Hi all, For those who have lots of experience with this: I need more hard drive space and a faster disc too. Assuming software based raid is reliable enough for a home user, and money is very tight, I have two options: 1. Buy a new fast disc 2. Use kernel software raid to mirror the current Western Digital Caviar 5400rpm 2mb cache 10gb drive with a 10gb partition on a new (but now very cheap) 5400rpm 40gb drive. The remaining 30gb on the new disc would be /home Which setup would be faster? Thanks Hans
For those who have lots of experience with this:
not lots, but i'll give it a shot.. :)
I need more hard drive space and a faster disc too. Assuming software based raid is reliable enough for a home user, and money is very tight, I have two options:
1. Buy a new fast disc
2. Use kernel software raid to mirror the current Western Digital Caviar 5400rpm 2mb cache 10gb drive with a 10gb partition on a new (but now very cheap) 5400rpm 40gb drive. The remaining 30gb on the new disc would be /home
Which setup would be faster?
by my understanding (and assumption that the "new fast disc" is not a 10k+ rpm drive) as long as the two drives are on separate ide channels, the raid-0 setup should be faster. if the two drives are on the same ide channel, there will be contention for the channel (as an ide channel can only talk to one device at a time, unlike scsi) which will drag down preformance. It still might be faster than the single fast disc, but not by much. take this with a grain of salt tho as i've only set up 2 raid arrays and they were on scsi systems purely for redundancy, not performance (raid-1 vs raid-0). -- trey
On Mon, 2003-09-29 at 20:36, Trey Gruel wrote:
by my understanding (and assumption that the "new fast disc" is not a 10k+ rpm drive) as long as the two drives are on separate ide channels, the raid-0 setup should be faster. That would be a normal 7200rpm 2mb cache IDE disc.
I've done this ony once so far. It was a Pentium with a 300mb disc and a 400. I made three partitions on the 400 mb disc - 10mb for /boot, 90mb for swap and the remaining 300 to go into the raid-0 with the other disc. This was in part for speed and in part to have enough space since I didn't really know how to devide the partitions for optimal usage. It took the machine (running RH 7.3 with 32mb ram at time time) from barely useable with a lot of patience (and time to attend to my coffee pot :-) to just normal barely useable. Then, those drives were what slow is all about... I've read a Tomshardware benchmark where two 4200rpm notebook ide hard discs in a raid-0 were compared to something else. Can't remember exactly what but it was at least a 7200rpm IDE drive. The two notebook drives outperformed the single disc and were a lot more quiet too. So I guess two 5400rpm ata100 drives should faster too. I should have thought of that before I wrote. Thanks for replying Hans
The 03.09.29 at 20:15, H du Plooy wrote:
2. Use kernel software raid to mirror the current Western Digital Caviar 5400rpm 2mb cache 10gb drive with a 10gb partition on a new (but now very cheap) 5400rpm 40gb drive. The remaining 30gb on the new disc would be /home
Which setup would be faster?
A mirror is not faster: write operations have to be done on both hard disks. At best, it will be the same. Read opertions could be faster, perhaps. A raid 5 setup (which is not a mirror) of three disks on three separate cables (IDE) should be faster. Another solution: distribute your partitions. For example, you can have /opt in one disk, and the rest on the other: X environment and applications start faster. Swap partitions can also be distributed. -- Cheers, Carlos Robinson
On Sat, 2003-10-04 at 00:31, Carlos E. R. wrote:
The 03.09.29 at 20:15, H du Plooy wrote:
2. Use kernel software raid to mirror the current Western Digital Caviar 5400rpm 2mb cache 10gb drive with a 10gb partition on a new (but now very cheap) 5400rpm 40gb drive. The remaining 30gb on the new disc would be /home
Which setup would be faster?
A mirror is not faster: write operations have to be done on both hard disks. At best, it will be the same. Read opertions could be faster, perhaps. A raid 5 setup (which is not a mirror) of three disks on three separate cables (IDE) should be faster.
Another solution: distribute your partitions. For example, you can have /opt in one disk, and the rest on the other: X environment and applications start faster. Swap partitions can also be distributed.
