"I realize that some of the memory in use is shared with other applications"
I ran a test on our systems here, the average for a Gnome application is around 85% shared, so only about 15% of the RAM is actually new memory, that doesn't stop Gnome having a large memory footprint overall though. I imagine it would be similar for KDE.
And it isn't *just* the size of the apps. Gnome is slower than KDE, XFCE, CDE at getting it's stuff up on the display both locally and running over the network.
OK, so It's a dupe, but this time you know it's a dupe.
I forsee a story bounce situation arising in the future. Someone sees a cool story, posts it on/., someone in a large news organisation sees story and files it for future research/publication. When it's published, someone reads it and posts it on/.
"Then 90% of our time is eating up just keeping the walls from falling down"
If 90% of your time is spent fighting fires, there's something fundamentally wrong with the way the systems are set up or you're chronically understaffed. Now, I can scale *myself* from 100 to 1000 systems with little additional effort on my behalf once they are set up.
"They don't exactly teach sysadmin in school, you know."
True, you have to teach yourself. http://www.infrastructures.org/
In general I see my job to automate everything I can. Repetitive work is what computers are good at, get them to do it for you. The sysadmin will still be required to oversee it.
And iiinnn the red corner is Tivo a small but deeeaaaadly black box, iiinnn the BLUEE corner, is the huuuulking eiight huuundddreeedd poound gorillla motionpictureassociation OF AMERICAAAAAAAA!
I run an SMTP server which goes through their servers.
Plusnet are doing the right thing, if a customer is spamming, acting as a relay, sending worms, viruses whether it's deliberate or not, they need to be quarantined.
This is a proven technique in medical circles, it works. If whole ISPs are refusing to quarantine abusive customers then the whole ISP needs to be quarantined as well.
Do your customers pay by the minute for the lights in your store? The air conditioning?
There are a dozen different payment methods, data rates, flat rate payment, by the megabyte payment, by the minute payment, encryption keys, it's almost not worth the hassle. If an ISP were to come along and standardise the lot it might be worth it.
At the moment without the standardisation, the only way wireless is going to work is as an infrastructure cost, perhaps with limited bandwidth and access, encourage people to come in and smell the coffee so to speak.
5 nines represents a loss of service of not more than about 5 minutes per year.
Each additional 9 requires about an order of magnitude increase in the cost of the systems.
e.g. A low cost server with no redundancy will be able to offer 2 nines availability for around $500 - $1,000. This allows around 3 days of downtime where components can be replaced, backups restored.
A low/medium cost server with mirrored drives should be able to offer 3 nines availability for around $2,000 - $10,000. That gives you around 8 hours to get the service back online.
For 4 nines availability, human beings have to be taken out of the equation, we don't work fast enough and make incorrect decisions. 4 nines gives you less than an hour downtime per year. It essentially requires a clustered approach with at least a hot standby machine with redundant access to the data in order to provide the service, you can expect to pay $50,000 -> several hundred thousand dollars for the systems, software etc.
5 nines availability requires multiple remote systems acting simultaneously and redundantly as a single machine, a global cluster if you like. The service has to be duplicated across all of the machines in a transactionally consistent manner. You only have 5mins for a failure to be detected and the service made available transparently from the rest of the cluster. You can expect to pay millions for a 5 nines service.
6 nines less than 30 seconds loss of service per year. Hmm. I'm not aware of anyone even claiming to *actually* provide a 6 nines service. The telcos and power utilities are aiming at it but experience tells me they're not there yet.
100%? Well that'd require an infinite amount of money.
One of the things about an N nines approach is that the whole thing can be compromised by not taking the same approach to *all* aspects of the service.
You realise we're talking about a guarantee of a backup having taken place?
The network failure is a case in point, the lan card on your server develops a fault and the backup fails or is silently corrupted as it's copied, no data has been lost but the backup has not been made successfully. Your hundred percent guarantee from the backup service is useless.
For every 9 of availability, you increase the cost of a system by an order of magnitude. You may well be able to be 99.999% sure that a backup will be made successfully you can never be 100% and anyone who says that they can is a fool or a liar.
The point another poster made about financial compensation is a valid one, yours isn't. Of course, the financial compensation is simply tacked onto the bottom of the bill for the kit required to support 99.99999% availability.
100% doesn't exist in the real world. In the real world there are media errors, drive failures, network failures, administration errors, power outages, disasters etc etc etc.
Go tell your system vendor that you want guaranteed 100% service and watch his beeming grin appear.
"I liked another poster's suggestion of preloading the cache by cat'ing selected binaries to/dev/null"
I mentioned in another post that was what I tried first.
It was taking too long to run through the filesystem and cat each file. The overhead of starting a process, seeking to the file, opening it cating it closing it and destroying the process hundreds or thousands of times is too slow, the ramdisk is faster at boot. Now what would be cool is a way to preload the buffer cache with an entire filesystem without having to go through the filesystem overhead and without resorting to a dedicated ramdisk.
