1 |
It depends. I'd like to see SW RAID keep up with dual 2GB fibre |
2 |
attached to an IBM FAStT600. Then again, I haven't been given a budget |
3 |
to test this with a Linux server. :) |
4 |
|
5 |
Also remember which RAID you're talking about. If you need parity, see |
6 |
my previous post about BAARF. |
7 |
|
8 |
Finally, saying the CPU is not usually the bottleneck in a database |
9 |
server is misleading at best. Being mainly an Oracle DBA these days, |
10 |
faster CPUs can *cause* CPU contention. Like I said before, "it |
11 |
depends". And like everything else, YMMV. |
12 |
|
13 |
Rich |
14 |
|
15 |
-----Original Message----- |
16 |
From: Sean Cook [mailto:scook@×××××.net] |
17 |
Sent: Wednesday, February 01, 2006 3:39 PM |
18 |
To: gentoo-server@l.g.o |
19 |
Subject: Re: [gentoo-server] RAID |
20 |
|
21 |
|
22 |
-----BEGIN PGP SIGNED MESSAGE----- |
23 |
Hash: SHA1 |
24 |
|
25 |
kashani wrote: |
26 |
> Sean Cook wrote: |
27 |
> > Redhat In their manual: |
28 |
> |
29 |
>> "With today's fast CPUs, Software RAID performance can excel against |
30 |
>> Hardware RAID." |
31 |
>> |
32 |
>> As I stated before, it depends on where your bottleneck is... if you |
33 |
>> are not cpu bound, software raid is great! and will boost IO through |
34 |
>> put on comparable hardware. If you are already CPU bound, forget |
35 |
>> software raid, it will degrade your system to a crawl... |
36 |
> |
37 |
> |
38 |
> Badly done tests circa 1998 without any sort of methodology, mention |
39 |
of |
40 |
> cluster sizes, etc is proof than any idiot can make a computer slower. |
41 |
> |
42 |
> I'll argue that a fully supported hard raid card is always |
43 |
superior |
44 |
> to a software raid by it's very nature, having local I/O cache and a |
45 |
> dedicated chip. However there are definitely workloads where a |
46 |
software |
47 |
> raid is good enough that spending money on a hardware raid card is |
48 |
> pointless. I can not imagine a case where all things being equal that |
49 |
> software raid would be measurably faster. |
50 |
> |
51 |
> In the event that removing your RAID card makes your disk 5x |
52 |
faster |
53 |
> I'd also recommend removing the admin who setup the original system as |
54 |
> well. :-) |
55 |
> |
56 |
> kashani |
57 |
|
58 |
My post did say "Back in the Day" and it was around 1999 that we did |
59 |
this however, then I was running unstable raid tools and they have come |
60 |
a long way, and I was running 2.2 kernel. |
61 |
|
62 |
However, the only people I am aware of that say the performance is |
63 |
better on hardware raid are hardware raid manufacturers and you... You |
64 |
don't work for LSI do you? Most of the linux software folks agree with |
65 |
me or take a mildly more conservative tone. |
66 |
|
67 |
eg: |
68 |
|
69 |
Redhat Enterprise Linux (circa 2005-2006) |
70 |
http://www.redhat.com/docs/manuals/enterprise/RHEL-4-Manual/pdf/rhel-isa |
71 |
-en.pdf |
72 |
|
73 |
Often the excess CPU power available for software RAID parity |
74 |
calculations greatly exceeds the processing power present on a RAID |
75 |
controller card. Therefore, some software RAID implementations |
76 |
actually have the capability for higher performance than hardware RAID |
77 |
implementations. |
78 |
|
79 |
|
80 |
|
81 |
Mysql: |
82 |
|
83 |
Hardware Versus Software |
84 |
|
85 |
Some operating systems can perform software RAID. Rather than buying a |
86 |
dedicated RAID controller, the operating system's kernel splits the I/O |
87 |
among multiple disks. Many users shy away from using these features |
88 |
because they've long been considered slow or buggy. |
89 |
|
90 |
In reality, software RAID is quite stable and performs rather well. The |
91 |
performance differences between hardware and software RAID tend not to |
92 |
be significant until they're under quite a bit of load. For smaller and |
93 |
medium-sized workloads, there's little discernible difference between |
94 |
them. Yes, the server's CPU must do a bit more work when using software |
95 |
RAID, but modern CPUs are so fast that the RAID operations consume a |
96 |
small fraction of the available CPU time. And, as we stressed earlier, |
97 |
the CPU is usually not the bottleneck in a database server anyway. |
98 |
|
99 |
-- |
100 |
gentoo-server@g.o mailing list |