If you feel that's a better test, I can certainly do that. Maybe even if it's prime stable for 24 and so forth, I could find the top speed it's stable at and then back it down 100Mhz or something as well. Basically as long as I can reproduce an error that has happened in the wild with a copy of the customer's db on this private machine I'll succeed in my endeavors. Thanks for the linX idea, btw. Maybe 24 hr prime, 24 hr linx...maybe even 48 hours. I'll hammer on it for a month or more maybe even total. I will probably start with a small OC and move upward as I get free time to test more. Start at 3.2 or so and move upward.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Navigation
Install the app
How to install the app on iOS
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
More options
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Building MySQL overclocked i7 for work
- Thread starter mystykmax
- Start date
If you feel that's a better test, I can certainly do that. Maybe even if it's prime stable for 24 and so forth, I could find the top speed it's stable at and then back it down 100Mhz or something as well. Basically as long as I can reproduce an error that has happened in the wild with a copy of the customer's db on this private machine I'll succeed in my endeavors. Thanks for the linX idea, btw. Maybe 24 hr prime, 24 hr linx...maybe even 48 hours. I'll hammer on it for a month or more maybe even total. I will probably start with a small OC and move upward as I get free time to test more. Start at 3.2 or so and move upward.
soulesschild
Supreme [H]ardness
- Joined
- Feb 18, 2007
- Messages
- 6,176
Another specific question I have is, will 12GB of RAM overclock fine? I mean if I'm aiming at hopefully 4ishGhz, will the 12GB of RAM I have withstand that?
I've always heard that more ram = more OCing headache. But since I don't have an i7 to OC with, I don't know if that changed at all or not.
Speaking as a DBA with MySQL experience, I'm pretty surprised your imports are CPU-bound. In my experience they're always I/O bound when done properly. There are tons of things you can do to speed them up, and I could go through them here, but if you've spoken with someone from sun/mysql you probably already tried all that stuff.
I still say that rather than spending all your budget on crazy CPUs and elaborate cooling, you should consider investing in a hardware RAID controller and setup a RAID1+0 array to hold the datafiles.
I still say that rather than spending all your budget on crazy CPUs and elaborate cooling, you should consider investing in a hardware RAID controller and setup a RAID1+0 array to hold the datafiles.
Speaking as a DBA with MySQL experience, I'm pretty surprised your imports are CPU-bound. In my experience they're always I/O bound when done properly. There are tons of things you can do to speed them up, and I could go through them here, but if you've spoken with someone from sun/mysql you probably already tried all that stuff.
I still say that rather than spending all your budget on crazy CPUs and elaborate cooling, you should consider investing in a hardware RAID controller and setup a RAID1+0 array to hold the datafiles.
yup, db = i/o bound
Speaking as a DBA with MySQL experience, I'm pretty surprised your imports are CPU-bound. In my experience they're always I/O bound when done properly. There are tons of things you can do to speed them up, and I could go through them here, but if you've spoken with someone from sun/mysql you probably already tried all that stuff.
I still say that rather than spending all your budget on crazy CPUs and elaborate cooling, you should consider investing in a hardware RAID controller and setup a RAID1+0 array to hold the datafiles.
Well I was testing on opterons with 8-disk RAID 10 and through all the testing of three different people we are CPU bound on the import. Since it's InnoDB I have to have my own instance of MySQL for this when doing a hot-copy, but in the case of loading a dump file I can't do a hot copy because the data transfer would be in the terabyte range and I'd have to shut down live hosted servers to make that copy.
nope. Not with RAID 10 8-disk array on a hardware controller.yup, db = i/o bound
Well today's CPUs are so damn fast that almost nothing is actually CPU bound these days.
A Windows example would be flight simulator X or try movie encoding or massive photoshop edits, indesign, etc. The other thing is if an app has pieces of it that are not multi-threaded or just not well-multi-threaded.
