Parallels Speed ISSUES!! - It became slow!!

Hi,

I run my usual tests in Maya to check Parallels speed.
Unluckily they are much slower than the first time I run them with beta 4

I have a mac mini core duo with latest firmware and OSX 10.4.6.
The windows image used for testing has always been the same: WinXP SP2
Virtualization is "ON" and in Parallels-about-more menu it says "Intel Vt-x"

I REINSTALLED ALL the 3 Parallels versions tested with their own Tools. also rebooting the mac mini.

I don't know when it started being slow, I realised today.

Maya scene 1 with software Render.
- Parallels Beta 4 68 sec. (tested ~ 1 month ago)
- Parallels Beta 5 83 sec. (tested today)
- Parallels Beta 6 83 sec. (tested today)
- Parallels RCandidate 93 sec. (tested today)
- Parallels RCandidate 93 sec. (tested today) WIN2000
- Maya WinXP (Bootcamp) 42 sec.(tested ~ 1 month ago)
- Maya OSX Rosetta 171 sec.(tested today - 178 ~ 1 month ago))

Maya scene 2 with mentalray Render.
- Parallels Beta 4 123 sec. (tested ~ 1 month ago)
- Parallels Beta 5 165 sec. (tested today)
- Parallels Beta 6 165 sec. (tested today)
- Parallels RCandidate 190 sec. (tested today)
- Maya WinXP (Bootcamp) 69 sec.(tested ~ 1 month ago)
- Maya OSX Rosetta 325 sec.(tested today - 340 ~ 1 month ago)

So the Mac Mini is running as fast, or even better than a month ago.
It is not my Image since a fresh Win2000 install is giving me the same result as my WinXP SP2.
Are you aware that Parallels is running very slow?
Can you replicate the problem?
If you look at the tests the speed I got time ago make sense. Since Parallels uses 1 processor I got slightly more than double compared to WinXP SP2 native with bootcamp.

If these are the results Parallels takes from 2.5 to 3 times Bootcamp.

Any Help or Light into this is welcome!!

 

FWIW, I see the same thing -- each beta has gotten slower for whatever reason.

 

Same here. Everything feels slower. But I have to say (as I reported already) network speed was improved in FC5 from B6 ro RC. Else in XP it feels slower.

 

Sloooow...

I have been running Parallels from the start and it definately feels slower for each release. I use some company specific asp software, this flew on the original release and is now crawling along. Any help on this would be appreciated.

 

Can anyone from parallels comment on this please?

Thank you very much.

L.

 

Speed issue

Everytime I install an update, the VM operates slower. This loss in speed (i.e. longer boot time) can be recovered by fresh installing windows and creating a new VM. I saw a definite improve when i did from Beta 4 and RC.

 

Excellent performance at the virtual HDD level

Using a XP SP2 guest, using an expanding disk image, I have seen significant impact on overall feeling of performance after running the "Disk Compacting Tool" which you will find in the Parallels Tools Center (in the XP guest, after having proceeded with the tools installation of course).

I run it in the "Execute all stages at once" mode. Then, I use the Parallels Disk Image Tools on Mac OS X and proceed with the defragmentation of the virtual HDD file. I don't know exactly what it does and how it does it (I am not even sure I want to know it), but I see a significant speed increase after these maintenance steps. My VM HDD also decreased from 13 GB to 10 GB after the last run. The boot time of Windows XP until the logon screen appears is much shorter (well under 10 sec) as well is the time between the logon and desktop being ready to interact with me.

One side note: This XP VM has a 60 GB NTFS formatted file system using a cluster size of 4 KB instead of the common default of 512 bytes. I did another VM configuration with such a default 512 bytes NTFS file system. The overall feeling was slower especially on high file I/O activities. Theorically you loose less disk space than with a 4 KB cluster size, __but__, I have the empirical feeling that the larger cluster size greatly optimizes XP own disk access while Parallels expanding imaging of the virtual HDD cuts on some of the space losses the 4 KB cluster size would introduce.

None of this experience is backed by any inside details on Parallels Desktop (which I don't have). Your mileage may vary. But I wanted to share: it really flies for me.

 

Installing XP on a non-default cluster-size partition

When installing XP in the classical way (booting its install CD) you won't have the choice to set the cluster size of your NTFS partition. There are two ways to do it.

1) If you have another XP VM already configured, you can create a new virtual disk through Parallels Desktop and attach it to your VM. Then you can create a NTFS partition on it, selecting the cluster size of your choice, using the standard tools of XP (either GUI - disk administrator - or command-line based - format). Then you shutdown this VM, unmount the disk, create a new VM, attach this new pre-made HDD to it and install on top of this empty partition by taking care of NOT re-formatting it again during the setup process.

2) Simpler method: do none of the above. Start XP setup, ask it to format your partition as FAT32. Unless the disk size is too small, it will opt for a 4 KB cluster size. After installation, use XP standard command-line "convertfs /NTFS" to turn this partition in an NTFS one. It will keep the cluster size unchanged.

