- Best way to use enum with classes - 8 Updates
- IMHO, some works of art... - 2 Updates
- Your eternal soul is at stake - 1 Update
| David Brown <david.brown@hesbynett.no>: Oct 26 10:39PM +0200 On 26/10/16 17:58, Jerry Stuckle wrote: >> it does not make sense to talk about "using RAID for backup" - except in >> the context of "don't do that". > No, YOU'VE been talking about RAID. And that is it's major usage. Are you really trying to claim that RAID is about backup? Or are you simply using the wrong words, when what you mean is availability? The primary function of most RAID setups is to minimise down-time - your storage continues to work correctly (though possibly more slowly) even if you have a disk failure. That is what the R for /redundancy/ means. It does not do /backup/, which gives you old versions of your files if something happens to your current versions (like corruption by virus, or user error), if there is a major disaster (fire in the server room), or if you simply need to refer to older versions for some reason. > I do > agree, however, that it can be used for other things, such as faster > data access. Typical benefits of RAID can include faster throughput for streamed reads or writes, faster access times for small reads or writes, and bigger volumes. More specialist uses may include continuous off-site replication of data. > And when that is necessary, once again RAID controllers > outperform RAID emulation because they are dedicated to data access. No, they don't outperform software raid. Hardware raid controllers are usually significantly slower. Modern cpus are simply much faster than the devices used on hardware raid controllers, and they usually have so many cores and bandwidth that it is not a bottleneck. Remember, the reason you are reading or writing data to or from the disk is that the data is needed in an application - the data has to pass through the cpu anyway. Now, if you have a large number of disks and have a layered setup with a RAID (or linear concatenation) on top of a set of RAID-1 pairs, then there is a good argument for having hardware raid controllers handle the mirroring. That will half the IO traffic through the cpu when writing (it makes no difference for reading, nor a difference to cache usage or memory costs). But you are probably cheaper by getting a slightly more powerful motherboard (say, a dual-socket board) or better cpus with higher bandwidth. If you are using SSD's, you definitely want to avoid hardware raid controllers - the extra latency they add will swamp all those lovely low-latency access times you paid for. Some will say that hardware raid cards with big caches and a battery unit make them faster. But those can cost more than just buying a UPS for the server, and you can add a hundred times the memory to the server for cache usage for the same price. > They don't need to steal cycles from the processor to access the data, > or wait for the processor to have time to do their work. The processor has more than enough cycles to spare. And if it doesn't, it is more cost-effective to buy more or faster cpus. >> raid, which is also true. > Not when fast access is important. And not for data which is required > 24/7 without fail. Too many things happen on the link. Obviously. It is a choice some companies make, for at least some of their data. Not all data has to be accessed at high speeds, and for many companies they are so reliant on Internet access that adding one more dependency is a minor issue. > And "may have meant"? Based on what facts? Oh, I forgot. You don't > have any facts. Ian mentioned Oracle's cloud storage - I said how I interpreted that reference. >> confusion about raid and backup again? > Good cloud storage providers use RAID for backup. They could be held > responsible for data loss in the case of a disk crash, for instance. Again, they use raid (usually software raid - either an explicit raid layer such as Linux md raid, or through a filesystem such as ZFS) to keep their systems running in the case of a disk crash. It is not backup. They will also have backup, mostly to protect against user errors but also to deal with more serious disasters. > over distance, network speeds are slower than disk. How many companies > do you know of have 100Gb links from their site to the storage provider, > and can maintain those speeds, for instance. I don't think you understand what I wrote. Until you learn the difference between raid and backup, there is little point in trying to explain about technologies such as DRBD or iSCSI. |
| Jerry Stuckle <jstucklex@attglobal.net>: Oct 26 05:55PM -0400 On 10/26/2016 4:11 PM, David Brown wrote: > systems for individual customers' needs. IBM and Dell certainly have > plenty of experts - but customers normally don't get to interact with > them, even with million dollar budgets. I KNOW they know a lot more about RAID than you do. And at IBM, they don't just read sales literature. They have to understand the products they sell. Not that you would know. >> Sure - there are dozens of ways to emulate a raid with them. I never >> said there wasn't. > Those are not emulations. (I know I am flogging a dead horse here.) I can run QEMU on my system here and load and execute code for a Cortex A-9 processor. So according to you I have an ARM processor on my laptop. But I should quit trying to teach the pig to sing. > is probably all right for many uses even if it is significantly more > expensive and often slower than a more optimised software raid setup > could be. Exactly. You don't need to change it, so your argument is as full of crap as you are. However, I know you need to use RAID emulation because you can't select the appropriate RAID device in the first place. >> RAID. You don't understand the difference, though. > You really are clueless, aren't you? "Small users sometimes use Linux". > I hope you aren't trying to say that /only/ small users use Linux! I never said anything of the sort. Your logic is as fallacious as the rest of your arguments. > optimisations or modifications, and use the standard software raid > options from those OS's. They run Linux with md raid, or FreeBSD with > gmirror, or ZFS or (on a few types) btrfs. Sure. But the code is in flash or other ROM, and not mixed in with user code - as I said. A huge difference. > level). Not that that knowledge helps me know what types of RAID and > disk setup might be suitable choices for a given workload, but I like to > know what is going on under the bonnet. No, you haven't designed anything, have you, David. But you are an expert on EVERYTHING! ROFLMAO! > of the drives haven't also died. Getting another generic computer is > easy - getting a specific hardware raid card can be hard if it is an old > or unusual device. Gee, that would be exactly the same thing you would do if your RAID controller died. And it's not so easy to just boot an OS for an old or unusual computer on a new one. So once again your argument is full of crap. > built-in raid, or whatever. When a virus attacks the file, it writes as > a client. Assuming it has the right permissions, your raid system will > do what it is designed to do - store the damaged file on the disks. Again you have no idea. Viruses can be spread throughout the system. There are many instances, especially recently, of entire networks being infected - including servers. But the viruses cannot destroy the partitioning, directories, etc. disk on a RAID device. It can corrupt data, but only data allowed by file permissions. > And since I use btrfs on most systems now, I would then find the last > undamaged version of the file from my hourly snapshots. For older > (pre-btrfs) systems, my snapshots are nightly. You can do the same on RAID, if you wish. > think is for backup? Watch the hardware raid card happily make exactly > the same copies of the damaged file as the software raid system does, > and then realise that you maybe should have had backups after all? Sure - but only the files my current ID has access to. It cannot, for instance, destroy the backups because this ID does not have permission to access those backups. -- ================== Remove the "x" from my email address Jerry Stuckle jstucklex@attglobal.net ================== |
| David Brown <david.brown@hesbynett.no>: Oct 26 11:57PM +0200 On 26/10/16 18:45, Christian Gollwitzer wrote: > electrodynamics, there is only one potential phi, and voltages are > simply the differences of that phi at different points in space. Because > the potential drops off at a rate of 1/r from static charges, one can Is it not 1/r² drop-off from a point charge? Or would that be the drop-off of the electric field, and the potential is the integral of that (in which case I understand the 1/r factor)? > between one of it's connectors and an infinitely distant point? The > answer is U/2, half of the battery voltage. > Christian I suppose the potential phi of an object is the energy it takes to move one Coulomb of charge from an infinite distance to the neutral object? |
| Jerry Stuckle <jstucklex@attglobal.net>: Oct 26 05:58PM -0400 On 10/26/2016 3:04 PM, Ian Collins wrote: > On 10/27/16 12:47 AM, Jerry Stuckle wrote: >> No, people who need high reliability backup use hardware RAID. > RAID != backup. No one ever said it was. However, you have to try to put words into other peoples' mouths to try to make a point. >> people know a lot more about it than you do. > So Oracle and its customers (who include the U.S. government) aren't > "people who need high reliability" or big companies? You don't know the difference there, either. In fact, you wouldn't know what high reliability is if it bit you on the arse. >> Or are you claiming you know more about it than the big companies with >> thousands of experts do? > The big companies who use Oracle? Yes, and the big companies who use real RAID devices for their critical reliability systems. But you wouldn't know what that is. -- ================== Remove the "x" from my email address Jerry Stuckle jstucklex@attglobal.net ================== |
| Jerry Stuckle <jstucklex@attglobal.net>: Oct 26 06:08PM -0400 On 10/26/2016 4:39 PM, David Brown wrote: >>> the context of "don't do that". >> No, YOU'VE been talking about RAID. And that is it's major usage. > Are you really trying to claim that RAID is about backup? I'm claiming that the major use of RAID around the world is for reliable data backup. > down-time - your storage continues to work correctly (though possibly > more slowly) even if you have a disk failure. That is what the R for > /redundancy/ means. Yes, which is a form of backup. > something happens to your current versions (like corruption by virus, or > user error), if there is a major disaster (fire in the server room), or > if you simply need to refer to older versions for some reason. Two copies of a file IS backup. Even copying a file to a different directory is a well-known method of backing up the file. > reads or writes, faster access times for small reads or writes, and > bigger volumes. More specialist uses may include continuous off-site > replication of data. Yes, and what do you think "continuous off-site replication of data" is, if it's not off-site backup. > reason you are reading or writing data to or from the disk is that the > data is needed in an application - the data has to pass through the cpu > anyway. Your proof? You don't have it, because once again you are full of crap. > memory costs). But you are probably cheaper by getting a slightly more > powerful motherboard (say, a dual-socket board) or better cpus with > higher bandwidth. Even with two disks you are halving the amount of data going through the CPU. And it is much more than just the data. It includes commands to two (or more) drives, additional wait times for seeks, etc. > If you are using