julian.kunkel
/
mistral-io-datasets
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Fix various typos.

This commit is contained in:
Julian M. Kunkel 2020-10-22 16:30:06 +01:00
parent bbad8f8272
commit 4ab98b256c
1 changed files with 58 additions and 55 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

107
paper/main.tex
View File

@ -69,7 +69,7 @@
\usepackage{cleveref}
\crefname{codecount}{Code}{Codes}
\title{A Workflow for Identifying Jobs with Similar I/O Behavior by Analyzing the Timeseries}
\title{A Workflow for Identifying Jobs with Similar I/O Behavior Utilizing Time Series Analyzing}
\author{Julian Kunkel\inst{2} \and Eugen Betke\inst{1}}
@ -85,17 +85,19 @@ DKRZ --
\begin{abstract}
One goal of support staff at a data center is to identify inefficient jobs and to improve their efficiency.
Therefore, a data center deploys monitoring systems which captures the behavior of the executed jobs.
While it is easy to utilize statistics to rank jobs based on the utilization of compute, storage, and network, it is tricky to find patterns in 100.000 of jobs, i.e., is there a class of jobs that aren't performing well.
When support staff investigates a single job, it is relevant to identify related jobs in order to understand the usage of the exhibited behavior better and assess the optimization potential.
Therefore, a data center deploys monitoring systems that capture the behavior of the executed jobs.
While it is easy to utilize statistics to rank jobs based on the utilization of computing, storage, and network, it is tricky to find patterns in 100.000 jobs, i.e., is there a class of jobs that aren't performing well.
Similarly, when support staff investigates a specific job in detail, e.g., because it is inefficient or highly efficient, it is relevant to identify related jobs.
This allows staff to understand the usage of the exhibited behavior better and to assess the optimization potential.
\medskip
In this paper, a methodology to rank the similarity of all jobs to a reference job based on their temporal IO behavior is described.
Practically, we apply several of previously developed time series based algorithms and also utilize Kolmogorov-Smirnov to compare the distribution of the statistics.
A study is conducted to explore the effectivity of the approach which starts starts from three reference jobs and investigates related jobs.
The data stems from DKRZ's supercomputer Mistral and includes more than 500.000 jobs that have been executed during 203 days of operation.
%Problem with definition of similarity.
Our analysis shows that the strategy and algorithms are effective to identify similar jobs and revealed some interesting patterns on the data.
Practically, we apply several previously developed time series algorithms and also utilize Kolmogorov-Smirnov to compare the distribution of the statistics.
A study is conducted to explore the effectivity of the approach which starts from three reference jobs and investigates related jobs.
The data stems from DKRZ's supercomputer Mistral and includes more than 500.000 jobs that have been executed for more than 6 months of operation %203 days.
%Problem with the definition of similarity.
Our analysis shows that the strategy and algorithms are effective to identify similar jobs and revealed interesting patterns in the data.
\end{abstract}
@ -109,27 +111,29 @@ Secondly, they aim to improve the efficiency of all workflows -- represented as
In order to optimize a single job, its behavior and resource utilization must be monitored and then assessed.
Rarely, users will liaise with staff and request a performance analysis and optimization explicitly.
Therefore, data centers deploy monitoring systems and staff must pro-actively identify candidates for optimization.
Monitoring tools such as \cite{Grafana} and \cite{XDMod} provide various statistics and time series data for the job execution.
Monitoring tools such as \cite{Grafana} and \cite{XDMod} provide various statistics and time-series data for job execution.
The support staff should focus on workloads for which optimization is beneficial, for instance, the analysis of a job that is executed once on a medium number of nodes costs human resources and is not a good return of investment.
By ranking jobs based on the statistics, it isn't difficult to find a job that exhibits extensive usage of compute, network, and IO resources.
By ranking jobs based on the statistics, it isn't difficult to find a job that exhibits extensive usage of computing, network, and IO resources.
