Re: Mysterious Lag With SATA Maxtor 250GB 7200RPM 16MB Cache UnderLinux using NFSv3 UDP

From: Manu Abraham
Date: Mon Jan 17 2005 - 16:38:54 EST

omes wrote:
I have the same problem as you. At least our problems are much alike. I got two Western digital hard disks. One 120GB 7200RPM 2MB Cache IDE, and one 80GB 7200 2MB Cache IDE. I get high loads when reading large files for some time, as well as when copying from one partition to another. All my partitions are in ext3, and i'm running 2.6.10.

I have a similar problem. I am running 2.6.9 on FC3 not the stock kernel, but a vanilla kernel.
I have 3 NFS shares mounted on this 2.6.9 machine (ASUS P4C800 ICH5 motherboard with 1024 MB RAM and sk98lin gigabit LOM (non-functional since thesk98lin driver broke apart) so i use an additional 3Com 3C59x 100 Mbps adapter, with a nvidia VGA adapter, but no nvidia modules but only VESAFB at 1024x768.

If i try to read/write some files from the NFS share maybe something around 10Megs or more, everything starts to crawl.. What i mean to say is not the n/w activity, but the console itself. Whatever i do later on, that effect seems to persist, without a reboot.

This looks more or less consistent, but not always. I was wondering what could be causing this. Previously i had 2.6.7 on FC2 without any problems, on another machine i have Suse 9.1 with the default kernel, without any problems.
In my case i have a 160GB Seagate SATA drive.


Manu




hdparm tests resulted in normal readings however. And when just reading small files the system looks and feels normal.

What motherboard have you got?

Regards,
- omes
On Monday 17 January 2005 21:56, you wrote:

Yes, only with NFS.

On Mon, 17 Jan 2005, Norbert van Nobelen wrote:

Only with NFS? I have a raid array of the same discs and the system just
sometimes seems to hang completely (for a second or less) and then to go
on again at a normal speed (110MB/s).
I am running a SuSE 9.1 stock kernel (2.6.5-7.111-smp) on that machine.

On Monday 17 January 2005 21:06, you wrote:

When writing to or from the drive via NFS, after 1GB or 2GB, it "feels"
like the system slows to a crawl, the mouse gets very slow, almost like
one is burning a CD at 52X under PIO mode. I originally had this disk
in my main system with an Intel ICH5 chipset (ABIT IC7-G mobo) and a
Pentium 4 2.6GHZ w/HT and I could not believe the hit the system took
when writing over NFS.

I put the drive on its own controller in another box to see if I could
replicate the problem. The drive is on a Promise 150TX2 Plus
controller. When writing locally on the disk, there is no problem. When
writing to or from the drive over NFS, the system becomes _very_ slow,
df -h lags, the system slows almost to a halt. I check /proc/interrupts
and
/var/log/messages but no errors result.

There is some type of system bottleneck when using NFS with this drive.

All the file systems are XFS on all systems.
Can anyone offer any suggestions?
I cannot explain or find the cause of this.

All systems are running 2.6.10.

I have an excerpt of both using dd if=/dev/zero of=file.img (locally)
and from a remote computer (on a full duplex gigabit network)

Locally: dd if=/dev/zero of=out.img

vmstat output:

procs -----------memory---------- ---swap-- -----io---- --system--
----cpu----
r b swpd free buff cache si so bi bo in cs us
sy id wa
1 0 412 4928 0 662612 0 0 0 0 1065 285 16
84 0 0
1 0 412 4608 0 662556 0 0 0 78932 1148 250 14
86 0 0
1 0 412 4672 0 662868 0 0 0 13 1068 274 16
84 0 0
1 0 412 5120 0 661644 0 0 0 78928 1116 245 13
87 0 0
1 0 412 4460 0 663020 0 0 0 0 1098 254 15
85 0 0
1 0 412 5376 0 660936 0 0 0 78932 1073 279 14
86 0 0
1 0 412 5376 0 662124 0 0 0 17 1147 267 18
82 0 0
2 0 412 5376 0 661580 0 0 44 31901 1052 286 15
85 0 0
1 0 412 5440 0 661740 0 0 48 47048 1168 265 15
85 0 0
1 0 412 5312 0 662152 0 0 4 14 1048 260 16
84 0 0
1 0 412 4664 0 662196 0 0 4 78941 1143 282 15
85 0 0
2 0 412 4600 0 662932 0 0 0 0 1104 251 14
86 0 0
1 0 412 5048 0 661448 0 0 0 78954 1111 269 16
84 0 0
1 0 412 5368 0 662120 0 0 0 0 1140 250 17
83 0 0
1 0 412 5432 0 660648 0 0 0 78928 1075 260 16
84 0 0
1 0 412 5176 0 662224 0 0 4 0 1217 266 14
86 0 0
1 0 412 5432 0 662104 0 0 0 0 1078 274 16
84 0 0
1 0 412 4984 0 662084 0 0 0 78932 1189 245 12
88 0 0
1 0 412 5440 0 662012 0 0 0 13 1080 271 17
83 0 0
1 0 412 5440 0 661168 0 0 0 78928 1109 242 14
86 0 0
1 0 412 5248 0 662232 0 0 0 0 1104 270 13
87 0 0

Remotely (writing over NFS, using NFS v3, not using direct I/O)

vmstat output:

procs -----------memory---------- ---swap-- -----io---- --system--
----cpu----
r b swpd free buff cache si so bi bo in cs us
sy id wa
1 0 596 4624 0 662460 0 0 0 448 4037 1099 1
99 0 0
1 0 596 4796 0 661564 0 0 0 96067 2877 788 1
75 0 24
0 0 596 5496 0 661692 0 0 0 160 2297 517 0
44 56 0
1 0 596 4824 0 662840 0 0 24 633 3896 1147 1
98 1 0
1 0 596 4696 0 662536 0 0 0 416 3987 1075 1
99 0 0
1 0 596 4892 0 662344 0 0 0 448 3746 1142 0
99 1 0
0 0 596 5868 0 660468 0 0 0 96064 3550 973 1
97 1 1
2 0 596 4600 0 663328 0 0 0 318 3390 823 1
71 28 0
1 0 596 5432 0 662336 0 0 0 448 3872 1066 0
100 0 0
1 0 596 5360 0 661888 0 0 0 453 3951 1144 1
99 0 0
1 0 596 4936 0 661568 0 0 0 94225 2942 726 0
75 7 18
1 0 596 4528 0 662944 0 0 0 259 2984 738 1
64 35 0
1 0 596 5368 0 661864 0 0 8 639 4234 1116 1
99 0 0
2 0 596 5368 0 662140 0 0 0 448 3919 1086 0
100 0 0
0 0 596 5512 0 662860 0 0 4 256 2747 650 0
56 43 1
0 0 596 5520 0 662860 0 0 0 0 1061 144 3 2 95 0
1 0 596 4688 0 663404 0 0 0 64 1455 312 0
13 87 0
1 0 596 5324 0 661964 32 0 40 96489 3325 921 1
88 11 0
1 0 596 4656 0 662852 24 0 144 1093 2228 650 1
38 58 3
1 0 596 4592 0 662648 0 0 0 448 3852 1098 0
99 1 0
1 0 596 5304 0 662232 0 0 0 448 3972 1100 1
99 0 0

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel"
in the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/




-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/