I'm sorry, I confused two things when writing my question. I meant (hope I have the number right!) Raid-0 i.e. two partitions on two seperate discs as one of double the size - and hopefully double the speed. What I meant to write is that I would not need or want to mirror the disc, as most of the stuff on the disc is replaceable: music collection which I own all the CDs of, e-mail foder which I back up to CD once or twice a week, same for docs and pics, the rest is just the OS installation and settings which would take some time to set up again, but no big loss if something breaks. Thing is I'm not thinking of buying an PCI Raid controller - for the price difference I could just as well get a 7200rpm 8mb cache drive. What I would really like to know is how stiping (sp?) the 5400rpm discs with kernel based software raid would perform compared to a new 7200rpm disc. The ones I've built into new machines lately - mostly the standard Seagate 7200rpm 2mb cache models - are really much faster than the drive I have. Thanks for the suggestions. Distributing partitions over discs is another option, but finding a good solution will take time. Which filesystems will get used more? I assume with local mail delivery + spamassasin, /var would get used heavily, and I use a fair bit of swap. Put them on seperate discs, or both on the faster of the two drives? Or shall I use the slower disc for the filesystems that get read more often and the faster disc for the filesystems that get written to more often? I have a feeling the time I would spend figuring out the best partition sceme wouldn't be worth the money I save. It is definitely something that I'd like to get to know inside out, but I can't afford the time for that just now. Thanks for the ideas! Hans
The 03.10.04 at 17:10, H du Plooy wrote:
I'm sorry, I confused two things when writing my question. I meant (hope I have the number right!) Raid-0 i.e. two partitions on two seperate discs as one of double the size - and hopefully double the speed.
Ah! That's different. From the howto: · RAID-0 · Also called ``stripe'' mode. The devices should (but need not) have the same size. Operations on the array will be split on the devices; for example, a large write could be split up as 4 kB to disk 0, 4 kB to disk 1, 4 kB to disk 2, then 4 kB to disk 0 again, and so on. If one device is much larger than the other devices, that extra space is still utilized in the RAID device, but you will be accessing this larger disk alone, during writes in the high end of your RAID device. This of course hurts performance. · Like linear, there is no redundancy in this level either. Unlike linear mode, you will not be able to rescue any data if a drive fails. If you remove a drive from a RAID-0 set, the RAID device will not just miss one consecutive block of data, it will be filled with small holes all over the device. e2fsck or other filesystem recovery tools will probably not be able to recover much from such a device. · The read and write performance will increase, because reads and writes are done in parallel on the devices. This is usually the main reason for running RAID-0. If the busses to the disks are fast enough, you can get very close to N*P MB/sec.
Thing is I'm not thinking of buying an PCI Raid controller - for the
The performance is similar to a software raid; they are not really hardware only raid systems, from what I read on this list some time ago.
price difference I could just as well get a 7200rpm 8mb cache drive. What I would really like to know is how stiping (sp?) the 5400rpm discs with kernel based software raid would perform compared to a new 7200rpm disc. The ones I've built into new machines lately - mostly the standard Seagate 7200rpm 2mb cache models - are really much faster than the drive I have.
Probably the 7200 rpm disk alone would be faster even without raid :-)
Thanks for the suggestions. Distributing partitions over discs is another option, but finding a good solution will take time. Which filesystems will get used more? I assume with local mail delivery + spamassasin, /var would get used heavily, and I use a fair bit of swap.
If you use swap, my guess is that having it on a completely different HD, or one seldom used for other things, would make the system faster. Or, you can distribute one swap partition on each HD, and have them all with the same priority: the kernel will distribute operations on them all. From the howto: 2.5. Swapping on RAID There's no reason to use RAID for swap performance reasons. The kernel itself can stripe swapping on several devices, if you just give them the same priority in the fstab file. By the way, read that howto: Software-RAID-HOWTO.gz The Software-RAID HOWTO Jakob Østergaard (jakob@unthought.net) v0.90.8, 2002-08-05
Put them on seperate discs, or both on the faster of the two drives? Or shall I use the slower disc for the filesystems that get read more often and the faster disc for the filesystems that get written to more often?
Er... I'm getting lost O:-)
I have a feeling the time I would spend figuring out the best partition sceme wouldn't be worth the money I save. It is definitely something that I'd like to get to know inside out, but I can't afford the time for that just now.
