DriveSpace effect in Veaam - is it possible?

[pizang]

Do you remember DriveSpace in Win95? It was a built in software for online compression of hard disk. What I call 'DriveSpace effect' appear having fast CPU and slow HDD. In such a situation online compression can actually improve your performance because system have to read less data from slow disk and it can decompress it fast using CPU.

The question is whether someone observed it in Veeam using best and optimal connection? I think it might be possible when using slow target or using for example 100Mb/s link between Veeam server and target.

Another question is how compression algorithm scales up? I have 1 CPU 4 Core server and Veeam make use of all four cores when using best compression. Does someone tried it on more powerful server like 2CPU per 4 cores or more?

[Vitaliy S.]

Hello Miroslaw,

There are two points in using compression you should be aware, let's take a look at both of them.

In case of using VCB mode, all data flows through your VCB proxy offloading the load on your hosts. Multiple CPUs on your proxy really help when using Best compression, especially with parallel jobs in VCB mode. If you are using default compression and no parallel jobs, then single CPU is more than enough. Faster clock speed will help with overall processing speed (faster data processing), but most likely insignificantly (data blocks hashing, and default compression are not really CPU-intensive operations).

However be aware, you shouldn't use compressed or encrypted drives for your TEMP folder, usually it is located on drive C, in this case it will affect the driver that is being responsible for transmitting data on-the-fly, all data will be stored locally on TEMP folder.

But of course using best compression, will affect your job time, cause less data should be transmitted, so the less VBK file will be in the end. On the other hand, compression depends heavily on specific VM's: amount of free disk space, files stored on it and such.

Also to have a full picture of compression feature, please have a look at the following topic regarding compression:

http://www.veeam.com/forums/viewtopic.p ... hili#p6566 (some facts about compression levels)

http://www.veeam.com/forums/viewtopic.p ... 56&start=0 (examples using compression and de-duplication)

Thank you.

[pizang]

Thank you for your explanation. I have posted my idea because I can see that best compression is pretty close to optimal one (22 MB/s to 29 MB/s in my case with 1 Xeon E5450 @ 3GHz) so maybe someone who has more powerful proxy observed that best compression improved performance.

I am also going to test it on 64-bit W2k3 - is it going to improve performance?

[Vitaliy S.]

Miroslaw,

Basically, 64-bit OS has more availabilities for scalability (memory, CPU usage e.t.c), there are essentially two reasons why a 64-bit Windows should be faster: compilers have more options for code optimization and all 64-bit CPUs support Streaming SIMD Extensions (command set that is supposed to be more efficient). However, the perfomance surely depends on how many jobs do you run at the same time, and of course all the perfomance rates fully relates on your Virtual Infrastructure (how everything is settled up, connected, e.t.c)

But, it'll be great to see your feedback here on the perfomance changes.

Thanks.

[Gostev]

The question is whether someone observed it in Veeam using best and optimal connection? I think it might be possible when using slow target or using for example 100Mb/s link between Veeam server and target.

Yes, most definitely - please see the link below - another customer reports exactly this.

Another question is how compression algorithm scales up? I have 1 CPU 4 Core server and Veeam make use of all four cores when using best compression. Does someone tried it on more powerful server like 2CPU per 4 cores or more?

It scales extremely well, check out some exact numbers from 2 CPU 4 core server in this thread:
http://www.veeam.com/forums/viewtopic.p ... 4848#p4848
(read down starting from this post)

[pizang]

Performance changes after 32 to 64 bit VCB proxy upgrade.

We have moved to vSphere 4.0 and W2K3 SR2 SP2 64-bit. As our host has 8GB of memory now it can see all of it.

I have benchmarked backup of one VM with best compression method. Performance didn't changed. Exactly the same number 23MB/s. I can assume that having more memory and 64 bit system does not influence compression algorithm. If our budget allow we will upgrade VCB proxy from 1CPU/4cores to 2CPU/4cores and I will post numbers here.

I do not expect any performance increase using none and optimal option. Is is just a function of source and target performance.

[Gostev]

Correct. Compression is math-heavy operation, so when using Best compression, performance is determined by raw computing power (the number of cores) solely. RAM speed also helps, but to less extent and this will be noticeable with 8 core computer only.