Just a thought, try taring the files to/dev/null rather than cating them individually. Avoids the process overhead but not the filesystem overhead.
Yeah, I tried that at first but you run into the filesystem overhead. The filesystem has to seek to each individual file and it was taking ages to run over the whole partition. dd's faster. It might be worth it for individual applications you have a discreet filesystem for though, i.e./opt/OO
When I click on Open Office or Netscape the CPU and I have to wait for the disk to finish the transfer before we can work. A 15k does it faster. The CPU cycles are wasted because on a desktop they're rarely used for something else. I'd agree with you if we were talking about a server.
You're right, it does still have to be read once from the disk into RAM, and that's where I take my performance hit, at boot time but if you use dd with the right options on the partition rather than the filesystem, the disk will stream it at whatever the maximum streaming speed of the disk is, 50MB/s, 100Mb/s? It adds around 20secs to the boot time for me.
Yup I have 2 copies of the apps in RAM. I have 4Gb of RAM, 2Gb of it as disk. My system doesn't swap, it still has 2Gb of RAM used as RAM and the performance is sensational.
It will... If you are repeatedly stopping and starting the same applications and the system stays up for extended periods. (sounds like a server eh?)
But lets face it... On a desktop system, you boot it up, start a browser, word processor, do some work save it, shut the system down and switch it off at the end. You aren't repeatedly stopping and starting the same apps over and over but you get hit with the N seconds delay every time you have to touch disk.
RAM is cheap, you'll still have 1-2Gb of RAM for all your applications to run in and a huge buffer cache for the rest of the partitions you left on disk anyway, it isn't a case of either/or.
The buffer cache does a great job with random files the second time they are accessed, applications aren't random files and I like Open Office and Gnome to start up in less than 3 seconds.
No it won't. With buffer cache you're always going to have to touch the disk at some point and that point wastes billions of CPU cycles. Minimise thatt by using a f*cking fast disk.
Slashdot Mirror
"I realize that some of the memory in use is shared with other applications"
I ran a test on our systems here, the average for a Gnome application is around 85% shared, so only about 15% of the RAM is actually new memory, that doesn't stop Gnome having a large memory footprint overall though. I imagine it would be similar for KDE.
Ram disk. Load your apps onto it at boot time. Slows down boot but significantly improves startup and running performance.
BTW, I've tried pre warming the buffer cache with libraries, binaries, misc app files etc, it doesn't help much and is slower than loading a ramdisk.
And it isn't *just* the size of the apps. Gnome is slower than KDE, XFCE, CDE at getting it's stuff up on the display both locally and running over the network.
OK, so It's a dupe, but this time you know it's a dupe.
/., someone in a large news organisation sees story and files it for future research/publication. When it's published, someone reads it and posts it on /.
I forsee a story bounce situation arising in the future. Someone sees a cool story, posts it on
Repeat.
"Then 90% of our time is eating up just keeping the walls from falling down"
If 90% of your time is spent fighting fires, there's something fundamentally wrong with the way the systems are set up or you're chronically understaffed. Now, I can scale *myself* from 100 to 1000 systems with little additional effort on my behalf once they are set up.
"They don't exactly teach sysadmin in school, you know."
True, you have to teach yourself. http://www.infrastructures.org/
Yeah. Until something breaks that is.
In general I see my job to automate everything I can. Repetitive work is what computers are good at, get them to do it for you. The sysadmin will still be required to oversee it.
And iiinnn the red corner is Tivo a small but deeeaaaadly black box, iiinnn the BLUEE corner, is the huuuulking eiight huuundddreeedd poound gorillla motionpictureassociation OF AMERICAAAAAAAA!
I run an SMTP server which goes through their servers.
Plusnet are doing the right thing, if a customer is spamming, acting as a relay, sending worms, viruses whether it's deliberate or not, they need to be quarantined.
This is a proven technique in medical circles, it works. If whole ISPs are refusing to quarantine abusive customers then the whole ISP needs to be quarantined as well.
I don't know anyone who uses "quarter of" when referring to time, and I'm a Scot in England. We use quarter to three, quarter past three. etc.
Do your customers pay by the minute for the lights in your store? The air conditioning?
There are a dozen different payment methods, data rates, flat rate payment, by the megabyte payment, by the minute payment, encryption keys, it's almost not worth the hassle. If an ISP were to come along and standardise the lot it might be worth it.
At the moment without the standardisation, the only way wireless is going to work is as an infrastructure cost, perhaps with limited bandwidth and access, encourage people to come in and smell the coffee so to speak.
"What is it with all the useless freecache links lately?"
Perhaps Freecache is unexpectedly useless.
If you can work 10% harder today then you were not working 100% yesterday.
3 nines is 99.9%, 5 nines is 99.999%
5 nines represents a loss of service of not more than about 5 minutes per year.
Each additional 9 requires about an order of magnitude increase in the cost of the systems.
e.g.
A low cost server with no redundancy will be able to offer 2 nines availability for around $500 - $1,000. This allows around 3 days of downtime where components can be replaced, backups restored.
