Questions on trace data

[sylab]

Questions on trace data

[zy1024cs]

Hi, I'm running the experiment as well but find it hard to reproduce the access pattern in this figure.

As far as I can see, the downloaded trace files are in this format: [ts in ns] [pid] [process] [lba] [size in 512 Bytes blocks] [Write or Read] [major device number] [minor device number] [MD5 per 4096 Bytes] Some sample data from webmail day 16 (webmail.cs.fiu.edu-110108-113008.16.blkparse) would like 1299602115063356 318 kjournald 220040 8 W 2 0 d186315aa2d4c75b753fe9ca98187cdd

To draw the same figure, we have the timestamp (first column) and the logic block address (fourth column). However, if we draw a scatter plot over the time and block address, the resulting figure would look like this: Very likely I misunderstood the dataset and drew it in the wrong way. It'd be very kind of you if you could explain a bit how to generate that figure. Thank you very much!

You should narrow the range of the coordinate axis, otherwise you often see a line. Because if you want to have that access mode, you can usually observe it in a small range.

[lia54]

Hi @Wkkkkk thanks for the interest.
For the purpose of building the access pattern, we mapped each lba value of the trace to a unique number to reduce the range on the y-axis of the plot. That's probably why you're seeing a different graph, since you're directly plotting the lba from the trace. The x-axis is the timestamp converted from nanoseconds to hours.

[Wkkkkk]

Hi,
I'm running the experiment as well but find it hard to reproduce the access pattern in this figure.

As far as I can see, the downloaded trace files are in this format:
[ts in ns] [pid] [process] [lba] [size in 512 Bytes blocks] [Write or Read] [major device number] [minor device number] [MD5 per 4096 Bytes]
Some sample data from webmail day 16 (webmail.cs.fiu.edu-110108-113008.16.blkparse) would like
1299602115063356 318 kjournald 220040 8 W 2 0 d186315aa2d4c75b753fe9ca98187cdd

To draw the same figure, we have the timestamp (first column) and the logic block address (fourth column). However, if we draw a scatter plot over the time and block address, the resulting figure would look like this:

Very likely I misunderstood the dataset and drew it in the wrong way. It'd be very kind of you if you could explain a bit how to generate that figure. Thank you very much!

[lia54]

Hi @brucechin,
Thank you for your interest in our paper!

Please find the datasets and the corresponding links/locations (when possible):

1. FIU SRC_Map (All traces are a one-day duration for each file).
2. MSR Cambridge (These traces are a one-week duration for each file. For the paper, we extracted the first day only based on the timestamp).
3. CloudVPS (These traces are a one-day duration for each file).
4. CloudCache is a collection of the webserver and moodle traces that were used in the CloudCache paper from FAST'16. (All one-day duration).
5. CloudPhysics are non-public traces used in the SHARDS paper from FAST'15 that were shared directly from the authors.

[Wkkkkk]

Hi,
Thank you both for the quick reply! Sure I'd like to try the mapping to narrow down the range in the y-axis. Would it be possible for you to share a bit about how you did the mapping?
Your paper is quite interesting and your insights about the access pattern mean a lot!

[Wkkkkk]

The latest plot if we just draw the ones blow 140K.

[zy1024cs]

The file types of CloudVPS seem to be the ".blktrace.1". Please ask me how I should convert them, after all the contents seem impossible to handle
.
includi

ngCloudCache

[jzx-bitdb]

Hi, Thank you both for the quick reply! Sure I'd like to try the mapping to narrow down the range in the y-axis. Would it be possible for you to share a bit about how you did the mapping? Your paper is quite interesting and your insights about the access pattern mean a lot!

I have the same question about how to range in the y-aixs. Have you solved this problem？

[jzx-bitdb]

Hi @Wkkkkk thanks for the interest. For the purpose of building the access pattern, we mapped each lba value of the trace to a unique number to reduce the range on the y-axis of the plot. That's probably why you're seeing a different graph, since you're directly plotting the lba from the trace. The x-axis is the timestamp converted from nanoseconds to hours.

Could you tell me how to map each lba value of the trace. I have two ideas about it. The first one is scale each lba value by the same factor. The second is iterator the request sequence to map them to an increment value. I have the similar picture as yours by the second method. Am I right? And what is the difference between the two methods?

[Wkkkkk]

Hi @xuangestallone,
Your results seem to be a great example. Would you mind sharing the second method in more detail or the code so I could repeat and try it myself as well? I want to keep experimenting and discussing it with you if possible.

My email is [email protected].

[zwh272581638]

The file types of CloudVPS seem to be the ".blktrace.1". Please ask me how I should convert them, after all the contents seem impossible to handle . includi ngCloudCache

Perhaps you should read the user manual on the fio and blktrace official website

[sylab]

Closing due to inactivity.