Slow install MGA5B2 on USB2 memory stick

[jaywalker]

..and eventually, from the MGA5 desktop, installing individual packages by uprmi takes an average 1.7+ minutes per package

Code: Select all

Jan 25 19:20:18 slug urpmi[1528]: called with: mediainfo
Jan 25 19:20:35 slug urpmi[1528]: transaction on / (remove=0, install=0, upgrade=4)
Jan 25 19:20:35 slug [RPM][1528]: Transaction ID 54c54203 started
Jan 25 19:22:08 slug [RPM][1528]: install lib64zen0-0.4.29-4.mga5.x86_64: success
Jan 25 19:23:48 slug [RPM][1528]: install lib64mms0-0.6.4-4.mga5.x86_64: success
Jan 25 19:25:43 slug [RPM][1528]: install lib64mediainfo0-0.7.70-4.mga5.x86_64: success
Jan 25 19:27:15 slug [RPM][1528]: install mediainfo-0.7.70-3.mga5.x86_64: success
Jan 25 19:27:36 slug [RPM][1528]: Transaction ID 54c54203 finished: 0


Observations:
I can't blame a slow mirror because the installation from a USB iso took in excess of three hours (I had to go to bed before it was finished) and the post-install stage took a further 2 hours (my fault for choosing the fglrx video stuff - it pulls in a heap of dev packages and kernel headers).

I know the slow install on USB stick has been mentioned before but I was wondering if anyone knows of a magic bullet to fix the glacial write speed (if that is what is wrong). It cannot be solely due to using ext4 on a cheap HP USB2 memory stick, can it?

Monitoring the system while the medianfo installation was proceeding shows only that the CPUs are all mostly idle as is the sole disc on the machine - the USB boot stick, and the network.

Richard

[doktor5000]

..and eventually, from the MGA5 desktop, installing individual packages by uprmi takes an average 1.7+ minutes per package

Code: Select all

Jan 25 19:20:18 slug urpmi[1528]: called with: mediainfo


If you would have given me that hostname, I'd also refuse to work faster

More on topic, with usb devices, you want to use ext2 and no journaling.

It cannot be solely due to using ext4 on a cheap HP USB2 memory stick, can it?

Yes it can.

[jiml8]

I have seen installs of Linux operating systems take as long as 30 hours on those USB sticks. The write speeds really are that low.

[jaywalker]

Well, I have been weighing the costs-v-gains of re-installing with an ext2 root partition and just bashing on regardless with the ext4 install (minus its journal).

On the one hand I cannot find any evidence that the installation will proceed "a lot" more quickly. I can write a 1.5 GiB file to the (ext4) stick in less than two minutes but it takes 5+ hours to install a couple of Gimme Bytes of programs, libraries and data on a 13 GiB root partition.

All those numbers are approximations but there seems to be no obvious direct correlation between the write speed and the time it takes to install.

On the other hand the installation, though not yet complete, is working. It may not be fun, but it does work. I have removed the ext4 journal, but that has not made any measurable difference. In short, it has done its limited job; to allow me to check the operation of a PCTV292E DVB-T2 tuner USB stick on bare hardware to see if it works compared to its performance on a VirtualBox MGA5. Well, it does. At least it will tune all the available DVB-T multiplexes. I don't see any DVB-T2 results, but I now know that running MGA5 in Virtualbox just to get access to the PCTV driver from an earlier MGA3 or MGA4 is a non-starter and not worth pursuing any further.

R

[filip]

All those numbers are approximations but there seems to be no obvious direct correlation between the write speed and the time it takes to install.

Write speed is marketing anyway. You can test the speed with a lot of small files in parallel if possible and I think you'll be somewhat surprised . Not necessary in the positive way . I have also seen some new devices with unreliable write/read. The fact is that performance and quality of USB sticks just varies wildly. Luckily the same gap in SSD is somewhat smaller .