Re: Blogbench RAID benchmarks
--0016363b7be26f17a104a89c3802
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
On Thu, Jul 21, 2011 at 3:35 PM, Freddie Cash <fjwcash@gmail.com> wrote:
> On Mon, Jul 18, 2011 at 7:06 PM, Matthew Dillon <
> dillon@apollo.backplane.com> wrote:
>
>> Ok, well this is interesting. Basically it comes down to whether we
>> want to starve read operations or whether we want to starve write
>> operations.
>>
>> The FreeBSD results starve read operations, while the DragonFly resul=
ts
>> starve write operations. That's the entirety of the difference betwe=
en
>> the two tests.
>>
>> Would using the disk scheduler's in FBSD/DFly help with this at all?
>
> FreeBSD includes a geom_sched class for enabling pluggable disk scheduler=
's
> (currently only round-robin algorithm is implemented).
> http://info.iet.unipi.it/~luigi/geom_sched/
>
> Page 39 of the presentation on GEOM_SCHED shows the following, indicating
that it should make a big difference in the blogbench results (note the
second result with greedy read and write):
Some preliminary results on scheduler=E2=80=99s performance in some easy
cases (the focus here is on the framework).
Measurement is using multiple dd instances on a =EF=AC=81lesystems, all
speeds in MiB/s.
two greedy readers, throughput improvement
NORMAL: 6.8 + 6.8 ; GSCHED RR: 27.0 + 27.0
one greedy reader, one greedy writer, capture e=EF=AC=80ect
NORMAL: R: 0.234 W:72.3 ; GSCHED RR: R:12.0 W:40.0
multiple greedy writers, only small loss of througput
NORMAL: 16+16; RR: 15.5 + 15.5
one sequential reader, one random reader (=EF=AC=81o)
NORMAL: Seq: 4.2 Rand: 4.2; RR: Seq: 30 Rand: 4.4
> And I believe DFly has dsched?
>
>
>> This is all with swapcache turned off. The only way to test in a
>> fair manner with swapcache turned on (with a SSD) is if the FreeBSD
>> test used a similar setup w/ZFS.
>>
>> ZFS includes it's own disk scheduler, so geom_sched wouldn't help in tha=
t
> case. Would be interesting to see a comparison of HAMMER+swapcache and
> ZFS+L2ARC, though.
>
> --
> Freddie Cash
> fjwcash@gmail.com
>
--=20
Freddie Cash
fjwcash@gmail.com
--0016363b7be26f17a104a89c3802
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
<div class=3D"gmail_quote">On Thu, Jul 21, 2011 at 3:35 PM, Freddie Cash <s=
pan dir=3D"ltr"><<a href=3D"mailto:fjwcash@gmail.com">fjwcash@gmail.com<=
/a>></span> wrote:<br><blockquote class=3D"gmail_quote" style=3D"margin:=
0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class=3D"gmail_quote"><div class=3D"im">On Mon, Jul 18, 2011 at 7:06 P=
M, Matthew Dillon <span dir=3D"ltr"><<a href=3D"mailto:dillon@apollo.bac=
kplane.com" target=3D"_blank">dillon@apollo.backplane.com</a>></span> wr=
ote:<br>
</div><div class=3D"im"><blockquote class=3D"gmail_quote" style=3D"margin:0=
0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
=C2=A0 =C2=A0Ok, well this is interesting. =C2=A0Basically it comes down t=
o whether we<br>
=C2=A0 =C2=A0want to starve read operations or whether we want to starve w=
rite<br>
=C2=A0 =C2=A0operations.<br>
<br>
=C2=A0 =C2=A0The FreeBSD results starve read operations, while the DragonF=
ly results<br>
=C2=A0 =C2=A0starve write operations. =C2=A0That's the entirety of the=
difference between<br>
=C2=A0 =C2=A0the two tests.<br>
<br></blockquote></div><div>Would using the disk scheduler's in FBSD/DF=
ly help with this at all?<br><br>FreeBSD includes a geom_sched class for en=
abling pluggable disk scheduler's (currently only round-robin algorithm=
is implemented).=C2=A0 <a href=3D"http://info.iet.unipi.it/%7Eluigi/geom_s=
ched/" target=3D"_blank">http://info.iet.unipi.it/~luigi/geom_sched/</a><br=
>
<br></div></div></blockquote><div>Page 39 of the presentation on GEOM_SCHED=
shows the following, indicating that it should make a big difference in th=
e blogbench results (note the second result with greedy read and write):<br=
>
<br>Some preliminary results on scheduler=E2=80=99s performance in some eas=
y<br>cases (the focus here is on the framework).<br><br>Measurement is usin=
g multiple dd instances on a =EF=AC=81lesystems, all<br>speeds in MiB/s.<br=
><br>two greedy readers, throughput improvement<br>
NORMAL: 6.8 + 6.8 ; GSCHED RR: 27.0 + 27.0<br><br>one greedy reader, one gr=
eedy writer, capture e=EF=AC=80ect<br>NORMAL: R: 0.234 W:72.3 ; GSCHED RR: =
R:12.0 W:40.0<br><br>multiple greedy writers, only small loss of througput<=
br>NORMAL: 16+16; RR: 15.5 + 15.5<br>
<br>one sequential reader, one random reader (=EF=AC=81o)<br>NORMAL: Seq: 4=
..2 Rand: 4.2; RR: Seq: 30 Rand: 4.4<br><br>=C2=A0</div><blockquote class=3D=
"gmail_quote" style=3D"margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rg=
b(204, 204, 204); padding-left: 1ex;">
<div class=3D"gmail_quote"><div>And I believe DFly has dsched?<br></div><di=
v class=3D"im"><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"=
margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204, 204, 204);padding-l=
eft:1ex">
=C2=A0=C2=A0 This is all with swapcache turned off. =C2=A0The only way to =
test in a<br>
=C2=A0 =C2=A0fair manner with swapcache turned on (with a SSD) is if the F=
reeBSD<br>
=C2=A0 =C2=A0test used a similar setup w/ZFS.<br>
<div><div></div><br></div></blockquote></div></div>ZFS includes it's ow=
n disk scheduler, so geom_sched wouldn't help in that case.=C2=A0 Would=
be interesting to see a comparison of HAMMER+swapcache and ZFS+L2ARC, thou=
gh.<br clear=3D"all">
<font color=3D"#888888">
<br>-- <br>Freddie Cash<br><a href=3D"mailto:fjwcash@gmail.com" target=3D"_=
blank">fjwcash@gmail.com</a><br>
</font></blockquote></div><br><br clear=3D"all"><br>-- <br>Freddie Cash<br>=
<a href=3D"mailto:fjwcash@gmail.com">fjwcash@gmail.com</a><br>
--0016363b7be26f17a104a89c3802--
討論串 (同標題文章)
完整討論串 (本文為第 20 之 23 篇)：