 

No no no
It is not about images, compress, and all that ..
A FRESH WIN2000 image gave me the same result as the OLD (1 month) WINXP SP2 image.

SO Parallels became slo .. and slow ... and those are the numbers ..
I would like womeone from Parallels to comment on this if possible ..
Thank you.

 

Set aside the discussion on disk I/O, I can not confirm a degradation in raw cpu performance when running the VM over the last betas until the RC1. I'm not denying that you see a performance degradation. I'm just saying that I don't see such, however I measure it.

With possibly one exception: with NO sound card in the VM configuration, my VM generally feels more responsive than with the sound card added.

On tests of pure CPU performance or pure HD I/O performance, I see no difference.

Super Pi 1.1e, running in my XP VM, still computes a million decimals of Pi in 31 to 33 secs, depending on the level of other activity on my 2 GHz MBP. This number has not changed since beta 4. So at least on raw processes demanding pure cpu/ram, no degradation of performances.

Sure a comment by Parallels would be useful as there have been reports of slowdowns.

 

Very good point, For example, once your web browser is used, it becomes choked with history and caches. A web browser performs 2 to 3 times better on a new Mac if you don't use the firewire transfer to bring over all the junk from the old mac. I have found nothing that improves speed more than formatting the HD and reinstalling the OS. Problem is the fix is short lived and I never do it on my Macs. Between malware & viruses, most the average none technical XP user get a HD format or new computer ever 6 months to a year.

 

I too cannot see any signs of raw CPU slowdown. I just re-ran a benchmark of my own that does a good bit of disk I/O and a lot of CPU work (batch debayering 3 megapixel files using a complex debayer routine) and I'm getting the same times I was getting on several previous versions. I run the test a few times and I actually turned in the fastest time by a second using RC1 (likely just a random sampling - I don't think it's actually faster).

Craig

 

Thank you for reporting. We will investigate and solve this prior to GA.

 

That's what we like to hear. Thanks Andrew!

 

Thank you Andrew.

Maya tests are pure raw CPU performance. NO I/O disk involved. No cache involved.
It is a 500kb Maya scne taking minutes to render ..
Also I do not agree on the reistalling side.
When it comes to cpu performances I've had my machines taking the same time it originally took to perform CPU calculation even if the disk image was used since 8 months, with no reboot since 15 or more days.
I agree after 3 days a mac with only 2GB of ram feels slower. But that's because of the swap disk size. Then rendering or video encoding takes the same.

And I don't mean it feels slower. I mean cronograph. I mean the same file tests, I mean no other app running, I mean the same "situation", I mean a bit of precision.
So if you feel something, well it is nice, but I havent seen people posting their tests here. They are the only thing that counts in this thread and in ivestigations (sorry to be this straight).

And it would be very welcome if some of you still have beta 4, beta 5, beta 6 & RC1 and can perform tests with chronograph using the same Win image showing the opposite.
After all the performace decrease I am experiencing is around 40%. So huge.

Thank you all.

 

I'd posted the numbers in another thread. I am a scientist by trade, so yes, I am not so foolish as to just go by some "feel something" report. I'd posted that my benchmarks were within a tolerance of a few seconds. Of course this is with the same files. In addition, Oliver noted that his test was stable -- calculating Pi to a million places, reporting 31-33 seconds. Again, do not assume that other users here are foolish.

If you think that in a multi-tasking environment your chronograph will always give the same exact number, I'm afraid you are mistaken. Many services run in the background that we cannot control. In truth, to be accurate, the tests should be repeated several times and we should report some description of the variance.

In any case, the per-run time has remained a constant 26-27 seconds across every version tested, dating I believe back to beta4. This is using a mix of floating point and integer math mainly with about 250M of RAM used directly in the calculations. It must read a total of five 4Mb files off the disk, perform all the calculations, and write out 19Mb files. As such it is a mixed CPU and disk test as noted.

Craig

 

Hi,

so the problem was a communication one.
Now it is clear what you did and how, before not.
Thank you.

Also, has anyone a beta 4 download link that I re test the same Maya scenes?
Thank you.

 

I too noticed that things were getting slower. However, I just installed another 1GB and now things are feeling snappy again. For people seeing the slowdown, how much memory do you have?

 

This is off-topic, but what's it like to run Maya in Parallels? Can you actually use it? This would be really great. Any info would be greatly appreciated.

 

I have all I can 2GB ... dedicated 512 to Parallels in any of the tests. But 2GB isn't enough for working. After a few days you have a 3GB swap file !!!!

Maya on Parallels I don't know because I use it only for rendering.
It was as fast as boot camp to beta 4 (1 processor only) then it became a lot slower ...
Still faster than Maya on Rosetta.
But nothing serious. If you're serious you buy something else to work with maya ... As soon as the scene starts to be a LITTLE complicate the viewport wouldn't rotate anymore .... As simple as that ....