I'm not trying to be insolent here, but I'd like to see what you guys can do with a MySQL dump...importing it via "zcat dump.gz | mysql dbname" or simply "mysql dbname < dump.sql"...and the db has to be (when actually imported) above 50GB, with an upper limit of 200GB.
Well I would set innodb_flush_log_at_trx_commit to 0 (very big deal), make sure to set innodb thread concurrency=4 to avoid the SMP bug (also a huuuge deal if you have >4 CPUs), increase the innodb normal and log buffers and the additional mem pool, set innodb log files to at least 512MB, put the logs and datafiles on separate storage, try doing several imports in parallel (but not more than 3 due to that SMP concurrency bug), turn off foreign key checks and autocommit on the import, take the mysqldump with extended inserts, try using load data infile/mysqlimport rather than SQL dumps, and finally try importing into myisam and then converting to innodb after the fact. But I'm sure talking to Sun you tried all that stuff, right?
When I say "nothing is CPU bound" I was talking about databases. Obviously media encoding and videogames aren't covered by that. You are coming across as condescending, and need to accept that on the internet you're talking to a larger audience and may actually run into people who aren't total imbeciles.
When I say "nothing is CPU bound" I was talking about databases. Obviously media encoding and videogames aren't covered by that. You are coming across as condescending, and need to accept that on the internet you're talking to a larger audience and may actually run into people who aren't total imbeciles.
Well I would set innodb_flush_log_at_trx_commit to 0 (very big deal), make sure to set innodb thread concurrency=4 to avoid the SMP bug (also a huuuge deal if you have >4 CPUs), increase the innodb normal and log buffers and the additional mem pool, set innodb log files to at least 512MB, put the logs and datafiles on separate storage, try doing several imports in parallel (but not more than 3 due to that SMP concurrency bug), turn off foreign key checks and autocommit on the import, take the mysqldump with extended inserts, try using load data infile/mysqlimport rather than SQL dumps, and finally try importing into myisam and then converting to innodb after the fact. But I'm sure talking to Sun you tried all that stuff, right?
When I say "nothing is CPU bound" I was talking about databases. Obviously media encoding and videogames aren't covered by that. You are coming across as condescending, and need to accept that on the internet you're talking to a larger audience and may actually run into people who aren't total imbeciles.
I probably was speaking out of some frustration if I came across that way so I apologize. My point was never that everyone was stupid, it was that no one was reading what I was asking for. Instead (not necessarily you though) most people did exactly what you are are saying I did...they accused me of being an imbecile and that overclocking was stupid without even asking me questions about the business case. You can probably expect a little defensiveness after being punched in the face verbally a few times without so much as some cursory information gathering. Do you realize you are the first person to actually suggest MySQL knobs get turned? I'd challenge you to start at the beginning of this thread and then put yourself in my shoes and start reading. Most (not all) people weren't helpful at all, simply condescending, insulting and never bothered to really ask a question. Again, I apologize if I come across the wrong way, but after the second page I was starting to get frustrated with the lack of common courtesy.
Yes, understood, but they weren't actually wrong, in that it shouldn't be CPU bound if done properly. That CPU load might be symptomatic of logical I/O, or the kernel interface to your disk controller spinning around as it does tons of tiny writes to disk with every single statement committing , or you might have >4 CPUs and hit that incredible SMP bug. Difficult to say exactly without looking at it.
If your prod DB is on some sort of SAN you can also try doing a brief global write lock, a BCV split, and then shutting down the split mirror (or replica) and copying the files over physically. That would be pretty much instantaneous and a far better choice than a logical restore of 50GB into mysql. Mysql doesn't handle larger datasets very well.
And there's also innobase's commercial mysql hotbackup tool. But I haven't actually tried that.
If your prod DB is on some sort of SAN you can also try doing a brief global write lock, a BCV split, and then shutting down the split mirror (or replica) and copying the files over physically. That would be pretty much instantaneous and a far better choice than a logical restore of 50GB into mysql. Mysql doesn't handle larger datasets very well.
