Is energy efficiency that important in most use cases? Best I can tell, copper vs. SFP chassis differ by about 60 watts.for a 24 port model.
SFP+ may be cheaper but the cabling sure isn't and it makes it more expensive yet if you're needing to blend in a handful of 1G devices to the same stack.
If "in a building" translates to "a lot of space" then nickel-iron batteries would make the most sense, since they have a very long lifespan and tolerate deep discharges. But I think their max discharge rate is lower than others, meaning you'd need a larger space for the array.
I think lead acid would suffer from similar issues due to the need to keep maximum discharge depth above 50%.
Lithium batteries have a good discharge profile and take less space.
I'm not sure an idealized 20th century version of hunter-gatherers -- the usual pastiche of peaceful, family-centric, eco-friendly tribes -- is really any more accurate or less stereotypical than Hobbes' view.
Some places maybe had more stable relationships, some maybe had less, but I think it's never really accurate to view them as universally peaceful, free of violent conflict or capable of pretty horrifying abuses. It's not like any of these kinds of groups didn't have weapons, warfare or domineering power relationships until Westerners showed up.
"during the time men live without a common power to keep them all in awe, they are in that condition which is called warre; and such a warre as is of every man against every man"
To seriously saturate a 1gig link, you have to get into 5 figures. At that point, the price of 10gig-e is not a big deal.
Not a big deal until the customer starts comparing quotes. The uptick on the SAN controller for 10 gig ports, the uptick on any servers for 10 gig ports, and then the huge uptick on the 10 gig Ethernet switching.
The SAN controller price increase is of course way beyond the price increase of a generic two-port Intel 10G card, as is the OEM price increase on the server ports. The switching is just breathtaking -- a Dell N series 2024 is $1500, an N4024 is $7500. So for bog-standard cluster with 2x24 port switches, the switching alone is a price increase of $13K and the total project price increase is probably $15k once you factor in controller NICs, host NICs, etc.
So 10 gig can ramp up your pricing by as much as 15%-25% depending on what stuff is baked into it.
I think anybody looking at storage upgrades or network upgrades should definitely eat the increase versus choking on 1 gig for disk transfers, but those customers are awfully stingy with their money.
And it really doesn't take much in terms of fancy disk setups to saturate 1 gig ethernet. More than a pair of SSDs can do it easily and even 7200 SATA arrays of 6 disks have no problem generating 250 mbyte/sec -- that's probably 4 gbit/sec on the wire. Now, I'd agree that most common workloads don't generate that kind of regular traffic but some do and backup is quite often a punishing workload even when it's done right.
10G is great for iSCSI storage and even all-hard disk arrays can overwhelm 1G links pretty easily. Multipathing helps, but like NIC teaming, doesn't result in anything like linear increases in throughput for additional paths.
A fair number of the storage installs I've done lately have been flash cached or tiered arrays and with 10G links, I've seen sequential throughput hit 600 MByte/sec.
Even if a given box can't completely saturate a 10G link, the gap between 1G and 10G is pretty wide -- I'd take 2-3G effective throughput over 1G.
Yeah, but that's one module type that tends to be a little unusual from a real-world usage scenario. Usually people who want 10G-T buy a switching platform that provides those ports built into the chassis or blade. People that buy chassis/blades with all SFP+ ports tend to use them as a fiber landing point or for terminating devices with SFP+ cabling.
The only place I see SFP-copper being used are on maxed-out blades or chassis where the only ports left are SFP and these tend to be 1G platforms anyway.
The 10G installs I've done have tended to be either/or -- all copper with a fiber uplink, or all SFP with a fiber uplink. I've seen one site with a pair of all-SFP Ciscos with a couple of 1G copper SFP modules.
I'm still not entirely convinced that the increased sophistication of 10G *still* accounts for a price differential of nearly 500%. I small profit-taking for obvious business use customers.
Maybe I'm crazy, but it seemed like there was a time when pure access points got strangely expensive and "routers" were cheap.