There is a section on this on one of the SuSE manuals. I like to separate the '/opt' partition to another HD: gnome, kde, and a few more things, like OO, are in there, while Xfree and the rest is mostly under '/usr'. Thus, when starting up kde, somethings will be read from '/opt' and some others from '/usr': the speedup gain is noticiable. Of course, it seems that raid 0 would be faster for everything, but at the same time, it needs some processing time overhead, and it is more vulnerable. -- Cheers, Carlos Robinson
> -----Original Message----- > From: Carlos E. R. [mailto:robin1.listas@tiscali.es] > Sent: Wednesday, October 08, 2003 2:11 PM > To: suse-linux-e@suse.com > Subject: Re: [SLE] Raid 0 speed question > > > > price difference I could just as well get a 7200rpm 8mb > cache drive. > > What I would really like to know is how striping (sp?) the 5400rpm > > discs with kernel based software raid would perform > compared to a new > > 7200rpm disc. The ones I've built into new machines lately > - mostly > > the standard Seagate 7200rpm 2mb cache models - are really > much faster > > than the drive I have. > > Probably the 7200 rpm disk alone would be faster even without raid :-) > If this is any help, I can tell you my experience with striped drives. I used striping a number of years back with Windows 2000 Pro. I had a 10GB and a 15GB drive so I partitioned the 5 of the 15 to the OS and swap file and striped the remaining 10GB with the 10GB drive. The result was much faster disk access on those things in the stripe. It was noticeable but not twice as fast. I quit using striping when the 10GB drive died. (Maxtor had to replace 8 drives in 2 years for me. A bad spell for some reason.) As for speed of mixed speed drives, I believe you will be hurting your speed by mixing a high speed with a slower speed HDD in striping. The 7200 speed drive will share write time with the 5400 resulting an average between the two drives. But it will leave you with a common average speed. Just for the fun of it, I did stripe three HDDs once. Now that was worth the effort, but with a CDR and DVD, I can't have 3 HDDs installed. I have been reading about RAID in Linux and see that with 3 HDDs, you can get the striping with error correction so you can get both speed and security. I can't tell you the best way to go, that's your decision. I am guessing that if it were me, I would go with the faster drive and leave open the possibility of expanding with a second and third drive in the future. That's where I would stripe them, giving me the security of the RAID 5 (I believe) at just less than 3x the speed of one drive. All this talk made me look at a pair of 20GB drives sitting next to me. Maybe striping them on this old IBM I am using to learn Linux on would be fun. lol Good luck Buck
On Thu, 2003-10-09 at 02:00, Buck wrote:
(Maxtor had to replace 8 drives in 2 years for me. A bad spell for some reason.) Saw the same thing here, over and over - I don't trust them any more.
I believe you will be hurting your speed by mixing a high speed with a slower speed HDD in striping. The 7200 speed drive will share write time with the 5400 resulting an average between the two drives. I won't do that much different - both 5400rpm, one with 2mb cache, one with 512k. 512k model is two years newer, and actually perform very similarly to the 2mb (but older) model.
But it will leave you with a common average speed. I don't think so, the slower drive can't raise it's own performance to halfway between itself and the faster one. The faster drive would be forced to wait for the slower one, so it would be like having two of the slower drive.
Hans
On Wed, 2003-10-08 at 20:10, Carlos E. R. wrote:
Put them on seperate discs, or both on the faster of the two drives? Or shall I use the slower disc for the filesystems that get read more often and the faster disc for the filesystems that get written to more often?
Er... I'm getting lost O:-)
Of course, it seems that raid 0 would be faster for everything, but at the same time, it needs some processing time overhead, and it is more vulnerable. Vulnerability is not really a worry, there won't be anything on that's
Sorry, I'm mixing all sorts of stuff together. I have two options: 1. one 7200rpm 2mb cache drive. Normal partitioning sceme, nothing exciting. 2. one 5400rpm 2mb cache 10gb drive, 10gb, one 5400rpm 512kb cache 40gb drive (a fair bit slower). Two sub options: a) 10gb drive and 10gb partition on 40gb drive in stripe mode raid. Linux on the resulting raid device, windows and space for goodies on the remaining non-raid 30gb. Linux wouldn't be using the non-raid part of the drive, so that shouldn't be a factor. b) linux partitions distributed over the two discs. In 2a) my question is: I have identified the two partitions that would be used most heavily - /var and swap. Should I put these both on the faster drive, or on seperate drives. After reading your reply (the part about /usr and /opt, I think those two on different drives would be better. By the way, your idea for /usr and /opt is pretty good - I'll definitely keep that in mind! perticularly valuable or irreplaceable. Just the time lost having to re-install, but that I do in an evening. No big deal. Processing power is another thing, but having done software raid on a P-I 133 with the same class disc, I'm not too worried. The overhead there certainly didn't present any problems, so it shouldn't on this one. Thanks for the replies... Hans
The 03.10.10 at 09:40, H du Plooy wrote:
In 2a) my question is: I have identified the two partitions that would be used most heavily - /var and swap. Should I put these both on the faster drive, or on seperate drives. After reading your reply (the part about /usr and /opt, I think those two on different drives would be better.
Both together on a diferent disk than the rest of the system, I think, and probably on the faster disk, if that is where most of your write operations will be.
By the way, your idea for /usr and /opt is pretty good - I'll definitely keep that in mind!
