I’ve been setting up a new development VMware server, and I’ve learned a few things about RAID performance. Our server is a brand new 3GHz Quad core Xeon, with 10 gigs of RAM. We initially set it up with 5 disks, 4 for RAID 5, and the fifth as a spare.
VMware virtual machines tend to be huge, especially if you’re pre-allocating the disks. In my case, I’ve pre-allocated them to have 15GB hard drives. I set up a couple on my laptop, and then tried to copy them up to the development server.
Much to my surprise, the write performance was horrible. Copying one large file effectively brings the server to a halt, and the copy isn’t really that fast to begin with.
Upon further investigation, I found out that RAID 5 write performance is less than desirable. Write performance is pretty important when you’re moving around huge files.
This morning we decided to convert the array to RAID 10, sometimes known as raid 1+0. Write performance is great, and read performance is just as good as RAID 5. We also increased the NTFS cluster size to 64KB, to better match the types of files we’re storing (big ones).
The results have been excellent. I can now copy a virtual machine in 10 minutes instead of 30+. The only disadvantage is that I had to lose 143GB (1 drive in the array).
If you’re setting up a VMware server, I recommend:
●Set the NTFS cluster size high (64k), and match that to your stripe size
●Use RAID 10
我的結論:
●VMWARE出來的一包不會小的,就算用內建功能把他切4G一包4G一包還是大,所以理所當然stripe和cluster的size要大。
●raid用哪種,還是要看你虛擬出來的機器的用途,測試的就放在raid0裡,當server用的就10(不要選5)。
不過我認為就算當server我也會選0,一方面省成本,再者反正要備份也不難,就整塊給他複製一下就可以了,但是要當資料庫等具互動式的server還是有容錯能力比較好。
沒有留言:
張貼留言