However, would it be beneficial to investigate this workload in detail and potentially optimize it?
A pattern that can be observed in many jobs bears a potential as the blueprint for optimizing one job may be applied to other jobs as well.
This is particularly true when running one application with similar inputs but also different applications may lead to a similar behavior.
Therefore, it is useful for support staff that investigates a resource hungry job to identify similar jobs that are executed on the supercomputer.
A pattern that can be observed in many jobs bears potential as the blueprint for optimizing one job may be applied to other jobs as well.
This is particularly true when running one application with similar inputs but also different applications may lead to similar behavior.
Therefore, it is useful for support staff that investigates a resource-hungry job to identify similar jobs that are executed on the supercomputer.
In our previous paper \cite{XXX}, we developed several distance metrics and algorithms for the clustering of jobs based on the time series of their IO behavior.
The distance metrics can be applied to jobs with different runtime and number of nodes utilized but differ in the way the define similarity.
The distance metrics can be applied to jobs with different runtime and number of nodes utilized but differ in the way they define similarity.
We showed that the metrics can be used to cluster jobs, however, it remains unclear if the method can be used by data center staff to explore jobs of a reference job effectively.
In this article, we refined these distance metrics slightly and apply them to rank jobs based on their similarity to a reference job.
Therefore, we perform a study on three reference jobs with different character.
Therefore, we perform a study on three reference jobs with a different character.
We also utilize Kolmogorov-Smirnov-Test to illustrate the benefit and drawbacks of the different methods.
This paper is structured as follows.
We start by introducing related work in \Cref{sec:relwork}.
%Then, in TODO we introduce the DKRZ monitoring systems and explain how I/O metrics are captured by the collectors.
In \Cref{sec:methodology} we describe briefly the data reduction and the machine learning approaches.
In \Cref{sec:evaluation}, we perform a study by applying the methodology on three jobs with different behavior, therewith, assessing the effectiveness of the approach to identify similar jobs.
In \Cref{sec:methodology} we describe briefly the data reduction and the algorithms for similarity analysis.
Then, we perform our study by applying the methodology to three jobs with different behavior, therewith, assessing the effectiveness of the approach to identify similar jobs.
In \Cref{sec:evaluation}, the reference jobs are introduced and quantitative analysis of the job pool is made based on job similarity.
In \Cref{sec:timelines}, the 100 most similar jobs are investigated in more detail, and selected timelines are presented.
Finally, we conclude our paper in \Cref{sec:summary}.
\section{Related Work}
@ -138,42 +142,41 @@ Finally, we conclude our paper in \Cref{sec:summary}.
\section{Methodology}
\label{sec:methodology}
The purpose of the methodology is to allow user and support staff to explore all executed jobs on a supercomputer in the order of their similarity to the reference job.
The purpose of the methodology is to allow users and support staff to explore all executed jobs on a supercomputer in order of their similarity to the reference job.
Therefore, we first need to define the job data, then describe the algorithms used to compute the similarity, and, finally, the methodology to investigate jobs is described.
\subsection{Job Data}
On the Mistral supercomputer at DKRZ, the monitoring system gathers in 10s intervals on all nodes nine IO metrics for the two Lustre file systems together with general job metadata from the SLURM workload manager.
The results are 4D data (time, nodes, metrics, file system) per job.
The distance metrics should handle jobs of different length and node count.
The distance metrics should handle jobs of different lengths and node count.
In \cite{TODOPaper}, we discussed a variety of options from 1D job-profiles to data reductions to compare time series data and the general workflow and pre-processing in detail.
In a nutshell, for each job executed on Mistral, we partition it into 10 minute segments and compute the arithmetic mean of each metric, categorize the value into non-IO (0), HighIO (1) and CriticalIO (4) for values below 99-percentile, up to 99.9-percentile, and above, respectively.