Thanks, but it is not mine: I read it on the SuSE manual years ago :-) Ah, by the way: it is possible to have /usr on a separate partition, but I found that sometimes -- when init sequence detects errors on the disk and doesn't mount anything but "/" -- that some programs fail to run. If those programs include things like fsck, you are hoosed and need the rescue CD. This happened to me. Just to clarify, I meant having '/opt' in a partition on a diferent HD than the partition where directory '/usr' resides (on its own, or as part of '/usr').
Of course, it seems that raid 0 would be faster for everything, but at the same time, it needs some processing time overhead, and it is more vulnerable. Vulnerability is not really a worry, there won't be anything on that's perticularly valuable or irreplaceable. Just the time lost having to re-install, but that I do in an evening. No big deal.
Have a good backup, or an image.
Processing power is another thing, but having done software raid on a P-I 133 with the same class disc, I'm not too worried. The overhead there certainly didn't present any problems, so it shouldn't on this one.
Thanks for the replies...
Welcome! -- Cheers, Carlos Robinson
On Fri, Oct 10, 2003 at 02:01:23PM +0200, Carlos E. R. wrote:
The 03.10.10 at 09:40, H du Plooy wrote:
In 2a) my question is: I have identified the two partitions that would be used most heavily - /var and swap. Should I put these both on the faster drive, or on seperate drives. After reading your reply (the part about /usr and /opt, I think those two on different drives would be better.
Both together on a diferent disk than the rest of the system, I think, and probably on the faster disk, if that is where most of your write operations will be.
By the way, your idea for /usr and /opt is pretty good - I'll definitely keep that in mind!
Thanks, but it is not mine: I read it on the SuSE manual years ago :-)
Ah, by the way: it is possible to have /usr on a separate partition, but I found that sometimes -- when init sequence detects errors on the disk and doesn't mount anything but "/" -- that some programs fail to run. If those programs include things like fsck, you are hoosed and need the rescue CD. This happened to me.
Just to clarify, I meant having '/opt' in a partition on a diferent HD than the partition where directory '/usr' resides (on its own, or as part of '/usr').
Of course, it seems that raid 0 would be faster for everything, but at the same time, it needs some processing time overhead, and it is more vulnerable. Vulnerability is not really a worry, there won't be anything on that's perticularly valuable or irreplaceable. Just the time lost having to re-install, but that I do in an evening. No big deal.
Have a good backup, or an image.
How do you folks recommend getting "an image" of an entire system for quick and easy restore?
Processing power is another thing, but having done software raid on a P-I 133 with the same class disc, I'm not too worried. The overhead there certainly didn't present any problems, so it shouldn't on this one.
Thanks for the replies...
The 03.10.10 at 12:52, Jim Norton wrote:
Have a good backup, or an image.
How do you folks recommend getting "an image" of an entire system for quick and easy restore?
I don't -- I still have to try one or two solutions (mentioned here) on my to-do list O:-) -- Cheers, Carlos Robinson
On Fri, 2003-10-03 at 18:31, Carlos E. R. wrote:
The 03.09.29 at 20:15, H du Plooy wrote:
2. Use kernel software raid to mirror the current Western Digital Caviar 5400rpm 2mb cache 10gb drive with a 10gb partition on a new (but now very cheap) 5400rpm 40gb drive. The remaining 30gb on the new disc would be /home
Which setup would be faster?
A mirror is not faster: write operations have to be done on both hard disks. At best, it will be the same. Read opertions could be faster, perhaps. A raid 5 setup (which is not a mirror) of three disks on three separate cables (IDE) should be faster.
Raid 5 is rarely if ever faster than anything. For reading the basic formula for i/o's per second for any form of raid is: ios = 1; efficiency = 1 / ios; ios per disk * disks * efficiency So a 6 disk Raid 0 or Raid 1 or Raid 5 will all have the same read performance. It is the capacity that will vary. ==== For writing it gets more complicated, so I'm using psuedo code: switch(raid_type) Raid 0: ios = 1; break; (write data) Raid 1: ios = 2; break; (write data, write data) Raid 5: ios = 3; break; (read parity, write data, write parity) efficiency = 1 / ios; ios per disk * disks * efficiency; So for a 6 disk array: Raid 0: 6 x ios of one disk Raid 1: 3 x ios of one disk Raid 5: 2 x ios of one disk Thus Raid 5 is really only good in a very read intensive environment. Greg -- Greg Freemyer
participants (6)
-
Buck -
Carlos E. R. -
Greg Freemyer -
H du Plooy -
Jim Norton -
Trey Gruel