And there's also innobase's commercial mysql hotbackup tool. But I haven't actually tried that.
Yes, understood, but they weren't actually wrong, in that it shouldn't be CPU bound if done properly. That CPU load might be symptomatic of logical I/O, or the kernel interface to your disk controller spinning around as it does tons of tiny writes to disk with every single statement committing , or you might have >4 CPUs and hit that incredible SMP bug. Difficult to say exactly without looking at it.
If your prod DB is on some sort of SAN you can also try doing a brief global write lock, a BCV split, and then shutting down the split mirror (or replica) and copying the files over physically. That would be pretty much instantaneous and a far better choice than a logical restore of 50GB into mysql. Mysql doesn't handle larger datasets very well.
And there's also innobase's commercial mysql hotbackup tool. But I haven't actually tried that.
They were saying I was stupid to have an overclocked system in production and that customers would not be happy with downtime. I had said to them that this was not customer facing nor was it in production. You're the only one who offered any actual implementable advice for MySQL. So the thing was, they were attacking me for something I said didn't exist...customer facing production server...which it was/is neither. They were worried about customers which in my situation is not an issue while I was worried about performance.
I think maybe I've confused some by now of what the process is for me. First a customer has an issue with our hosted solution. I ask for a copy of that particular db if I can't figure out what it is by looking at the log files from our product. Eventually they (hosting dept) gives me a gzipped dump file of the db. I have to load this into my own instance of MySQL that other devs are currently using so I can't mess around with knobs on that server too much as far as I know. Having my own MySQL server I'd be able to fiddle with the knobs as much as I need. Also I'll be running unit testing on very large db's on this machine eventually...logical errors, memory leaks, etc would get tested on large datasets. I know our hosting guys do the cutover thing and swap back and forth with failover servers and so forth, but I don't have much control over how they pull the data. I know one of the guys on the team actually submits a fair amount of bugs to MySQL and also gets features implemented in special builds for us. I'm not saying we know everything but that's why I've trusted the guy(s) over there so far anyway. I'm going to try the knobs you suggested if I can find a machine with no users here somewhere. I'd like to prove this part to myself. The guy that worked for Sun tested it on his own and didn't really tell me the knobs he turned on it so I can't speak for him in that regard. I appreciate you taking the time to converse here and give me some thoughts.
The other thing is that the db, when downloaded, will never be pushed back to production. I use it for debugging, fix the logical error that caused the bug and check into cvs. Database gets thrown away or stored...
In reference to your hardware, since that's what you originally asked:
The EVGA X58 board is what I am running, and it is a nice board and I have had a pleasant experience with it. Since the EVGA X58 Classified was just released and was designed for people running watercooled/phase changed hardware and (supposedly) has better power regulation for the CPU, if you are looking for the cream of the crop overclocking hardware I might look into that board. The ASUS P6T offerings are also good, as are Gigabyte's. I do not feel you will go wrong with boards from either of those three vendors.
For RAM, Corsair's Dominator memory is excellent. It's more expensive but their support is great - I got a bad stick of cheaper Corsair memory and I had to swap it, and they replaced my entire kit for free with exception of paying shipping to them. Do you need the 1866MHz kits? Maybe, maybe not. If you plan on getting a 965, you can adjust the multiplier upwards and thus do not need to increase your QPI bus to overclock as much, in which case the 1600MHz memory is fine. For the 920 CPU, you hit a wall at 200x21 = 4200MHz. If you plan on going above that you will need to run RAM faster than the 1600MHz modules are rated for. The decision is really up to you, the processor you select and the goals you are looking to achieve.
Unlike some of these posters I have an i7 system and if you want more info on my particular experiences with these products give me a shout. I am not a DBA so I can't help with any of your MySQL tuning questions but I can vouch for the hardware I run.