So I just bought a router and used one of the LAN ports as its uplink, avoiding the "WAN" port and all the routing functionality together. I already have a firewall/router elsewhere.
The only downside I've seen of doing this is some of the devices I've used seem to have some of their ancillary functionality, like NTP, hardcoded to only use the WAN port for outgoing traffic.
How long are they going to keep charging such ridiculous prices for 10 gig networking?
10 gig copper has been out now longer than it took 1 gig copper to go from being "ooh, enterprise" expensive to being in every $499 laptop you could find. Yet they've managed to prop up 10 gig switch and NIC prices forever.
Are 10 gig parts that complicated that they're staying so expensive for so long?
Or are we waiting for the next big "ooh, enterprise" speed bump to come along?
Wasn't Playboy always kind of going out of business once Penthouse and Hustler came around and starting showing women with their legs spread? After that Playboy became kind of irrelevant as pornography.
I wonder if there's a certain amount of buyer's remorse -- people who thought "I really only drive 70 miles a day, this $electric_car has a range of 120 miles, and I can charge it at home easily".
Then they discover well, they actually do drive past the range frequently, don't get the range they thought they would, or if they did take those things into account, thought that "all those free charging points" would make up for it.
Now they're pissed -- they made a big commitment to an electric car, combined with a lot of self-righteous praise of their choice to everyone who would listen, and they're finding out it that it's not panning out like they thought it would.
Now they have a whole new set of hassles to deal with on a daily basis because their car doesn't have the range they need and recharging on the go has turned into a Mad Max situation.
Yeah, and they have been merging CML and EQL now for 2-3 years with little to show for it. CML is still a great tiering system but overly complicated for just farming out block storage to hypervisors.
IMHO, CML should have overhauled configuration and management more in line with EQL and phased out some of the legacy cruft in I/O management and configuration and streamlined the user interface.
AFAICT, CML is supposed to be competitive with EMC so it's hard to see where EMC is high end and CML is only mid-tier. If EQL had ever added SAS expansion instead of being only whole-controller-plus-disk expansion they would be seriously eroding CML on the low-mid side.
I haven't drunk the VSAN kool-aid so I'm not sold on any of it. I don't think it does much for storage efficiency, reliability or hardware migration, there seems to be a lot of backend I/O overhead.
I do think the emergence of new flash technology (and the continuing decline in current flash prices) will do a lot to upend the storage marketplace. For just about forever, SAN has been about aggregating lots of disks for capacity and performance, with a heavy emphasis on specialized hardware and controllers to manage the performance side. When storage is all flash (ie, a shelf of 24 SSDs supplying 2.4M IOPS and SAS-choking throughput), the need for all the controller complexity seems to go out the window. Beefy interfaces will be necessary, but not necessarily a lot of complexity (like tiering).
I have yet to find a use case for VMware SRM where Veeam replication wasn't simpler, more transparent and more flexible.
The one feature missing from Essentials Plus that is useful is storage VMotion. The gotcha is that most small sites don't use it often enough to justify the licensing charge for it. But when you need it, there's really no third party VMware solution that can match moving a running VMs storage from one storage location to another without disruption.
My worry about this is that Dell will corrupt VMware with hardware-specific "features".
One thing that's been kid of nice about VMware up to this point is that they have been fairly hardware agnostic, which I think improves the innovation of pure virtualization. My worry is that with Dell ownership, they will put pressure on VMware to develop features which give an advantage to Dell hardware solutions.
Maybe it'd be ultimately beneficial to further SDN or SDS, which seem hobbled by a lack of standards at the hardware interface level but I doubt such integration would be oriented at vendor-neutral public standards and more oriented towards monopoly standards.
It's hard to see how they will deal with EMC's storage lineup when they have barely dealt with the Equallogic/Compellent situation.