In a nutshell, for each job executed on Mistral, we partition it into 10-minute segments and compute the arithmetic mean of each metric, categorize the value into non-IO (0), HighIO (1), and CriticalIO (4) for values below 99-percentile, up to 99.9-percentile, and above, respectively.
After data is reduced across nodes, we quantize the timelines either using binary or hexadecimal representation which is then ready for similarity analysis.
By pre-filtering jobs with no I/O activity -- their sum across all dimensions and time series is equal to zero, we are reducing the dataset from about 1 million jobs to about 580k jobs.
\subsection{Algorithms for Computing Similarity}
In this paper, we reuse the algorithms developed in \cite{TODO}: bin\_all, bin\_aggzeros, hex\_native, hex\_lev, and hex\_quant.
They differ in the way data similarity is defined; either the binary or hexadecimal coding is used, the distance metrics is mostly the Euclidean distance or the Levensthein-distance.
For jobs with different length, we apply a sliding-windows approach which finds the location for the shorter job in the longer job with the highest similarity.
The hex\_quant algorithm extracts I/O phases and computes similarity between the most similar I/O phases of both jobs.
They differ in the way data similarity is defined; either the binary or hexadecimal coding is used, the distance metrics is mostly the Euclidean distance or the Levenshtein-distance.
For jobs with different lengths, we apply a sliding-windows approach which finds the location for the shorter job in the long job with the highest similarity.
The hex\_quant algorithm extracts I/O phases and computes the similarity between the most similar I/O phases of both jobs.
\paragraph{Kolmogorov-Smirnov (kv) algorithm}
In this paper, we add a new similarity definition based on Kolmogorov-Smirnov-Test that compares the probability distribution of the observed values which we describe in the following.
% Summary
For the analysis, we perform two preparation steps.
Dimension reduction by computing mean across the two file systems and by concatenating the time series data of the individual nodes.
This reduces the four dimensional dataset to two dimensions (time, metrics).
Dimension reduction by computing means across the two file systems and by concatenating the time series data of the individual nodes.
This reduces the four-dimensional dataset to two dimensions (time, metrics).
% Aggregation
The reduction of the file system dimension by the mean function ensures the time series values stay in the range between 0 and 4, independently how many file systems are present on an HPC system.
The fixed interval of 10 minutes also ensure the portability of the approach to other HPC systems.
Unlike the previous similarity definitions, the concatenation of time series on the node dimension preserves the individual I/O information of all nodes while it allows comparison of jobs with different number of nodes.
The fixed interval of 10 minutes also ensures the portability of the approach to other HPC systems.
Unlike the previous similarity definitions, the concatenation of time series on the node dimension preserves the individual I/O information of all nodes while it allows comparison of jobs with a different number of nodes.
We apply no aggregation function to the metric dimension.
% Filtering
%Zero-jobs are jobs with no sign of significant I/O load are of little interest in the analysis.
%Their sum across all dimensions and time series is equal to zero.
%Furthermore, we filter those jobs whose time series have less than 8 values.
% Oben beschrieben
% Similarity
For the analysis we use the kolmogorov-smirnov-test 1.1.0 Rust library from the official Rust Package Registry ``cargo.io''.
@ -199,12 +202,12 @@ Start the inspection of the jobs looking at the most similar jobs first.
The user can decide about the criterion when to stop inspecting jobs; based on the similarity, the number of investigated jobs, or the distribution of the job similarity.
For the latter, it is interesting to investigate clusters of similar jobs, e.g., if there are many jobs between 80-90\% similarity but few between 70-80\%.
For the inspection of the jobs, a user may explore the job metadata, searching for similarities and explore the time series of a job's IO metrics.
For the inspection of the jobs, a user may explore the job metadata, searching for similarities, and explore the time series of a job's IO metrics.
\section{Evaluation}
\label{sec:evaluation}
For each reference job and algorithm, we created a CSV files with the computed similarity for all other jobs.
For each reference job and algorithm, we created CSV files with the computed similarity for all other jobs.
