2x RAID-1 vs. RAID-10
Question: with four 18GB drives on a Mylex RAID controller, is it better to use two separate RAID-1 mirrored pairs (one for OS, one for dataspaces) or one RAID-10 across all four? Art Kagel answered that with only two mirrored pairs it makes little practical difference — the performance gain from striping just two pairs is minimal and rarely worth the extra administrative complexity, with benefits only really appearing from about four or five mirrored pairs upward. The rest of the thread is banter plus a definition of "checkering" (striping across multiple stripe sets).
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
What's the general opinion on two mirrors (one for OS and one for dataspaces) vs one big RAID-10? Controller: Mylex AccelRAID 353 (dual channel Ultra160) Disk: 4x IBM Ultrastar 36LZX (18GB) My initial though was that the controller could pull more out of a RAID-10 than two mirror-pairs. I don't have the time for real life testing, so even a clowns opinion would be worth something... THomas
With only 4 drives, 2 mirrored pairs, it doesnot much matter which way you go. The performance gained from striping two pairs is rarely worth the administration headache. Now if you have 4 or five pairs... Art S. Kagel Thomas Parsli wrote: > > What's the general opinion on two mirrors (one for OS and one for > dataspaces) vs one big RAID-10? > > Controller: Mylex AccelRAID 353 (dual channel Ultra160) > Disk: 4x IBM Ultrastar 36LZX (18GB) > > My initial though was that the controller could pull more out > of a RAID-10 than two mirror-pairs. > I don't have the time for real life testing, so even a clowns > opinion would be worth something... > > THomas
"Art S. Kagel" <kagel@bloomberg.net> writes: > With only 4 drives, 2 mirrored pairs, it doesnot much matter which way you > go. The performance gained from striping two pairs is rarely worth the > administration headache. Now if you have 4 or five pairs... Yes, my RAID-arrays _are_ smaller then yours... Thomas
Thomas Parsli wrote: > > "Art S. Kagel" <kagel@bloomberg.net> writes: > > > With only 4 drives, 2 mirrored pairs, it doesnot much matter which way you > > go. The performance gained from striping two pairs is rarely worth the > > administration headache. Now if you have 4 or five pairs... > > Yes, my RAID-arrays _are_ smaller then yours... Let's not get testy now! If I were to laud it up I'd have mentioned that my arrays are constructed of from 10-30 drive pairs and that we actually checker 4 such arrays and carve chunks out of that. I was merely pointing out that the performance gain from a two drive (or pair) stripe is minimal and that the gains seem to tip over between 4 and 5 drive pairs. Art S. Kagel
"Art S. Kagel" <kagel@bloomberg.net> writes: > Thomas Parsli wrote: > > > > "Art S. Kagel" <kagel@bloomberg.net> writes: > > > > > With only 4 drives, 2 mirrored pairs, it doesnot much matter which way you > > > go. The performance gained from striping two pairs is rarely worth the > > > administration headache. Now if you have 4 or five pairs... > > > > Yes, my RAID-arrays _are_ smaller then yours... > > Let's not get testy now! You talkin' to me? You talkin' to me? You talkin' to me? Then who the hell else are you talkin' to? You talkin' to me? Well I'm the only one here. Who do you think you're talking to? Oh yeah? Huh? Ok. > If I were to laud it up I'd have mentioned that my > arrays are constructed of from 10-30 drive pairs and that we actually > checker 4 such arrays and carve chunks out of that. I was merely pointing out > that the performance gain from a two drive (or pair) stripe is minimal > and that the gains seem to tip over between 4 and 5 drive pairs. I got that, really:) Thomas (yes, the quote is from Taxi Driver -I'm not [that] insane)
Thomas Parsli wrote in message ... >"Art S. Kagel" <kagel@bloomberg.net> writes: > >> Thomas Parsli wrote: >> > >> > "Art S. Kagel" <kagel@bloomberg.net> writes: >> > >> > > With only 4 drives, 2 mirrored pairs, it doesnot much matter which way you >> > > go. The performance gained from striping two pairs is rarely worth the >> > > administration headache. Now if you have 4 or five pairs... >> > >> > Yes, my RAID-arrays _are_ smaller then yours... Oh, poor baby. Let's not compare now gentleman.. >> >> Let's not get testy now! > >You talkin' to me? You talkin' to me? You talkin' to me? >Then who the hell else are you talkin' to? You talkin' to me? >Well I'm the only one here. Who do you think you're talking to? Oh yeah? >Huh? Ok. > >> If I were to laud it up I'd have mentioned that my >> arrays are constructed of from 10-30 drive pairs and that we actually >> checker 4 such arrays and carve chunks out of that. I was merely Checker?? pointing out >> that the performance gain from a two drive (or pair) stripe is minimal >> and that the gains seem to tip over between 4 and 5 drive pairs. > >I got that, really:) > >Thomas >(yes, the quote is from Taxi Driver -I'm not [that] insane) What kind of neighbourhood do you live in if even the taxi drivers talk like that?
smooth1 wrote: > > Thomas Parsli wrote in message ... > >"Art S. Kagel" <kagel@bloomberg.net> writes: > > > >> Thomas Parsli wrote: > >> > > >> > "Art S. Kagel" <kagel@bloomberg.net> writes: [SNIP] > > > >> If I were to laud it up I'd have mentioned that my > >> arrays are constructed of from 10-30 drive pairs and that we actually > >> checker 4 such arrays and carve chunks out of that. I was merely > > Checker?? [SNIP] Checkering: A term coined to describe the product of creating a stripe set from multiple stripe sets. (I didn't coin it, so don't blame me.) Art S. Kagel