Compellent still reeks of a lot of old school fiber channel oriented mindset and offers a lot of features and complexity that a lot of small-midsize organizations don't need or can't really use. Plus there is the longer-term question of what a tiering oriented storage system means in a future that seems increasingly destined to be all-flash based.
Equallogic seems to cut out a lot of that cruft at the expense of more costly expansion and I'd wager the hybrid (SSD+HDD) units with 10 gig interfaces probably are performance competitive with CML units with similar disk and network configurations.
I think the core virtualization services are tough to beat. Where I work we do both VMware and Hyper-V, and while Hyper-V is pretty close in features most of us that work with both prefer VMware -- there are just too many weird cases where Hyper-V just doesn't work right, but it is popular among cost-cutters.
Not that VMWare is bad, it is a good product, it is just really expensive, to the point where you can consider getting additional low end servers vs virtualizing them.
The problem with this is that with virtualization you get VM backup and a lot of storage efficiencies you'd never get with single servers, not to mention high availability or hardware replacement. I'm pretty sure low end licenses like Essentials Plus are still cheaper even at 3-4 VMs than additional physical hardware, especially if you have to do better than "low end servers". I'm sure there's some extreme cheapskate math that works in the more hardware method, but you sacrifice a lot of flexibility and you have to also deal with great switching port counts.
IMHO, the problem VMware has is largely that the problem of virtualization is mostly solved and it's hard to grow the business without coming up with lots of add-ons.
I think you're mostly right about organized crime types havng some kind of advantage due to their experience with violence, But I think you underestimate the number of non-ctiminals with experience with violence.
We have at least 100,000 veterans with recent experience in a war zone, many with first-hand experience in urban combat against irregulars. Their training and experience gives them a huge advantage in terms of organization, tactics and strategy. You could expand that 100,000 number into the millions if you factor in non-combat veterans (Navy and Air Force vets, etc) who may not have specific combat experience or deep training, but have experience in military organization and general training.
The other problem with organized crime as a potential survivor is while they may be experienced with violence, their methods of organization and leadership tend to be chaotic, lacking in trust or reliability. Internal conflicts over leadership and spoils tend to be common, with members often turning on or stealing each other.
How long until the gated community phenomenon becomes the walled/domed city phenomenon? The doming part might be unrealistic, but it's not hard to see the the gated community become the "bunker community" where the master plan includes bunkers and defensive barriers to make the whole thing one huge luxury fortress?
I'm not sure how easy it would be to normalize different vehicle suspensions. You'd have to have some kind of test road with a known surface quality that the car could be driven on the calibrate the app to the car.
I'd like to see municipal vehicles and maybe buses equipped with scanners that could map the actual road surface.
In Minneapolis we have a "311" system which is supposed to serve a similar purpose -- report potholes, etc. I was just thinking how much better it would be if I could go stand right on top of this one pothole and get the GPS coordinates of it and send in a picture of it.
the problem with wireless data isn't so much how slow it is; but how costly it is(in part because of scarcity, which more efficient RF technology might actually alleviate, the 'because we can' part is a separate issue)
I'd like to know the difference between the high costs due to real scarcity and the high costs due to profit-taking.
I'd love to see a heat map of cell sites based on RF congestion and backhaul congestion to get an idea if the limits being imposed are really about site limits or mostly about extracting maximum profit.
Many websites only exist because of user generated content (like/.).
Don't impose your idea of what's fair to the content I provide for your site.
Web sites had the chance to go the NPR route and be low key about advertising but by and large they went the obnoxious way and embraced pop ups, pop unders, Flash, animation, and widespread invasive tracking.
Fuck that, I'm not participating in your scheme to get rich off my content, at least the part where I provide you with content and am then expected to be shouted at by ads and tracked. That's not even remotely fair.
Slashdot Mirror
Is energy efficiency that important in most use cases? Best I can tell, copper vs. SFP chassis differ by about 60 watts.for a 24 port model.