Next, we analyzed the performance of the algorithm.
Then the quantitative behavior and the correlation between chosen similarity and number of found jobs, and, finally, the quality of the 100 most similar jobs.
@ -213,16 +216,16 @@ Then the quantitative behavior and the correlation between chosen similarity and
In the following, we assume a job is given and we aim to identify similar jobs.
We chose several reference jobs with different compute and IO characteristics:
\begin{itemize}
\item Job-S: performs post-processing on a single node. This is a typical process in climate science where data products are reformatted and annotated with metadata to a standard representation (so called CMORization). The post-processing is IO intensive.
\item Job-M: a typical MPI parallel 8-hour compute job on 128 nodes which writes time series data after some spin up. %CHE.ws12
\item Job-S: performs post-processing on a single node. This is a typical process in climate science where data products are reformatted and annotated with metadata to a standard representation (so-called CMORization). The post-processing is IO intensive.
\item Job-M: a typical MPI parallel 8-hour compute job on 128 nodes which write time series data after some spin up. %CHE.ws12
\item Job-L: a 66-hour 20-node job.
The initialization data is read at the beginning.
Then only a single master node writes constantly a small volume of data; in fact, the generated data is too small to be categorized as IO relevant.
\end{itemize}
The segmented timeline of the jobs are visualized in \Cref{fig:refJobs}.
The segmented timelines of the jobs are visualized in \Cref{fig:refJobs}.
This coding is also used for the HEX class of algorithms (BIN algorithms merge all timelines together as described in \jk{TODO}.
The figures show the values of active metrics ($\neq 0$) only; if few are active then they are shown in one timeline, otherwise they are rendered individually to provide a better overview.
The figures show the values of active metrics ($\neq 0$) only; if few are active then they are shown in one timeline, otherwise, they are rendered individually to provide a better overview.
For example, we can see in \Cref{fig:job-S}, that several metrics increase in Segment\,6.
In \Cref{fig:refJobsHist}, the histograms of all job metrics are shown.
@ -310,7 +313,7 @@ A boxplot for the runtimes is shown in \Cref{fig:performance}.
The runtime is normalized for 100k jobs, i.e., for bin\_all it takes about 41\,s to process 100k jobs out of the 500k total jobs that this algorithm will process.
Generally, the bin algorithms are fastest, while the hex algorithms take often 4-5x as long.
Hex\_phases is slow for Job-S and Job-M while it is fast for Job-L, the reason is that just one phase is extracted for Job-L.
The Levensthein based algorithms take longer for longer jobs -- proportional to the job length as it applies a sliding window.
The Levenshtein based algorithms take longer for longer jobs -- proportional to the job length as it applies a sliding window.
Note that the current algorithms are sequential and executed on just one core.
For computing the similarity to one (or a small set of reference jobs), they could easily be parallelized.
We believe this will then allow a near-online analysis of a job.
@ -343,7 +346,7 @@ We believe this will then allow a near-online analysis of a job.
\subsection{Quantitative Analysis}
In the quantitative analysis, we explore for the different algorithms how the similarity of our pool of jobs behaves to our three reference jobs (Job-S, Job-M, and Job-L).
In the quantitative analysis, we explore the different algorithms how the similarity of our pool of jobs behaves to our three reference jobs (Job-S, Job-M, and Job-L).
The cumulative distribution of similarity to the reference jobs is shown in \Cref{fig:ecdf}.
For example, in \Cref{fig:ecdf-job-S}, we see that about 70\% have a similarity of less than 10\% to Job-S for HEX\_native.
BIN\_aggzeros shows some steep increases, e.g., more than 75\% of jobs have the same low similarity below 2\%.
@ -353,7 +356,7 @@ The different algorithms lead to different curves for our reference jobs, e.g.,
The support team in a data center may have time to investigate the most similar jobs.