The EVGA X58 board is what I am running, and it is a nice board and I have had a pleasant experience with it. Since the EVGA X58 Classified was just released and was designed for people running watercooled/phase changed hardware and (supposedly) has better power regulation for the CPU, if you are looking for the cream of the crop overclocking hardware I might look into that board. The ASUS P6T offerings are also good, as are Gigabyte's. I do not feel you will go wrong with boards from either of those three vendors.
For RAM, Corsair's Dominator memory is excellent. It's more expensive but their support is great - I got a bad stick of cheaper Corsair memory and I had to swap it, and they replaced my entire kit for free with exception of paying shipping to them. Do you need the 1866MHz kits? Maybe, maybe not. If you plan on getting a 965, you can adjust the multiplier upwards and thus do not need to increase your QPI bus to overclock as much, in which case the 1600MHz memory is fine. For the 920 CPU, you hit a wall at 200x21 = 4200MHz. If you plan on going above that you will need to run RAM faster than the 1600MHz modules are rated for. The decision is really up to you, the processor you select and the goals you are looking to achieve.
Unlike some of these posters I have an i7 system and if you want more info on my particular experiences with these products give me a shout. I am not a DBA so I can't help with any of your MySQL tuning questions but I can vouch for the hardware I run.
In reference to your hardware, since that's what you originally asked:
The EVGA X58 board is what I am running, and it is a nice board and I have had a pleasant experience with it. Since the EVGA X58 Classified was just released and was designed for people running watercooled/phase changed hardware and (supposedly) has better power regulation for the CPU, if you are looking for the cream of the crop overclocking hardware I might look into that board. The ASUS P6T offerings are also good, as are Gigabyte's. I do not feel you will go wrong with boards from either of those three vendors.
For RAM, Corsair's Dominator memory is excellent. It's more expensive but their support is great - I got a bad stick of cheaper Corsair memory and I had to swap it, and they replaced my entire kit for free with exception of paying shipping to them. Do you need the 1866MHz kits? Maybe, maybe not. If you plan on getting a 965, you can adjust the multiplier upwards and thus do not need to increase your QPI bus to overclock as much, in which case the 1600MHz memory is fine. For the 920 CPU, you hit a wall at 200x21 = 4200MHz. If you plan on going above that you will need to run RAM faster than the 1600MHz modules are rated for. The decision is really up to you, the processor you select and the goals you are looking to achieve.
Unlike some of these posters I have an i7 system and if you want more info on my particular experiences with these products give me a shout. I am not a DBA so I can't help with any of your MySQL tuning questions but I can vouch for the hardware I run.
I may check out that evga board. I have heard good things about that company. I think the 1600Mhz RAM is probably good enough for what I'm doing. If I can hit 4.2, I'll be happy.
in my opinion and i think most would agree the biggest appeal for overclocking aside from just personal satisfaction of tweaking something and flexing the e-peen around is to try to maximize resources you have because of a limited budget. (ie. buying a celeron 300a and doing a o/c to 450mhz) and getting the higher end cpu performance but at the low end price.
that being said i'm not sure what business purpose you fill when you spend big money on midrange parts and cooling in order to tweak it to top end performance when you could just spend the money and get the top end parts. it is like buying a civic for $20,000 and then throwing another $20,000 $30,000 worth of mods to get it to perform like an m3. why not just get the m3, if you are going to spend the money anyways.
i would say the money is better served grabbing xenons and then outfitting the system with as much ram as it can handle,
it sounds like you were able to convince someone to buy a bunch of toys for you to play with at work, which is great and more power to you, but i don't know if i would agree that this would be the best way to spend the money to do this job
that being said i'm not sure what business purpose you fill when you spend big money on midrange parts and cooling in order to tweak it to top end performance when you could just spend the money and get the top end parts. it is like buying a civic for $20,000 and then throwing another $20,000 $30,000 worth of mods to get it to perform like an m3. why not just get the m3, if you are going to spend the money anyways.