SFP+ may be cheaper but the cabling sure isn't and it makes it more expensive yet if you're needing to blend in a handful of 1G devices to the same stack.
If "in a building" translates to "a lot of space" then nickel-iron batteries would make the most sense, since they have a very long lifespan and tolerate deep discharges. But I think their max discharge rate is lower than others, meaning you'd need a larger space for the array.
I think lead acid would suffer from similar issues due to the need to keep maximum discharge depth above 50%.
Lithium batteries have a good discharge profile and take less space.
I'm not sure an idealized 20th century version of hunter-gatherers -- the usual pastiche of peaceful, family-centric, eco-friendly tribes -- is really any more accurate or less stereotypical than Hobbes' view.
Some places maybe had more stable relationships, some maybe had less, but I think it's never really accurate to view them as universally peaceful, free of violent conflict or capable of pretty horrifying abuses. It's not like any of these kinds of groups didn't have weapons, warfare or domineering power relationships until Westerners showed up.
What about the Hobbesian state of nature?
"during the time men live without a common power to keep them all in awe, they are in that condition which is called warre; and such a warre as is of every man against every man"
To seriously saturate a 1gig link, you have to get into 5 figures. At that point, the price of 10gig-e is not a big deal.
Not a big deal until the customer starts comparing quotes. The uptick on the SAN controller for 10 gig ports, the uptick on any servers for 10 gig ports, and then the huge uptick on the 10 gig Ethernet switching.
The SAN controller price increase is of course way beyond the price increase of a generic two-port Intel 10G card, as is the OEM price increase on the server ports. The switching is just breathtaking -- a Dell N series 2024 is $1500, an N4024 is $7500. So for bog-standard cluster with 2x24 port switches, the switching alone is a price increase of $13K and the total project price increase is probably $15k once you factor in controller NICs, host NICs, etc.
So 10 gig can ramp up your pricing by as much as 15%-25% depending on what stuff is baked into it.
I think anybody looking at storage upgrades or network upgrades should definitely eat the increase versus choking on 1 gig for disk transfers, but those customers are awfully stingy with their money.
And it really doesn't take much in terms of fancy disk setups to saturate 1 gig ethernet. More than a pair of SSDs can do it easily and even 7200 SATA arrays of 6 disks have no problem generating 250 mbyte/sec -- that's probably 4 gbit/sec on the wire. Now, I'd agree that most common workloads don't generate that kind of regular traffic but some do and backup is quite often a punishing workload even when it's done right.
10G is great for iSCSI storage and even all-hard disk arrays can overwhelm 1G links pretty easily. Multipathing helps, but like NIC teaming, doesn't result in anything like linear increases in throughput for additional paths.
A fair number of the storage installs I've done lately have been flash cached or tiered arrays and with 10G links, I've seen sequential throughput hit 600 MByte/sec.
Even if a given box can't completely saturate a 10G link, the gap between 1G and 10G is pretty wide -- I'd take 2-3G effective throughput over 1G.
Yeah, but that's one module type that tends to be a little unusual from a real-world usage scenario. Usually people who want 10G-T buy a switching platform that provides those ports built into the chassis or blade. People that buy chassis/blades with all SFP+ ports tend to use them as a fiber landing point or for terminating devices with SFP+ cabling.
The only place I see SFP-copper being used are on maxed-out blades or chassis where the only ports left are SFP and these tend to be 1G platforms anyway.
The 10G installs I've done have tended to be either/or -- all copper with a fiber uplink, or all SFP with a fiber uplink. I've seen one site with a pair of all-SFP Ciscos with a couple of 1G copper SFP modules.
I'm still not entirely convinced that the increased sophistication of 10G *still* accounts for a price differential of nearly 500%. I small profit-taking for obvious business use customers.
Maybe I'm crazy, but it seemed like there was a time when pure access points got strangely expensive and "routers" were cheap.
So I just bought a router and used one of the LAN ports as its uplink, avoiding the "WAN" port and all the routing functionality together. I already have a firewall/router elsewhere.