Time for the analysis is typically bound, for instance, the team may analyze the 100 most similar ranked jobs; we refer to them as the Top\,100 jobs, and Rank\,i refers to the job that has the i-th highest similarity to the reference job -- sometimes these values can be rather close together as we see in the following histogram.
In \Cref{fig:hist}, the histograms with the actual number of jobs for a given similarity are shown.
As we focus on a feasible number of jobs, the diagram should be read from right (100\% similarity) to left; and for a bin we show at most 100 jobs (total number is still given).
As we focus on a feasible number of jobs, the diagram should be read from the right (100\% similarity) to left; and for a bin we show at most 100 jobs (total number is still given).
It turns out that both BIN algorithms produce nearly identical histograms and we omit one of them.
In the figures, we can see again a different behavior of the algorithms depending on the reference job.
Especially for Job-S, we can see clusters with jobs of higher similarity (e.g., at hex\_lev at SIM=75\%) while for Job-M, the growth in the relevant section is more steady.
@ -426,21 +429,21 @@ Therefore, they are not published in our data repository.
To understand if the analysis is inclusive and identifies different applications, we use two approaches with our Top\,100 jobs:
We explore the distribution of users (and groups), runtime, and node count across jobs.
The algorithms should include different users, node counts, and across runtime.
To confirm hypotheses presented, we analyzed the job metadata comparing job names which validates our quantitative results discussed in the following.
To confirm the hypotheses presented, we analyzed the job metadata comparing job names which validate our quantitative results discussed in the following.
\paragraph{User distribution.}
To understand how the Top\,100 are distributed across users, the data is grouped by userid and counted.
\Cref{fig:userids} shows the stacked user information, where the lowest stack is the user with the most jobs and the top most user in the stack has the smallest number of jobs.
\Cref{fig:userids} shows the stacked user information, where the lowest stack is the user with the most jobs and the topmost user in the stack has the smallest number of jobs.
For Job-S, we can see that about 70-80\% of jobs stem from one user, for the hex\_lev and hex\_native algorithms, the other jobs stem from a second user while bin includes jobs from additional users (5 in total).
For Job-M, jobs from more users are included (13); about 25\% of jobs stem from the same user, here, hex\_lev, hex\_native, and ks is including more users (29, 33, and 37, respectively) than the other three algorithms.
For Job-L, the two hex algorithms include with (12 and 13) a bit more diverse user community than the bin algorithms (9) but hex\_phases covers 35 users.
For Job-L, the two hex algorithms include with (12 and 13) a bit more diverse user community than the bin algorithms (9) but hex\_phases cover 35 users.
We didn't include the group analysis in the figure as user count and group id is proportional, at most the number of users is 2x the number of groups.
Thus, a user is likely from the same group and the number of groups is similar to the number of unique users.
\paragraph{Node distribution.}
All algorithms reduce over the node dimensions, therefore, we naturally expect a big inclusion across node range -- as long as the average I/O behavior of the jobs are similar.
All algorithms reduce over the node dimensions, therefore, we naturally expect a big inclusion across node range -- as long as the average I/O behavior of the jobs is similar.
\Cref{fig:nodes-job} shows a boxplot for the node counts in the Top\,100 -- the red line marks the reference job.
For Job-M and Job-L, we can observe that indeed the range of similar nodes is between 1 and 128.
For Job-S, all 100 top-ranked jobs use one node.
@ -518,11 +521,11 @@ For Job-L, the job itself isn't included in the chosen Top\,100 (see \Cref{fig:h
\end{figure}
\subsubsection{Algorithmic differences}
To verify that the different algorithms behave differently, the intersection for the Top\,100 is computed for all combination of algorithms and visualized in \Cref{fig:heatmap-job}.
To verify that the different algorithms behave differently, the intersection for the Top\,100 is computed for all combinations of algorithms and visualized in \Cref{fig:heatmap-job}.
Bin\_all and bin\_aggzeros overlap with at least 99 ranks for all three jobs.