i would say the money is better served grabbing xenons and then outfitting the system with as much ram as it can handle,
it sounds like you were able to convince someone to buy a bunch of toys for you to play with at work, which is great and more power to you, but i don't know if i would agree that this would be the best way to spend the money to do this job
A 965 is the fastest, in terms of overall CPU speed, processor you can buy, especially if overclocked to 4GHz+.in my opinion and i think most would agree the biggest appeal for overclocking aside from just personal satisfaction of tweaking something and flexing the e-peen around is to try to maximize resources you have because of a limited budget. (ie. buying a celeron 300a and doing a o/c to 450mhz) and getting the higher end cpu performance but at the low end price.
that being said i'm not sure what business purpose you fill when you spend big money on midrange parts and cooling in order to tweak it to top end performance when you could just spend the money and get the top end parts. it is like buying a civic for $20,000 and then throwing another $20,000 $30,000 worth of mods to get it to perform like an m3. why not just get the m3, if you are going to spend the money anyways.
i would say the money is better served grabbing xenons and then outfitting the system with as much ram as it can handle,
it sounds like you were able to convince someone to buy a bunch of toys for you to play with at work, which is great and more power to you, but i don't know if i would agree that this would be the best way to spend the money to do this job
There are no Nehalem-based Xeons available yet AFAIK.
A 965 is the fastest, in terms of overall CPU speed, processor you can buy, especially if overclocked to 4GHz+.
There are no Nehalem-based Xeons available yet AFAIK.
that is correct, the Xeon chips are not out yet, only one available is inside a Mac Pro
http://www.apple.com/macpro/
mrmylanman
[H]ard|Gawd
- Joined
- Jul 13, 2008
- Messages
- 1,484
If you are going to overclock your CPU (I'm sure you will) be sure to only populate 3 DIMMs. My i7's maximum stable overclock went WAY down when I populated all 6.
The highest I could get was 3.8 beforehand (although I never really pushed it, heat was an issue at that speed for my case). Afterwards? 3.2 at the same voltage. So heat is still an issue. But anything higher than that was unstable anyway so I'll be buying some different RAM and selling this at some point...
However, if I were you I would wait on the Xeons to come out, and get one of those for a grand, and get a server board (if there are any available that allow good overclocking) when those come out. They really are meant more for that sort of thing. That way you can always pop another processor in there whenever you need to upgrade and it'll be a lot cheaper than buying a whole new server.
Since Apple is selling the new Mac Pro with the Xeons in it, it's only a short matter of time until they hit retail.
The highest I could get was 3.8 beforehand (although I never really pushed it, heat was an issue at that speed for my case). Afterwards? 3.2 at the same voltage. So heat is still an issue. But anything higher than that was unstable anyway so I'll be buying some different RAM and selling this at some point...
However, if I were you I would wait on the Xeons to come out, and get one of those for a grand, and get a server board (if there are any available that allow good overclocking) when those come out. They really are meant more for that sort of thing. That way you can always pop another processor in there whenever you need to upgrade and it'll be a lot cheaper than buying a whole new server.
Since Apple is selling the new Mac Pro with the Xeons in it, it's only a short matter of time until they hit retail.
THERMAL-REACTOR
n00b
- Joined
- Nov 27, 2008
- Messages
- 53
I have not had any problems at all with 12gb`s of ram (6 slots filled). I have a nice stable everyday OC at 4.0Ghz. I have even hit 4.5 with all slots filled.
I think the above poster just has some bad luck or should get a different mobo.