The only downside I've seen of doing this is some of the devices I've used seem to have some of their ancillary functionality, like NTP, hardcoded to only use the WAN port for outgoing traffic.
How long are they going to keep charging such ridiculous prices for 10 gig networking?
10 gig copper has been out now longer than it took 1 gig copper to go from being "ooh, enterprise" expensive to being in every $499 laptop you could find. Yet they've managed to prop up 10 gig switch and NIC prices forever.
Are 10 gig parts that complicated that they're staying so expensive for so long?
Or are we waiting for the next big "ooh, enterprise" speed bump to come along?
Wasn't Playboy always kind of going out of business once Penthouse and Hustler came around and starting showing women with their legs spread? After that Playboy became kind of irrelevant as pornography.
Won't somebody just game the system and force page reloads or some other statistic-generating scheme?
A pager was cool to have then? I would have thought that you would have needed a Motorola MicroTAC to be at least halfway cool.
I wonder if there's a certain amount of buyer's remorse -- people who thought "I really only drive 70 miles a day, this $electric_car has a range of 120 miles, and I can charge it at home easily".
Then they discover well, they actually do drive past the range frequently, don't get the range they thought they would, or if they did take those things into account, thought that "all those free charging points" would make up for it.
Now they're pissed -- they made a big commitment to an electric car, combined with a lot of self-righteous praise of their choice to everyone who would listen, and they're finding out it that it's not panning out like they thought it would.
Now they have a whole new set of hassles to deal with on a daily basis because their car doesn't have the range they need and recharging on the go has turned into a Mad Max situation.
Yeah, and they have been merging CML and EQL now for 2-3 years with little to show for it. CML is still a great tiering system but overly complicated for just farming out block storage to hypervisors.
IMHO, CML should have overhauled configuration and management more in line with EQL and phased out some of the legacy cruft in I/O management and configuration and streamlined the user interface.
AFAICT, CML is supposed to be competitive with EMC so it's hard to see where EMC is high end and CML is only mid-tier. If EQL had ever added SAS expansion instead of being only whole-controller-plus-disk expansion they would be seriously eroding CML on the low-mid side.
I haven't drunk the VSAN kool-aid so I'm not sold on any of it. I don't think it does much for storage efficiency, reliability or hardware migration, there seems to be a lot of backend I/O overhead.
I do think the emergence of new flash technology (and the continuing decline in current flash prices) will do a lot to upend the storage marketplace. For just about forever, SAN has been about aggregating lots of disks for capacity and performance, with a heavy emphasis on specialized hardware and controllers to manage the performance side. When storage is all flash (ie, a shelf of 24 SSDs supplying 2.4M IOPS and SAS-choking throughput), the need for all the controller complexity seems to go out the window. Beefy interfaces will be necessary, but not necessarily a lot of complexity (like tiering).
I have yet to find a use case for VMware SRM where Veeam replication wasn't simpler, more transparent and more flexible.
The one feature missing from Essentials Plus that is useful is storage VMotion. The gotcha is that most small sites don't use it often enough to justify the licensing charge for it. But when you need it, there's really no third party VMware solution that can match moving a running VMs storage from one storage location to another without disruption.
My worry about this is that Dell will corrupt VMware with hardware-specific "features".
One thing that's been kid of nice about VMware up to this point is that they have been fairly hardware agnostic, which I think improves the innovation of pure virtualization. My worry is that with Dell ownership, they will put pressure on VMware to develop features which give an advantage to Dell hardware solutions.
Maybe it'd be ultimately beneficial to further SDN or SDS, which seem hobbled by a lack of standards at the hardware interface level but I doubt such integration would be oriented at vendor-neutral public standards and more oriented towards monopoly standards.
It's hard to see how they will deal with EMC's storage lineup when they have barely dealt with the Equallogic/Compellent situation.