SSD's, you definitely want to avoid hardware raid > controllers - the extra latency they add will swamp all those lovely > low-latency access times you paid for. Yes, if you're using RAID controllers from 20 years ago, I agree. Current controllers use processors just as fast as modern computers. In fact, some are faster because they can use RISC processors and code optimized to those processors. > unit make them faster. But those can cost more than just buying a UPS > for the server, and you can add a hundred times the memory to the server > for cache usage for the same price. Some are right. But it's more than just power and cache - a lot more. >> or wait for the processor to have time to do their work. > The processor has more than enough cycles to spare. And if it doesn't, > it is more cost-effective to buy more or faster cpus. Of course it does, when all you use it for is to play FreeCell. >> have any facts. > Ian mentioned Oracle's cloud storage - I said how I interpreted that > reference. And "may have meant"? Based on what facts? At least you admitted you don't have any facts - just an opinion. > keep their systems running in the case of a disk crash. It is not backup. > They will also have backup, mostly to protect against user errors but > also to deal with more serious disasters. Sure, they use RAID. And as I said earlier, RAID uses software. But these devices are dedicated to backups - no user code is running on them. A huge difference. > I don't think you understand what I wrote. Until you learn the > difference between raid and backup, there is little point in trying to > explain about technologies such as DRBD or iSCSI. Oh, I understand *exactly* what you wrote. But you don't understand RAID, that's obvious. -- ================== Remove the "x" from my email address Jerry Stuckle jstucklex@attglobal.net ================== |
| Ian Collins <ian-news@hotmail.com>: Oct 27 11:34AM +1300 On 10/27/16 10:58 AM, Jerry Stuckle wrote: >> RAID != backup. > No one ever said it was. However, you have to try to put words into > other peoples' mouths to try to make a point. You keep conflating the terms. >> So Oracle and its customers (who include the U.S. government) aren't >> "people who need high reliability" or big companies? > You don't know the difference there, either. Difference between what? >> The big companies who use Oracle? > Yes, and the big companies who use real RAID devices for their critical > reliability systems. Oracle engineered systems all use ZFS. -- Ian |
| Ian Collins <ian-news@hotmail.com>: Oct 27 11:52AM +1300 On 10/27/16 10:55 AM, Jerry Stuckle wrote: >> Those are not emulations. (I know I am flogging a dead horse here.) > I can run QEMU on my system here and load and execute code for a Cortex > A-9 processor. So according to you I have an ARM processor on my laptop. No, QEMU *is* an emulator. "QEMU is a generic and open source machine emulator and virtualizer." http://wiki.qemu.org/Main_Page >> gmirror, or ZFS or (on a few types) btrfs. > Sure. But the code is in flash or other ROM, and not mixed in with user > code - as I said. A huge difference. Ah, so now you agree that software RAID is RAID. Good start. > Again you have no idea. Viruses can be spread throughout the system. > There are many instances, especially recently, of entire networks being > infected - including servers. Reference (assuming you aren't referring to Windows servers)? > But the viruses cannot destroy the partitioning, directories, etc. disk > on a RAID device. It can corrupt data, but only data allowed by file > permissions. A rootkit can destroy everything. -- Ian |
| Tim Rentsch <txr@alumni.caltech.edu>: Oct 26 04:05PM -0700 > [...] I will add to the usual saying. Engineers defer only to > physicists, physicists defer only to mathematicians, and mathematicians > defer only to God. You left off doctors. ;) |
| "Chris M. Thomasson" <invalid@invalid.invalid>: Oct 26 03:21PM -0700 On 10/25/2016 6:00 PM, Chris M. Thomasson wrote: >> https://youtu.be/FaHJOkOrfNo >> I can code RCU in C++! Nice... > Well, I should say, we can code a user-space RCU impl. Check is out: https://groups.google.com/d/topic/lock-free/X3fuuXknQF0/discussion (please, read all if interested!) Proxy GC is RCU in nature... ;^) |
| "Chris M. Thomasson" <invalid@invalid.invalid>: Oct 26 03:22PM -0700 On 10/26/2016 3:21 PM, Chris M. Thomasson wrote: > https://groups.google.com/d/topic/lock-free/X3fuuXknQF0/discussion > (please, read all if interested!) > Proxy GC is RCU in nature... My code can be found here wrt this subject of proxy gc: http://pastebin.com/f71480694 |
| "Rick C. Hodgin" <rick.c.hodgin@gmail.com>: Oct 26 05:22AM -0700 On Tuesday, October 25, 2016 at 5:13:45 PM UTC-4, Daniel wrote: > Do a random act of kindness for a stranger. Your comment implies I don't already. You do not know me, Daniel. You cannot make that assessment. > Do something nice for your wife. Ibid. :-) > Answer a question on comp.lang.c++ :-) I've answered a few. My knowledge of c++ is lacking significantly because I don't really care for c++ in whole, just in part. I think it gets to be too complex and the source code looks like gobbly-gook when people code too deeply in c++. I've currently mapped out a new programming language called CAlive which is designed to inject some new life into C. It introduces the class, and some new things I've created. I'm here mostly to learn more about c++ as I do have interest in it, just not as much as other things. And, I do use a c++ compiler to compile my C code, which really seems to offend some of the people in the comp.lang.c group. But, I prefer the stricter typing, and relaxed syntaxes of c++. Best regards, Rick C. Hodgin |
| You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page. To unsubscribe from this group and stop receiving emails from it send an email to comp.lang.c+++unsubscribe@googlegroups.com. |
No comments:
Post a Comment