I think the above poster just has some bad luck or should get a different mobo.
mrmylanman
[H]ard|Gawd
- Joined
- Jul 13, 2008
- Messages
- 1,484
It's a pretty commonly known fact that the more DIMM slots you fill the less overclock you generally achieve. Have you tested your overclock using a stability program? Mine would boot just fine and even had ok temperatures, but once I ran Prime95 it shut down in less than 2 minutes. Or it just locked the computer up.
in my opinion and i think most would agree the biggest appeal for overclocking aside from just personal satisfaction of tweaking something and flexing the e-peen around is to try to maximize resources you have because of a limited budget. (ie. buying a celeron 300a and doing a o/c to 450mhz) and getting the higher end cpu performance but at the low end price.
that being said i'm not sure what business purpose you fill when you spend big money on midrange parts and cooling in order to tweak it to top end performance when you could just spend the money and get the top end parts. it is like buying a civic for $20,000 and then throwing another $20,000 $30,000 worth of mods to get it to perform like an m3. why not just get the m3, if you are going to spend the money anyways.
i would say the money is better served grabbing xenons and then outfitting the system with as much ram as it can handle,
it sounds like you were able to convince someone to buy a bunch of toys for you to play with at work, which is great and more power to you, but i don't know if i would agree that this would be the best way to spend the money to do this job
They don't make a xeon i7 yet. We've built a couple MySQL servers already in the $10k to $20k range and that's fine for most multiple user scenarios. I'm the db upgrade guy. Get big db, load it, test. fix. test. This has nothing to do with e-peen. It has to do with db peen. it's not like buying a civic and putting 20 or 30 k into it. That's an analogy that falls completely flat. Not to mention they are looking at possibly water-cooling more of our current xeon servers so this is also a bit of R&D on that end. The nice thing about the Koolance systems is that they are pretty forward compatible. Sometimes you have to buy a new heatsink, but that's it. I'm still using the cases/systems at home that I bought 7 years ago and they work great.
in my opinion and i think most would agree the biggest appeal for overclocking aside from just personal satisfaction of tweaking something and flexing the e-peen around is to try to maximize resources you have because of a limited budget. (ie. buying a celeron 300a and doing a o/c to 450mhz) and getting the higher end cpu performance but at the low end price.
that being said i'm not sure what business purpose you fill when you spend big money on midrange parts and cooling in order to tweak it to top end performance when you could just spend the money and get the top end parts. it is like buying a civic for $20,000 and then throwing another $20,000 $30,000 worth of mods to get it to perform like an m3. why not just get the m3, if you are going to spend the money anyways.
i would say the money is better served grabbing xenons and then outfitting the system with as much ram as it can handle,
it sounds like you were able to convince someone to buy a bunch of toys for you to play with at work, which is great and more power to you, but i don't know if i would agree that this would be the best way to spend the money to do this job
I'd say this is more like buying an Enzo and tuning the engine to get more (if that were possible) power out of it. There isn't anything else out beyond a 965 and I doubt the xeon i7's will OC better.
To OP: Ignore most of these people. You should do what you want to do that you can convince your employer to let you do without wasting a lot of money/resources. In the grand scheme of things, even if you set the thing on fire and burn it to ash, it's very little money. That's not going to happen anyway.
I've been using dozens of the evga x58 boards here with 12gb of 1600 dominator and either 920's or 940's and they usually hit between 3.6 and 4.2 on air with a TRUE, so I'd say a 965 on water should easily do 4.5. I may try that sometime... Anyway, I've only had one doa board so far. I've not had a chance to get my hands on a classified one yet but I will with the next order.
Bottom line, whatever you do with this system, assuming it's stable, it's not going to perform worse OC'd versus not.
Keep us updated and have fun!
RaphaelVinceti
Gawd
- Joined
- Apr 7, 2005
- Messages
- 600
On the cooling end, I would recommend a 240x120 rad as well, 3/8" ID tubing (make sure you get a rad with those fittings as well), and a D-Tek cpu block. I have used D-Tek in a couple high-end customer builds, they are all nicely built and work very well transferring heat from the CPU. Also make sure you get a good high-quality TIM. OCZ Freeze won the last review I looked at for it, I believe.. Course, you cant really go wrong with AS-5.
