver.
personally, i assumed the numbers were measuring the monitors, and i was having a hard time figuring them out.
that's the truth.
I live in Ann Arbor, MI.
Ann Arbor is a great town, and has a reasonably good tech talent pool, and a major research university with a strong engineering school, CS department, and "school of information". We're essentially part of the metro detroit area, and as such, the entire economy here is going down the tubes courtesy of the auto industry (and unfortunately, the Pfizer research lab here is closing too). The place just isn't vibrant like i've experienced when I visit the coasts.
Even before the auto industry crash, we got no respect. Try finding venture capital for a tech company based in Michigan (one with a half-decent idea, even).
I love the place but I really doubt my wife and I will be here in 10 years.
samsung announced a line of LED-based DLPs at the recent CES:
http://www.bigscreen.com/journal.php?id=426
I've had the same problem - duracell nimh's (just purchased a couple weeks ago) were too tight in my camera flash.
I had the brilliant idea to remove the plastic outer wrapper to make them a bit slimmer. Don't try this, it's a fine way to short out the batteries against one another. Luckily i didn't damage the flash.
Does anyone happen to have a set of recent energizer nimh's and a fairly precise measurement tool, to check the diameter for me? Sam's Club has them at an attractive price and the battery test web site mentioned above reviewed them strongly.
i'll try to check back here but i'm danpritts on yahoo if anyone happens to be able to do this. thanks!
I'm sure that a 30k RPM drive wouldn't be low-power. but it would be high performance, which is what he claimed this version of the device would be.
Notably, he did NOT claim that all versions of the drive would run at 30k.
Can I have some of what you are smoking? "most windows users" wouldn't have a clue what this sentence means.
Saying that Internet2 is paid for by the public is just not true.
i es.cfm
Internet2 is paid for by its members (primarily universities). Many of these are public institutions but many are private.
http://members.internet2.edu/university/universit
The name of the carrier will be announced once the agreement is finalized. There are various reasons for the carrier's name being kept private but the bottom line is that the carrier has asked that it not be named until the agreement is finalized.
Disclaimer: I work for Internet2.
> Aside, is it true that Diesel fuel in the US is more expensive than regular unleaded?
yes - on a recent car trip i noticed diesel prices were maybe 10-20% more than 87 octane unleaded.