While there is some reordering, both algorithms lead to a comparable set.
All algorithms have significant overlap for Job-S.
For Job\-M, however, they lead to a different ranking and Top\,100, particularly ks determines a different set.
All algorithms have a significant overlap for Job-S.
For Job\-M, however, they lead to a different ranking, and Top\,100, particularly ks determines a different set.
Generally, hex\_lev and Hex\_native are generating more similar results than other algorithms.
From this analysis, we conclude that one representative from binary quantization is sufficient as it generates very similar results while the other algorithms identify mostly disjoint behavioral aspects and, therefore, should be analyzed individually.
@ -552,7 +555,7 @@ From this analysis, we conclude that one representative from binary quantization
%%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%% %%%%%%%%%%%
\section{Assessing Timelines for Similar Jobs}
\label{sec:timelines}
To verify the suitability of the similarity metrics, for each algorithm, we investigated the timelines of all Top\,100 jobs.
We subjectively found that the approach works very well and identifies suitable similar jobs.
To demonstrate this, we include a selection of job timelines -- typically Rank\,2, Rank\,15, and Rank\,100, and selected interesting job profiles.
@ -566,7 +569,7 @@ The job name of Job-S suggests that is applied to the control variable.
In the metadata, we found 22,580 jobs with “cmor” in the name of which 367 jobs mention “control”.
The bin and ks algorithms identify one job which name doesn't include “cmor”,
All other algorithm identify only “cmor” jobs and 26-38 of these jobs are applied to “control” (see \Cref{tbl:control-jobs}) -- only the ks algorithm doesn't identify any job with control.
All other algorithms identify only “cmor” jobs and 26-38 of these jobs are applied to “control” (see \Cref{tbl:control-jobs}) -- only the ks algorithm doesn't identify any job with control.
A selection of job timelines is given in \Cref{fig:job-S-hex-lev}; all of these jobs are jobs on control variables.
The single non-cmor job and a high-ranked non-control cmor job is shown in \Cref{fig:job-S-bin-agg}.
While we cannot visually see much differences between these two jobs compared to the cmor job processing the control variables, the algorithms indicate that jobs processing the control variables must be more similar as they appear much more frequently in the Top\,100 jobs than in all jobs labeled with “cmor”.
@ -678,14 +681,14 @@ For Job-S, we found that all algorithms work well and, therefore, omit further t
\subsection{Job-M}
Inspecting the Top\,100 for this reference jobs is highlighting the differences between the algorithms.
Inspecting the Top\,100 for this reference job is highlighting the differences between the algorithms.
All algorithms identify a diverse range of job names for this reference job in the Top\,100.
Firstly, the name of the reference job appears 30 times in the whole dataset so this kind job type isn't necessarily executed frequently and, therefore, our Top\,100 is expected to contain other names.
Firstly, the name of the reference job appears 30 times in the whole dataset so this job type isn't necessarily executed frequently and, therefore, our Top\,100 is expected to contain other names.
Some applications are more prominent in these sets, e.g., for bin\_aggzero, 32\,jobs contain WRF (a model) in the name.
The number of unique names is 19, 38, 49 to 51 for bin\_aggzero, hex\_phases, hex\_native and hex\_lev, respectively.
The jobs that are similar according to the bin algorithms differ from our expectation.
The jobs that are similar according to the bin algorithms differ from our expectations.
@ -907,7 +910,7 @@ The ks algorithm finds 71 jobs ending with t127, which is a typical model config
One consideration could be to identify jobs that are found by all algorithms, i.e., jobs that meet a certain (rank) threshold for different algorithms.
That would increase the likelihood that these jobs are very similar and what the user is looking for.
The ks algorithm finds jobs with similar histograms which is not necessarily what we are looking for.
The ks algorithm finds jobs with similar histograms which are not necessarily what we are looking for.
%\printbibliography
\end{document}