On the hardware front, as someone else said, the 920 is going to be a bit harder to overclock above 4GHz than the 965, simply because of the locked multiplier, and if you add 6 sticks of ram, its 50/50 on if you'll get the same OC as 3 sticks. (Oh, and I recommend the Corsair Dominator ram as well. Expensive, but overclocks like MAD!) I would love to see how this turns out, so, despite the skeptics, post results if you can from the Overclocking, and what the speed differences importing SQL dumps into MySQL are. Good Luck!
On the hardware front, as someone else said, the 920 is going to be a bit harder to overclock above 4GHz than the 965, simply because of the locked multiplier, and if you add 6 sticks of ram, its 50/50 on if you'll get the same OC as 3 sticks. (Oh, and I recommend the Corsair Dominator ram as well. Expensive, but overclocks like MAD!) I would love to see how this turns out, so, despite the skeptics, post results if you can from the Overclocking, and what the speed differences importing SQL dumps into MySQL are. Good Luck!
THERMAL-REACTOR
n00b
- Joined
- Nov 27, 2008
- Messages
- 53
It's a pretty commonly known fact that the more DIMM slots you fill the less overclock you generally achieve. Have you tested your overclock using a stability program? Mine would boot just fine and even had ok temperatures, but once I ran Prime95 it shut down in less than 2 minutes. Or it just locked the computer up.
Yes it primed for 6 hours at 4.0ghz. I could run it at 4.2 but the temp's get too high for my liking.
mrmylanman
[H]ard|Gawd
- Joined
- Jul 13, 2008
- Messages
- 1,484
That is strange. I don't need 12GB but I will probably end up selling my 1GB sticks for 2GB sticks and just have 3 slots populated, and go back to my 3.8GHz or so. I can probably get it higher but I also ran in to heat issues.
It's pretty funny to hear all these people saying "Overclocking a machine whose reliability is important is so stupid" - this is HardOCP - do you all think that you can't overclock a machine reliably?
Now don't get me wrong - I'm not saying that it's easy, or even that the OP will succeed with his OC, but to suggest that it is not possible to OC a machine and have a stable, reliable platform is pretty pathetic.
You're basically saying "Overclocking is an unreliable way to increase system performance, at the cost of reliability" - which translates to "I don't know how to overclock well enough to trust the stability of my machine, so I suggest others do not as well"
Just sayin.
Now don't get me wrong - I'm not saying that it's easy, or even that the OP will succeed with his OC, but to suggest that it is not possible to OC a machine and have a stable, reliable platform is pretty pathetic.
You're basically saying "Overclocking is an unreliable way to increase system performance, at the cost of reliability" - which translates to "I don't know how to overclock well enough to trust the stability of my machine, so I suggest others do not as well"
Just sayin.
Well I would set innodb_flush_log_at_trx_commit to 0 (very big deal), make sure to set innodb thread concurrency=4 to avoid the SMP bug (also a huuuge deal if you have >4 CPUs), increase the innodb normal and log buffers and the additional mem pool, set innodb log files to at least 512MB, put the logs and datafiles on separate storage, try doing several imports in parallel (but not more than 3 due to that SMP concurrency bug), turn off foreign key checks and autocommit on the import, take the mysqldump with extended inserts, try using load data infile/mysqlimport rather than SQL dumps, and finally try importing into myisam and then converting to innodb after the fact. But I'm sure talking to Sun you tried all that stuff, right?
So I'm catching the bit of sarcasm there at the end. But for the official word, we tried all the items you mentioned already. We don't use db-level foreign keys so the foreign key checks was definitely off and autocommit is turned off in our .my.cnf files of course. Converting from myisam to innodb after the fact takes quite awhile so nothing really gained there. Also, we are using Percona as well to see if that would speed things up appreciably. http://www.percona.com/percona-lab.html. FWIW bouncing the MySQL server before importing seemed to help a fair amount with import speed.