A low/medium cost server with mirrored drives should be able to offer 3 nines availability for around $2,000 - $10,000. That gives you around 8 hours to get the service back online.
For 4 nines availability, human beings have to be taken out of the equation, we don't work fast enough and make incorrect decisions. 4 nines gives you less than an hour downtime per year. It essentially requires a clustered approach with at least a hot standby machine with redundant access to the data in order to provide the service, you can expect to pay $50,000 -> several hundred thousand dollars for the systems, software etc.
5 nines availability requires multiple remote systems acting simultaneously and redundantly as a single machine, a global cluster if you like. The service has to be duplicated across all of the machines in a transactionally consistent manner. You only have 5mins for a failure to be detected and the service made available transparently from the rest of the cluster. You can expect to pay millions for a 5 nines service.
6 nines less than 30 seconds loss of service per year. Hmm. I'm not aware of anyone even claiming to *actually* provide a 6 nines service. The telcos and power utilities are aiming at it but experience tells me they're not there yet.
100%? Well that'd require an infinite amount of money.
One of the things about an N nines approach is that the whole thing can be compromised by not taking the same approach to *all* aspects of the service.
You realise we're talking about a guarantee of a backup having taken place?
The network failure is a case in point, the lan card on your server develops a fault and the backup fails or is silently corrupted as it's copied, no data has been lost but the backup has not been made successfully. Your hundred percent guarantee from the backup service is useless.
For every 9 of availability, you increase the cost of a system by an order of magnitude. You may well be able to be 99.999% sure that a backup will be made successfully you can never be 100% and anyone who says that they can is a fool or a liar.
The point another poster made about financial compensation is a valid one, yours isn't. Of course, the financial compensation is simply tacked onto the bottom of the bill for the kit required to support 99.99999% availability.
"In the real world, yes, 100% does exist"
Quantum mechanics pretty much says you're wrong.
And what is it about those f*cking idiots who talk about "giving 110%". You can't give even 100%, if you did you'd be *dead*.
"Yeah, man we worked 10% harder than we could possibly have done".
Of course 110% isn't enough, now *everyone* gives 110% and we now have to give 120%, 200%, 1000%!
100% doesn't exist in the real world. In the real world there are media errors, drive failures, network failures, administration errors, power outages, disasters etc etc etc.
Go tell your system vendor that you want guaranteed 100% service and watch his beeming grin appear.
SecureDigital cards have SDMI built in. MMC are completely digital rights management free. You can use MMC cards in devices which can read SD cards.
"I liked another poster's suggestion of preloading the cache by cat'ing selected binaries to /dev/null"
/dev/null rather than cating them individually. Avoids the process overhead but not the filesystem overhead.
I mentioned in another post that was what I tried first.
It was taking too long to run through the filesystem and cat each file. The overhead of starting a process, seeking to the file, opening it cating it closing it and destroying the process hundreds or thousands of times is too slow, the ramdisk is faster at boot. Now what would be cool is a way to preload the buffer cache with an entire filesystem without having to go through the filesystem overhead and without resorting to a dedicated ramdisk.
Just a thought, try taring the files to
Yeah, I tried that at first but you run into the filesystem overhead. The filesystem has to seek to each individual file and it was taking ages to run over the whole partition. dd's faster. It might be worth it for individual applications you have a discreet filesystem for though, i.e. /opt/OO
We're talking desktops here.
When I click on Open Office or Netscape the CPU and I have to wait for the disk to finish the transfer before we can work. A 15k does it faster. The CPU cycles are wasted because on a desktop they're rarely used for something else. I'd agree with you if we were talking about a server.
You're right, it does still have to be read once from the disk into RAM, and that's where I take my performance hit, at boot time but if you use dd with the right options on the partition rather than the filesystem, the disk will stream it at whatever the maximum streaming speed of the disk is, 50MB/s, 100Mb/s? It adds around 20secs to the boot time for me.
Yup I have 2 copies of the apps in RAM. I have 4Gb of RAM, 2Gb of it as disk. My system doesn't swap, it still has 2Gb of RAM used as RAM and the performance is sensational.
It will... If you are repeatedly stopping and starting the same applications and the system stays up for extended periods. (sounds like a server eh?)
But lets face it... On a desktop system, you boot it up, start a browser, word processor, do some work save it, shut the system down and switch it off at the end. You aren't repeatedly stopping and starting the same apps over and over but you get hit with the N seconds delay every time you have to touch disk.
RAM is cheap, you'll still have 1-2Gb of RAM for all your applications to run in and a huge buffer cache for the rest of the partitions you left on disk anyway, it isn't a case of either/or.
The buffer cache does a great job with random files the second time they are accessed, applications aren't random files and I like Open Office and Gnome to start up in less than 3 seconds.
No it won't. With buffer cache you're always going to have to touch the disk at some point and that point wastes billions of CPU cycles. Minimise thatt by using a f*cking fast disk.