Compellent still reeks of a lot of old school fiber channel oriented mindset and offers a lot of features and complexity that a lot of small-midsize organizations don't need or can't really use. Plus there is the longer-term question of what a tiering oriented storage system means in a future that seems increasingly destined to be all-flash based.
Equallogic seems to cut out a lot of that cruft at the expense of more costly expansion and I'd wager the hybrid (SSD+HDD) units with 10 gig interfaces probably are performance competitive with CML units with similar disk and network configurations.
I think the core virtualization services are tough to beat. Where I work we do both VMware and Hyper-V, and while Hyper-V is pretty close in features most of us that work with both prefer VMware -- there are just too many weird cases where Hyper-V just doesn't work right, but it is popular among cost-cutters.
Not that VMWare is bad, it is a good product, it is just really expensive, to the point where you can consider getting additional low end servers vs virtualizing them.
The problem with this is that with virtualization you get VM backup and a lot of storage efficiencies you'd never get with single servers, not to mention high availability or hardware replacement. I'm pretty sure low end licenses like Essentials Plus are still cheaper even at 3-4 VMs than additional physical hardware, especially if you have to do better than "low end servers". I'm sure there's some extreme cheapskate math that works in the more hardware method, but you sacrifice a lot of flexibility and you have to also deal with great switching port counts.
IMHO, the problem VMware has is largely that the problem of virtualization is mostly solved and it's hard to grow the business without coming up with lots of add-ons.
I think you're mostly right about organized crime types havng some kind of advantage due to their experience with violence, But I think you underestimate the number of non-ctiminals with experience with violence.
We have at least 100,000 veterans with recent experience in a war zone, many with first-hand experience in urban combat against irregulars. Their training and experience gives them a huge advantage in terms of organization, tactics and strategy. You could expand that 100,000 number into the millions if you factor in non-combat veterans (Navy and Air Force vets, etc) who may not have specific combat experience or deep training, but have experience in military organization and general training.
The other problem with organized crime as a potential survivor is while they may be experienced with violence, their methods of organization and leadership tend to be chaotic, lacking in trust or reliability. Internal conflicts over leadership and spoils tend to be common, with members often turning on or stealing each other.
How long until the gated community phenomenon becomes the walled/domed city phenomenon? The doming part might be unrealistic, but it's not hard to see the the gated community become the "bunker community" where the master plan includes bunkers and defensive barriers to make the whole thing one huge luxury fortress?
I read an interview with a local municipal maintenance guy who said that they use plastic shims on manholes to get them up to level with the road.
They must not use them much, because seldom are they flush. Usually they're obnoxiously below grade.
I'm not sure how easy it would be to normalize different vehicle suspensions. You'd have to have some kind of test road with a known surface quality that the car could be driven on the calibrate the app to the car.
I'd like to see municipal vehicles and maybe buses equipped with scanners that could map the actual road surface.
In Minneapolis we have a "311" system which is supposed to serve a similar purpose -- report potholes, etc. I was just thinking how much better it would be if I could go stand right on top of this one pothole and get the GPS coordinates of it and send in a picture of it.
the problem with wireless data isn't so much how slow it is; but how costly it is(in part because of scarcity, which more efficient RF technology might actually alleviate, the 'because we can' part is a separate issue)
I'd like to know the difference between the high costs due to real scarcity and the high costs due to profit-taking.
I'd love to see a heat map of cell sites based on RF congestion and backhaul congestion to get an idea if the limits being imposed are really about site limits or mostly about extracting maximum profit.
Many websites only exist because of user generated content (like /.).
Don't impose your idea of what's fair to the content I provide for your site.
Web sites had the chance to go the NPR route and be low key about advertising but by and large they went the obnoxious way and embraced pop ups, pop unders, Flash, animation, and widespread invasive tracking.
Fuck that, I'm not participating in your scheme to get rich off my content, at least the part where I provide you with content and am then expected to be shouted at by ads and tracked. That's not even remotely fair.