It's pretty funny to hear all these people saying "Overclocking a machine whose reliability is important is so stupid" - this is HardOCP - do you all think that you can't overclock a machine reliably?
Now don't get me wrong - I'm not saying that it's easy, or even that the OP will succeed with his OC, but to suggest that it is not possible to OC a machine and have a stable, reliable platform is pretty pathetic.
You're basically saying "Overclocking is an unreliable way to increase system performance, at the cost of reliability" - which translates to "I don't know how to overclock well enough to trust the stability of my machine, so I suggest others do not as well"
Just sayin.
Good points! It seemed many here were very worried about what the customer would think of a database server they would never see nor use nor rely upon. *shakes head*
It's pretty funny to hear all these people saying "Overclocking a machine whose reliability is important is so stupid" - this is HardOCP - do you all think that you can't overclock a machine reliably?
Now don't get me wrong - I'm not saying that it's easy, or even that the OP will succeed with his OC, but to suggest that it is not possible to OC a machine and have a stable, reliable platform is pretty pathetic.
You're basically saying "Overclocking is an unreliable way to increase system performance, at the cost of reliability" - which translates to "I don't know how to overclock well enough to trust the stability of my machine, so I suggest others do not as well"
Just sayin.
Pathetic? Sounds to me like you just don't have any experience with servers or other mission-critical or scientific computing applications. "Just sayin"
It seems to me like there are two kinds of people in this thread: Those who have experience with mission-critical applications (servers, etc.) and those who have only dealt with consumer-level systems. The definitions of 'stability' and 'reliability' are vastly different between these two camps, and for good reason.
For a consumer or the average HardOCP user, 'stable' and 'reliable' means that the machine can pass Prime95 for 24 hours, or run games without crashing indefinitely, or some other metric that basically boils down to not crashing and not suffering any errors. This is perfect for a personal application.
For a server application, there are more factors that come into play. Not only must the system not crash and not encounter any data corruption, but it must do so indefinitely, meaning for many years to come without further adjustment. It also must do so whether someone is there to attend to it or not. That is to say, the hardware should not degrade over time. It is well known among the HardOCP and overclocking communities that overclocking and, more specifically, overvolting reduce the life of a CPU. A system which was stable at XX GHz when you built it might very well not be stable 2 years down the road and you start getting unknown crashes. And yes, I'm sure your cousin Bob overclocked his Pentium 4 six years ago and hasn't touched it since and it runs just fine. Doesn't matter, you can't count on that.
There's a reason why people pay more for Xeon hardware, ECC RAM, and server-class motherboards. To suggest that a ragged-edge, overclocked prosumer-level system could match the long-term stability of server-class hardware is ridiculous.
Now if you KNOW you can tolerate possible errors creeping into your data, and you won't be bothered if your processor dies seven months from now and you have to spend several hundred dollars, a few days of waiting for shipping, and countless hours restoring the system to working order, then feel free to overclock. But you also have to KNOW that the server will not be moved to a closet or rack with a higher ambient temperature. You've got to KNOW that no one will be depending on this server for their data or their business needs. You've got to be absolutely sure that YOU will be around to fix it when it breaks. If that's the case, and you understand and do not underestimate the risks involved, then go ahead.
One more thing, before I forget:
You've got to make sure that absolutely no one else can touch the computer. If you overclock the machine and, down the road, someone else decides to double the amount of RAM in the machine, for example, you're VERY likely to run into issues. With a stock-clocked system, upgrades and changes aren't an issue.
You've got to make sure that absolutely no one else can touch the computer. If you overclock the machine and, down the road, someone else decides to double the amount of RAM in the machine, for example, you're VERY likely to run into issues. With a stock-clocked system, upgrades and changes aren't an issue.
Way to resurrect a thread that has been dead for over a month...
Oh wow, my apologies. I didn't even notice. Not sure how I got here, since I usually stick to the first page anyway. I just had a hard time leaving a thread like this with misinformation at the end.