Date: Wed, 29 Jan 1997 11:13:14 -0800 (PST) From: Simon Shapiro <Shimon@i-Connect.Net> To: hackers@freebsd.org Subject: More 2.2-BETA goodies... Message-ID: <XFMail.970129120959.Shimon@i-Connect.Net>
next in thread | raw e-mail | index | archive | help
[ I am cross-posting this to scsi as well, as some of the issues here are
SCSI related and I am not sure that readership is overlapping... ]
Hi Y'all,
Since we are all bored with nothing better to do I decided to burden this
nice quorum with some more good ways to use 2.2-BETA. Maybe you can
clarify some of them for me:
1. An easy way to panic the system:
* Put ``devfs /devs devfs rw 0 0''
in /etc/fstab
* Do ``mount -a''
* Do ``mount -a'' (again)
* Do ``df'' - You will see /devs mounted twice.
Actually you will see /kern and /dev twice
if you setup kernfs and/or fdesc.
* do ``umount /devs''
You will get a nice little panic with a complaint about dangling inodes.
2. Not a show stopper. More of a question. Under HEAVY disk I/O, with
an AHA-2940W, you see all keyboard, mouse, modem activity freeze for
few seconds and then resume. I suspected SCSI timeouts, but saw no
kernel or driver complaints. It appears as if the system stops
processing interrupts for a while. Quiet SCSI timeout (no error
messages)?
3. Back to the Iomega Jaz: If it spins down (it likes to do that), and
then you try to access it (say mount a file system), you get dropped
the kernel debugger (``db>''), with ``.. sd... no slices...'' error
message.
Now, if you are lucky, you were not in X11, and can see the ``db>'' on
the screen. All you need to do is type ``c'' once and all will be well.
If you do it too quickly, you will repeat the event. Is this problem:
a. The aic7xxx driver timing out too soon
b. The SCSI code above the driver deciding to timeout too soon
c. Iomega not building hardware to FreeBSD specs :-)
If, on the other hand, you were in an X11 session, you are hosed.
I think the kernel debugger stops interrupts at this point. What is the
way to get at the debugger at this point. Aside from doing the proper
thing, which is to configure a serial console, etc.
4. Pppd drops the connection, without warning, under very heavy load.
5. Question: Using xperfmon++ (cool), one observes that a system equipped
with AHA2940W enjoys about 3-15 times as many interrupts/sec as disk
transfers/sec. The system, does nothing but SCSI right now (cvs
checkout in parallel with a tape backup (cpio, I know...).
Nothing else.
Assumin the #9 S3 card generates no interrupts, keeping my hands off the
keyboard and mouse, no Ethernet traffic, no PPP traffic; The system
does a peak of 246 disk T/s and about 625 interrupts/sec. Discounting
100 interrupts for the heartbeat, We still have two interrupts per sec
charged to the disk I/O. Under HEAVY load, the numbers climb
logarithmically; Up to 300 T/s with over 2,500 I/s.
Heavy disk load in this note means 8``st'' threads in parallel. So that
xperfmon++ will say 300+ T/s, which seems to be the maximum the AHA2940
can sustain. Ts is a Seek Test program we use to excite disks and stress
them at the same time. It generates a random (or sequential) stream of I/O
requests to a file/device and then measures how it does. It can either read
or write, or read-modify-write and do a whole set of nifty things.
If there is interest, I can post it. It runs on Slowlaris 2.5.1, linux 2.x
and FreeBSD 9of courcse). We prefer it to others because it does true
random seeks, very accurate and I wrote it, so it must be good :-). It is
also short!
Simon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.970129120959.Shimon>