| 1 |
Richard Fish wrote: |
| 2 |
|
| 3 |
> On 3/2/06, Marton Gabor <gabor.marton@××××××××××.hu> wrote: |
| 4 |
>>let's say a ~100MB /boot in RAID1, 512MB swap not in RAID on every disk, |
| 5 |
|
| 6 |
Actually, if you make 512MB "non-raid" swap on each disk with equal |
| 7 |
priority, its like having swap on raid0 (it will be stripped over |
| 8 |
swap-partitions on all disks). But disadvantage is, that if swap |
| 9 |
is used and some of your disks fails, your system probably crushes |
| 10 |
and will have to be restarted. If stability is your concern, you could |
| 11 |
maybe think about swap on raid1. In such a case you would survive |
| 12 |
disk failure even if swap had been already used (because it is be |
| 13 |
mirrored too). |
| 14 |
|
| 15 |
> You should also consider what kind of IO throughput you require from |
| 16 |
> this system. RAID5 will require an IO to every drive for each write |
| 17 |
> operation. Additionally, reads can only be satisfied by a single |
| 18 |
> drive. This means your write performance will max out at around |
| 19 |
> 33MB/s, and reads will max out at the speed of the disks (70MB/s |
| 20 |
> typical today) |
| 21 |
|
| 22 |
Frankly, I dont understand this. Why should the write speed be so |
| 23 |
degraded? If you have 4 disks in raid5, and you want to write |
| 24 |
1.5 GB of data, you actually write 500MB on disk1, 500MB on disk2, |
| 25 |
500MB on disk3 and 500MB on disk4 (1.5 GB data + 0.5 GB parity). |
| 26 |
And because they are sata-disks, they do not share i/o channel, |
| 27 |
as 2 pata-disks on one cable. In other words, write operations |
| 28 |
are parallel. There is of course some overhead caused by parity |
| 29 |
calculation and synchronisation, but with today's cpu it is not |
| 30 |
problem. I'm sure with 4 todays equal sata disks read/write speed |
| 31 |
of raid5-array would be much higher... |
| 32 |
|
| 33 |
> either RAID1 set, so you should easily be able to saturate the bus |
| 34 |
> bandwidth at 132MB/s. |
| 35 |
|
| 36 |
Nope. Today disks controllers are not attached to southbridge |
| 37 |
through pci, but rather through a few pci-express lines - 2, 4, or |
| 38 |
even more, depending on mobo configuration. For example nForce4 has |
| 39 |
20 pci-express flexible lines, it means mobo-producers can use them |
| 40 |
as they want, but most cheap boards have 2pci-express lines |
| 41 |
assigned to sata disk controller). |
| 42 |
|
| 43 |
FYI, peak transfer rates: |
| 44 |
pci-express x1 = ~500MB/s unencoded data rate (1st gen. 250MB/s), |
| 45 |
33MHz pci = 133 MB/s |
| 46 |
|
| 47 |
And moreover unlike pci, pci-express is bi-directional, at the same |
| 48 |
time data can be read/written... |
| 49 |
|
| 50 |
But the other (and rather sad) thing is pci-express and sata-II/NCQ |
| 51 |
support in linux... :-( |
| 52 |
|
| 53 |
Jarry |
| 54 |
-- |
| 55 |
gentoo-user@g.o mailing list |
Archives